1 <!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.01//EN"
2 "http://www.w3.org/TR/html4/strict.dtd">
5 <title>Writing an LLVM Pass</title>
6 <link rel="stylesheet" href="llvm.css" type="text/css">
10 <div class="doc_title">
15 <li><a href="#introduction">Introduction - What is a pass?</a></li>
16 <li><a href="#quickstart">Quick Start - Writing hello world</a>
18 <li><a href="#makefile">Setting up the build environment</a></li>
19 <li><a href="#basiccode">Basic code required</a></li>
20 <li><a href="#running">Running a pass with <tt>opt</tt>
21 or <tt>analyze</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="#Pass">The <tt>Pass</tt> class</a>
28 <li><a href="#run">The <tt>run</tt> method</a></li>
30 <li><a href="#CallGraphSCCPass">The <tt>CallGraphSCCPass</tt> class</a>
32 <li><a href="#doInitialization_scc">The <tt>doInitialization(Module
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(Module
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="#BasicBlockPass">The <tt>BasicBlockPass</tt> class</a>
48 <li><a href="#doInitialization_fn">The <tt>doInitialization(Function
49 &)</tt> method</a></li>
50 <li><a href="#runOnBasicBlock">The <tt>runOnBasicBlock</tt>
52 <li><a href="#doFinalization_fn">The <tt>doFinalization(Function
53 &)</tt> method</a></li>
55 <li><a href="#MachineFunctionPass">The <tt>MachineFunctionPass</tt>
58 <li><a href="#runOnMachineFunction">The
59 <tt>runOnMachineFunction(MachineFunction &)</tt> method</a></li>
62 <li><a href="#registration">Pass Registration</a>
64 <li><a href="#print">The <tt>print</tt> method</a></li>
66 <li><a href="#interaction">Specifying interactions between passes</a>
68 <li><a href="#getAnalysisUsage">The <tt>getAnalysisUsage</tt>
70 <li><a href="#AU::addRequired">The <tt>AnalysisUsage::addRequired<></tt> and <tt>AnalysisUsage::addRequiredTransitive<></tt> methods</a></li>
71 <li><a href="#AU::addPreserved">The <tt>AnalysisUsage::addPreserved<></tt> method</a></li>
72 <li><a href="#AU::examples">Example implementations of <tt>getAnalysisUsage</tt></a></li>
73 <li><a href="#getAnalysis">The <tt>getAnalysis<></tt> and <tt>getAnalysisToUpdate<></tt> methods</a></li>
75 <li><a href="#analysisgroup">Implementing Analysis Groups</a>
77 <li><a href="#agconcepts">Analysis Group Concepts</a></li>
78 <li><a href="#registerag">Using <tt>RegisterAnalysisGroup</tt></a></li>
80 <li><a href="#passmanager">What PassManager does</a>
82 <li><a href="#releaseMemory">The <tt>releaseMemory</tt> method</a></li>
84 <li><a href="#debughints">Using GDB with dynamically loaded passes</a>
86 <li><a href="#breakpoint">Setting a breakpoint in your pass</a></li>
87 <li><a href="#debugmisc">Miscellaneous Problems</a></li>
89 <li><a href="#future">Future extensions planned</a>
91 <li><a href="#SMP">Multithreaded LLVM</a></li>
92 <li><a href="#ModuleSource">A new <tt>ModuleSource</tt> interface</a></li>
93 <li><a href="#PassFunctionPass"><tt>Pass</tt>es requiring
94 <tt>FunctionPass</tt>es</a></li>
98 <div class="doc_author">
99 <p>Written by <a href="mailto:sabre@nondot.org">Chris Lattner</a></p>
102 <!-- *********************************************************************** -->
103 <div class="doc_section">
104 <a name="introduction">Introduction - What is a pass?</a>
106 <!-- *********************************************************************** -->
108 <div class="doc_text">
110 <p>The LLVM Pass Framework is an important part of the LLVM system, because LLVM
111 passes are where most of the interesting parts of the compiler exist. Passes
112 perform the transformations and optimizations that make up the compiler, they
113 build the analysis results that are used by these transformations, and they are,
114 above all, a structuring technique for compiler code.</p>
116 <p>All LLVM passes are subclasses of the <tt><a
117 href="http://llvm.cs.uiuc.edu/doxygen/classllvm_1_1Pass.html">Pass</a></tt>
118 class, which implement functionality by overriding virtual methods inherited
119 from <tt>Pass</tt>. Depending on how your pass works, you may be able to
120 inherit from the <tt><a href="#CallGraphSCCPass">CallGraphSCCPass</a></tt>,
121 <tt><a href="#FunctionPass">FunctionPass</a></tt>, or <tt><a
122 href="#BasicBlockPass">BasicBlockPass</a></tt> classes, which gives the system
123 more information about what your pass does, and how it can be combined with
124 other passes. One of the main features of the LLVM Pass Framework is that it
125 schedules passes to run in an efficient way based on the constraints that your
126 pass meets (which are indicated by which class they derive from).</p>
128 <p>We start by showing you how to construct a pass, everything from setting up
129 the code, to compiling, loading, and executing it. After the basics are down,
130 more advanced features are discussed.</p>
134 <!-- *********************************************************************** -->
135 <div class="doc_section">
136 <a name="quickstart">Quick Start - Writing hello world</a>
138 <!-- *********************************************************************** -->
140 <div class="doc_text">
142 <p>Here we describe how to write the "hello world" of passes. The "Hello" pass
143 is designed to simply print out the name of non-external functions that exist in
144 the program being compiled. It does not modify the program at all, it just
145 inspects it. The source code and files for this pass are available in the LLVM
146 source tree in the <tt>lib/Transforms/Hello</tt> directory.</p>
150 <!-- ======================================================================= -->
151 <div class="doc_subsection">
152 <a name="makefile">Setting up the build environment</a>
155 <div class="doc_text">
157 <p>First thing you need to do is create a new directory somewhere in the LLVM
158 source base. For this example, we'll assume that you made
159 "<tt>lib/Transforms/Hello</tt>". The first thing you must do is set up a build
160 script (Makefile) that will compile the source code for the new pass. To do
161 this, copy this into "<tt>Makefile</tt>":</p>
166 # Makefile for hello pass
168 # Path to top level of LLVM heirarchy
171 # Name of the library to build
174 # Build a dynamically loadable shared object
177 # Include the makefile implementation stuff
178 include $(LEVEL)/Makefile.common
181 <p>This makefile specifies that all of the <tt>.cpp</tt> files in the current
182 directory are to be compiled and linked together into a
183 <tt>lib/Debug/libhello.so</tt> shared object that can be dynamically loaded by
184 the <tt>opt</tt> or <tt>analyze</tt> tools. If your operating system uses a
185 suffix other than .so (such as windows or Mac OS/X), the appropriate extension
188 <p>Now that we have the build scripts set up, we just need to write the code for
193 <!-- ======================================================================= -->
194 <div class="doc_subsection">
195 <a name="basiccode">Basic code required</a>
198 <div class="doc_text">
200 <p>Now that we have a way to compile our new pass, we just have to write it.
204 <b>#include</b> "<a href="http://llvm.cs.uiuc.edu/doxygen/Pass_8h-source.html">llvm/Pass.h</a>"
205 <b>#include</b> "<a href="http://llvm.cs.uiuc.edu/doxygen/Function_8h-source.html">llvm/Function.h</a>"
208 <p>Which are needed because we are writing a <tt><a
209 href="http://llvm.cs.uiuc.edu/doxygen/classllvm_1_1Pass.html">Pass</a></tt>, and
210 we are operating on <tt><a
211 href="http://llvm.cs.uiuc.edu/doxygen/classllvm_1_1Function.html">Function</a></tt>'s.</p>
215 <b>using namespace llvm;</b>
217 <p>... which is required because the functions from the include files
218 live in the llvm namespace.
227 <p>... which starts out an anonymous namespace. Anonymous namespaces are to C++
228 what the "<tt>static</tt>" keyword is to C (at global scope). It makes the
229 things declared inside of the anonymous namespace only visible to the current
230 file. If you're not familiar with them, consult a decent C++ book for more
233 <p>Next, we declare our pass itself:</p>
236 <b>struct</b> Hello : <b>public</b> <a href="#FunctionPass">FunctionPass</a> {
239 <p>This declares a "<tt>Hello</tt>" class that is a subclass of <tt><a
240 href="http://llvm.cs.uiuc.edu/doxygen/structllvm_1_1FunctionPass.html">FunctionPass</a></tt>.
241 The different builtin pass subclasses are described in detail <a
242 href="#passtype">later</a>, but for now, know that <a
243 href="#FunctionPass"><tt>FunctionPass</tt></a>'s operate a function at a
247 <b>virtual bool</b> <a href="#runOnFunction">runOnFunction</a>(Function &F) {
248 std::cerr << "<i>Hello: </i>" << F.getName() << "\n";
251 }; <i>// end of struct Hello</i>
254 <p>We declare a "<a href="#runOnFunction"><tt>runOnFunction</tt></a>" method,
255 which overloads an abstract virtual method inherited from <a
256 href="#FunctionPass"><tt>FunctionPass</tt></a>. This is where we are supposed
257 to do our thing, so we just print out our message with the name of each
261 RegisterOpt<Hello> X("<i>hello</i>", "<i>Hello World Pass</i>");
262 } <i>// end of anonymous namespace</i>
265 <p>Lastly, we register our class <tt>Hello</tt>, giving it a command line
266 argument "<tt>hello</tt>", and a name "<tt>Hello World Pass</tt>". There are
267 several different ways of <a href="#registration">registering your pass</a>,
268 depending on what it is to be used for. For "optimizations" we use the
269 <tt>RegisterOpt</tt> template.</p>
271 <p>As a whole, the <tt>.cpp</tt> file looks like:</p>
274 <b>#include</b> "<a href="http://llvm.cs.uiuc.edu/doxygen/Pass_8h-source.html">llvm/Pass.h</a>"
275 <b>#include</b> "<a href="http://llvm.cs.uiuc.edu/doxygen/Function_8h-source.html">llvm/Function.h</a>"
277 <b>using namespace llvm;</b>
280 <b>struct Hello</b> : <b>public</b> <a href="#FunctionPass">FunctionPass</a> {
281 <b>virtual bool</b> <a href="#runOnFunction">runOnFunction</a>(Function &F) {
282 std::cerr << "<i>Hello: </i>" << F.getName() << "\n";
287 RegisterOpt<Hello> X("<i>hello</i>", "<i>Hello World Pass</i>");
291 <p>Now that it's all together, compile the file with a simple "<tt>gmake</tt>"
292 command in the local directory and you should get a new
293 "<tt>lib/Debug/libhello.so</tt> file. Note that everything in this file is
294 contained in an anonymous namespace: this reflects the fact that passes are self
295 contained units that do not need external interfaces (although they can have
296 them) to be useful.</p>
300 <!-- ======================================================================= -->
301 <div class="doc_subsection">
302 <a name="running">Running a pass with <tt>opt</tt> or <tt>analyze</tt></a>
305 <div class="doc_text">
307 <p>Now that you have a brand new shiny shared object file, we can use the
308 <tt>opt</tt> command to run an LLVM program through your pass. Because you
309 registered your pass with the <tt>RegisterOpt</tt> template, you will be able to
310 use the <tt>opt</tt> tool to access it, once loaded.</p>
312 <p>To test it, follow the example at the end of the <a
313 href="GettingStarted.html">Getting Started Guide</a> to compile "Hello World" to
314 LLVM. We can now run the bytecode file (<tt>hello.bc</tt>) for the program
315 through our transformation like this (or course, any bytecode file will
319 $ opt -load ../../../lib/Debug/libhello.so -hello < hello.bc > /dev/null
325 <p>The '<tt>-load</tt>' option specifies that '<tt>opt</tt>' should load your
326 pass as a shared object, which makes '<tt>-hello</tt>' a valid command line
327 argument (which is one reason you need to <a href="#registration">register your
328 pass</a>). Because the hello pass does not modify the program in any
329 interesting way, we just throw away the result of <tt>opt</tt> (sending it to
330 <tt>/dev/null</tt>).</p>
332 <p>To see what happened to the other string you registered, try running
333 <tt>opt</tt> with the <tt>--help</tt> option:</p>
336 $ opt -load ../../../lib/Debug/libhello.so --help
337 OVERVIEW: llvm .bc -> .bc modular optimizer
339 USAGE: opt [options] <input bytecode>
342 Optimizations available:
344 -funcresolve - Resolve Functions
345 -gcse - Global Common Subexpression Elimination
346 -globaldce - Dead Global Elimination
347 <b>-hello - Hello World Pass</b>
348 -indvars - Canonicalize Induction Variables
349 -inline - Function Integration/Inlining
350 -instcombine - Combine redundant instructions
354 <p>The pass name get added as the information string for your pass, giving some
355 documentation to users of <tt>opt</tt>. Now that you have a working pass, you
356 would go ahead and make it do the cool transformations you want. Once you get
357 it all working and tested, it may become useful to find out how fast your pass
358 is. The <a href="#passManager"><tt>PassManager</tt></a> provides a nice command
359 line option (<tt>--time-passes</tt>) that allows you to get information about
360 the execution time of your pass along with the other passes you queue up. For
364 $ opt -load ../../../lib/Debug/libhello.so -hello -time-passes < hello.bc > /dev/null
368 ===============================================================================
369 ... Pass execution timing report ...
370 ===============================================================================
371 Total Execution Time: 0.02 seconds (0.0479059 wall clock)
373 ---User Time--- --System Time-- --User+System-- ---Wall Time--- --- Pass Name ---
374 0.0100 (100.0%) 0.0000 ( 0.0%) 0.0100 ( 50.0%) 0.0402 ( 84.0%) Bytecode Writer
375 0.0000 ( 0.0%) 0.0100 (100.0%) 0.0100 ( 50.0%) 0.0031 ( 6.4%) Dominator Set Construction
376 0.0000 ( 0.0%) 0.0000 ( 0.0%) 0.0000 ( 0.0%) 0.0013 ( 2.7%) Module Verifier
377 <b> 0.0000 ( 0.0%) 0.0000 ( 0.0%) 0.0000 ( 0.0%) 0.0033 ( 6.9%) Hello World Pass</b>
378 0.0100 (100.0%) 0.0100 (100.0%) 0.0200 (100.0%) 0.0479 (100.0%) TOTAL
381 <p>As you can see, our implementation above is pretty fast :). The additional
382 passes listed are automatically inserted by the '<tt>opt</tt>' tool to verify
383 that the LLVM emitted by your pass is still valid and well formed LLVM, which
384 hasn't been broken somehow.</p>
386 <p>Now that you have seen the basics of the mechanics behind passes, we can talk
387 about some more details of how they work and how to use them.</p>
391 <!-- *********************************************************************** -->
392 <div class="doc_section">
393 <a name="passtype">Pass classes and requirements</a>
395 <!-- *********************************************************************** -->
397 <div class="doc_text">
399 <p>One of the first things that you should do when designing a new pass is to
400 decide what class you should subclass for your pass. The <a
401 href="#basiccode">Hello World</a> example uses the <tt><a
402 href="#FunctionPass">FunctionPass</a></tt> class for its implementation, but we
403 did not discuss why or when this should occur. Here we talk about the classes
404 available, from the most general to the most specific.</p>
406 <p>When choosing a superclass for your Pass, you should choose the <b>most
407 specific</b> class possible, while still being able to meet the requirements
408 listed. This gives the LLVM Pass Infrastructure information necessary to
409 optimize how passes are run, so that the resultant compiler isn't unneccesarily
414 <!-- ======================================================================= -->
415 <div class="doc_subsection">
416 <a name="ImmutablePass">The <tt>ImmutablePass</tt> class</a>
419 <div class="doc_text">
421 <p>The most plain and boring type of pass is the "<tt><a
422 href="http://llvm.cs.uiuc.edu/doxygen/structllvm_1_1ImmutablePass.html">ImmutablePass</a></tt>"
423 class. This pass type is used for passes that do not have to be run, do not
424 change state, and never need to be updated. This is not a normal type of
425 transformation or analysis, but can provide information about the current
426 compiler configuration.</p>
428 <p>Although this pass class is very infrequently used, it is important for
429 providing information about the current target machine being compiled for, and
430 other static information that can affect the various transformations.</p>
432 <p><tt>ImmutablePass</tt>es never invalidate other transformations, are never
433 invalidated, and are never "run".</p>
437 <!-- ======================================================================= -->
438 <div class="doc_subsection">
439 <a name="Pass">The <tt>Pass</tt> class</a>
442 <div class="doc_text">
445 href="http://llvm.cs.uiuc.edu/doxygen/classllvm_1_1Pass.html">Pass</a></tt>"
446 class is the most general of all superclasses that you can use. Deriving from
447 <tt>Pass</tt> indicates that your pass uses the entire program as a unit,
448 refering to function bodies in no predictable order, or adding and removing
449 functions. Because nothing is known about the behavior of direct <tt>Pass</tt>
450 subclasses, no optimization can be done for their execution.</p>
452 <p>To write a correct <tt>Pass</tt> subclass, derive from <tt>Pass</tt> and
453 overload the <tt>run</tt> method with the following signature:</p>
457 <!-- _______________________________________________________________________ -->
458 <div class="doc_subsubsection">
459 <a name="run">The <tt>run</tt> method</a>
462 <div class="doc_text">
465 <b>virtual bool</b> run(Module &M) = 0;
468 <p>The <tt>run</tt> method performs the interesting work of the pass, and should
469 return true if the module was modified by the transformation, false
474 <!-- ======================================================================= -->
475 <div class="doc_subsection">
476 <a name="CallGraphSCCPass">The <tt>CallGraphSCCPass</tt> class</a>
479 <div class="doc_text">
482 href="http://llvm.cs.uiuc.edu/doxygen/structllvm_1_1CallGraphSCCPass.html">CallGraphSCCPass</a></tt>"
483 is used by passes that need to traverse the program bottom-up on the call graph
484 (callees before callers). Deriving from CallGraphSCCPass provides some
485 mechanics for building and traversing the CallGraph, but also allows the system
486 to optimize execution of CallGraphSCCPass's. If your pass meets the
487 requirements outlined below, and doesn't meet the requirements of a <tt><a
488 href="#FunctionPass">FunctionPass</a></tt> or <tt><a
489 href="#BasicBlockPass">BasicBlockPass</a></tt>, you should derive from
490 <tt>CallGraphSCCPass</tt>.</p>
492 <p><b>TODO</b>: explain briefly what SCC, Tarjan's algo, and B-U mean.</p>
494 <p>To be explicit, <tt>CallGraphSCCPass</tt> subclasses are:</p>
498 <li>... <em>not allowed</em> to modify any <tt>Function</tt>s that are not in
499 the current SCC.</li>
501 <li>... <em>allowed</em> to inspect any Function's other than those in the
502 current SCC and the direct callees of the SCC.</li>
504 <li>... <em>required</em> to preserve the current CallGraph object, updating it
505 to reflect any changes made to the program.</li>
507 <li>... <em>not allowed</em> to add or remove SCC's from the current Module,
508 though they may change the contents of an SCC.</li>
510 <li>... <em>allowed</em> to add or remove global variables from the current
513 <li>... <em>allowed</em> to maintain state across invocations of
514 <a href="#runOnSCC"><tt>runOnSCC</tt></a> (including global data).</li>
517 <p>Implementing a <tt>CallGraphSCCPass</tt> is slightly tricky in some cases
518 because it has to handle SCCs with more than one node in it. All of the virtual
519 methods described below should return true if they modified the program, or
520 false if they didn't.</p>
524 <!-- _______________________________________________________________________ -->
525 <div class="doc_subsubsection">
526 <a name="doInitialization_scc">The <tt>doInitialization(Module &)</tt>
530 <div class="doc_text">
533 <b>virtual bool</b> doInitialization(Module &M);
536 <p>The <tt>doIninitialize</tt> method is allowed to do most of the things that
537 <tt>CallGraphSCCPass</tt>'s are not allowed to do. They can add and remove
538 functions, get pointers to functions, etc. The <tt>doInitialization</tt> method
539 is designed to do simple initialization type of stuff that does not depend on
540 the SCCs being processed. The <tt>doInitialization</tt> method call is not
541 scheduled to overlap with any other pass executions (thus it should be very
546 <!-- _______________________________________________________________________ -->
547 <div class="doc_subsubsection">
548 <a name="runOnSCC">The <tt>runOnSCC</tt> method</a>
551 <div class="doc_text">
554 <b>virtual bool</b> runOnSCC(const std::vector<CallGraphNode *> &SCCM) = 0;
557 <p>The <tt>runOnSCC</tt> method performs the interesting work of the pass, and
558 should return true if the module was modified by the transformation, false
563 <!-- _______________________________________________________________________ -->
564 <div class="doc_subsubsection">
565 <a name="doFinalization_scc">The <tt>doFinalization(Module
566 &)</tt> method</a>
569 <div class="doc_text">
572 <b>virtual bool</b> doFinalization(Module &M);
575 <p>The <tt>doFinalization</tt> method is an infrequently used method that is
576 called when the pass framework has finished calling <a
577 href="#runOnFunction"><tt>runOnFunction</tt></a> for every function in the
578 program being compiled.</p>
582 <!-- ======================================================================= -->
583 <div class="doc_subsection">
584 <a name="FunctionPass">The <tt>FunctionPass</tt> class</a>
587 <div class="doc_text">
589 <p>In contrast to direct <tt>Pass</tt> subclasses, direct <tt><a
590 href="http://llvm.cs.uiuc.edu/doxygen/classllvm_1_1Pass.html">FunctionPass</a></tt>
591 subclasses do have a predictable, local behavior that can be expected by the
592 system. All <tt>FunctionPass</tt> execute on each function in the program
593 independent of all of the other functions in the program.
594 <tt>FunctionPass</tt>'s do not require that they are executed in a particular
595 order, and <tt>FunctionPass</tt>'s do not modify external functions.</p>
597 <p>To be explicit, <tt>FunctionPass</tt> subclasses are not allowed to:</p>
600 <li>Modify a Function other than the one currently being processed.</li>
601 <li>Add or remove Function's from the current Module.</li>
602 <li>Add or remove global variables from the current Module.</li>
603 <li>Maintain state across invocations of
604 <a href="#runOnFunction"><tt>runOnFunction</tt></a> (including global data)</li>
607 <p>Implementing a <tt>FunctionPass</tt> is usually straightforward (See the <a
608 href="#basiccode">Hello World</a> pass for example). <tt>FunctionPass</tt>'s
609 may overload three virtual methods to do their work. All of these methods
610 should return true if they modified the program, or false if they didn't.</p>
614 <!-- _______________________________________________________________________ -->
615 <div class="doc_subsubsection">
616 <a name="doInitialization_mod">The <tt>doInitialization(Module &)</tt>
620 <div class="doc_text">
623 <b>virtual bool</b> doInitialization(Module &M);
626 <p>The <tt>doIninitialize</tt> method is allowed to do most of the things that
627 <tt>FunctionPass</tt>'s are not allowed to do. They can add and remove
628 functions, get pointers to functions, etc. The <tt>doInitialization</tt> method
629 is designed to do simple initialization type of stuff that does not depend on
630 the functions being processed. The <tt>doInitialization</tt> method call is not
631 scheduled to overlap with any other pass executions (thus it should be very
634 <p>A good example of how this method should be used is the <a
635 href="http://llvm.cs.uiuc.edu/doxygen/LowerAllocations_8cpp-source.html">LowerAllocations</a>
636 pass. This pass converts <tt>malloc</tt> and <tt>free</tt> instructions into
637 platform dependent <tt>malloc()</tt> and <tt>free()</tt> function calls. It
638 uses the <tt>doInitialization</tt> method to get a reference to the malloc and
639 free functions that it needs, adding prototypes to the module if necessary.</p>
643 <!-- _______________________________________________________________________ -->
644 <div class="doc_subsubsection">
645 <a name="runOnFunction">The <tt>runOnFunction</tt> method</a>
648 <div class="doc_text">
651 <b>virtual bool</b> runOnFunction(Function &F) = 0;
654 <p>The <tt>runOnFunction</tt> method must be implemented by your subclass to do
655 the transformation or analysis work of your pass. As usual, a true value should
656 be returned if the function is modified.</p>
660 <!-- _______________________________________________________________________ -->
661 <div class="doc_subsubsection">
662 <a name="doFinalization_mod">The <tt>doFinalization(Module
663 &)</tt> method</a>
666 <div class="doc_text">
669 <b>virtual bool</b> doFinalization(Module &M);
672 <p>The <tt>doFinalization</tt> method is an infrequently used method that is
673 called when the pass framework has finished calling <a
674 href="#runOnFunction"><tt>runOnFunction</tt></a> for every function in the
675 program being compiled.</p>
679 <!-- ======================================================================= -->
680 <div class="doc_subsection">
681 <a name="BasicBlockPass">The <tt>BasicBlockPass</tt> class</a>
684 <div class="doc_text">
686 <p><tt>BasicBlockPass</tt>'s are just like <a
687 href="#FunctionPass"><tt>FunctionPass</tt></a>'s, except that they must limit
688 their scope of inspection and modification to a single basic block at a time.
689 As such, they are <b>not</b> allowed to do any of the following:</p>
692 <li>Modify or inspect any basic blocks outside of the current one</li>
693 <li>Maintain state across invocations of
694 <a href="#runOnBasicBlock"><tt>runOnBasicBlock</tt></a></li>
695 <li>Modify the constrol flow graph (by altering terminator instructions)</li>
696 <li>Any of the things verboten for
697 <a href="#FunctionPass"><tt>FunctionPass</tt></a>es.</li>
700 <p><tt>BasicBlockPass</tt>es are useful for traditional local and "peephole"
701 optimizations. They may override the same <a
702 href="#doInitialization_mod"><tt>doInitialization(Module &)</tt></a> and <a
703 href="#doFinalization_mod"><tt>doFinalization(Module &)</tt></a> methods that <a
704 href="#FunctionPass"><tt>FunctionPass</tt></a>'s have, but also have the following virtual methods that may also be implemented:</p>
708 <!-- _______________________________________________________________________ -->
709 <div class="doc_subsubsection">
710 <a name="doInitialization_fn">The <tt>doInitialization(Function
711 &)</tt> method</a>
714 <div class="doc_text">
717 <b>virtual bool</b> doInitialization(Function &F);
720 <p>The <tt>doIninitialize</tt> method is allowed to do most of the things that
721 <tt>BasicBlockPass</tt>'s are not allowed to do, but that
722 <tt>FunctionPass</tt>'s can. The <tt>doInitialization</tt> method is designed
723 to do simple initialization type of stuff that does not depend on the
724 BasicBlocks being processed. The <tt>doInitialization</tt> method call is not
725 scheduled to overlap with any other pass executions (thus it should be very
730 <!-- _______________________________________________________________________ -->
731 <div class="doc_subsubsection">
732 <a name="runOnBasicBlock">The <tt>runOnBasicBlock</tt> method</a>
735 <div class="doc_text">
738 <b>virtual bool</b> runOnBasicBlock(BasicBlock &BB) = 0;
741 <p>Override this function to do the work of the <tt>BasicBlockPass</tt>. This
742 function is not allowed to inspect or modify basic blocks other than the
743 parameter, and are not allowed to modify the CFG. A true value must be returned
744 if the basic block is modified.</p>
748 <!-- _______________________________________________________________________ -->
749 <div class="doc_subsubsection">
750 <a name="doFinalization_fn">The <tt>doFinalization(Function &)</tt>
754 <div class="doc_text">
757 <b>virtual bool</b> doFinalization(Function &F);
760 <p>The <tt>doFinalization</tt> method is an infrequently used method that is
761 called when the pass framework has finished calling <a
762 href="#runOnBasicBlock"><tt>runOnBasicBlock</tt></a> for every BasicBlock in the
763 program being compiled. This can be used to perform per-function
768 <!-- ======================================================================= -->
769 <div class="doc_subsection">
770 <a name="MachineFunctionPass">The <tt>MachineFunctionPass</tt> class</a>
773 <div class="doc_text">
775 <p>A <tt>MachineFunctionPass</tt> is a part of the LLVM code generator that
776 executes on the machine-dependent representation of each LLVM function in the
777 program. A <tt>MachineFunctionPass</tt> is also a <tt>FunctionPass</tt>, so all
778 the restrictions that apply to a <tt>FunctionPass</tt> also apply to it.
779 <tt>MachineFunctionPass</tt>es also have additional restrictions. In particular,
780 <tt>MachineFunctionPass</tt>es are not allowed to do any of the following:</p>
783 <li>Modify any LLVM Instructions, BasicBlocks or Functions.</li>
784 <li>Modify a MachineFunction other than the one currently being processed.</li>
785 <li>Add or remove MachineFunctions from the current Module.</li>
786 <li>Add or remove global variables from the current Module.</li>
787 <li>Maintain state across invocations of <a
788 href="#runOnMachineFunction"><tt>runOnMachineFunction</tt></a> (including global
794 <!-- _______________________________________________________________________ -->
795 <div class="doc_subsubsection">
796 <a name="runOnMachineFunction">The <tt>runOnMachineFunction(MachineFunction
797 &MF)</tt> method</a>
800 <div class="doc_text">
803 <b>virtual bool</b> runOnMachineFunction(MachineFunction &MF) = 0;
806 <p><tt>runOnMachineFunction</tt> can be considered the main entry point of a
807 <tt>MachineFunctionPass</tt>; that is, you should override this method to do the
808 work of your <tt>MachineFunctionPass</tt>.</p>
810 <p>The <tt>runOnMachineFunction</tt> method is called on every
811 <tt>MachineFunction</tt> in a <tt>Module</tt>, so that the
812 <tt>MachineFunctionPass</tt> may perform optimizations on the machine-dependent
813 representation of the function. If you want to get at the LLVM <tt>Function</tt>
814 for the <tt>MachineFunction</tt> you're working on, use
815 <tt>MachineFunction</tt>'s <tt>getFunction()</tt> accessor method -- but
816 remember, you may not modify the LLVM <tt>Function</tt> or its contents from a
817 <tt>MachineFunctionPass</tt>.</p>
821 <!-- *********************************************************************** -->
822 <div class="doc_section">
823 <a name="registration">Pass registration</a>
825 <!-- *********************************************************************** -->
827 <div class="doc_text">
829 <p>In the <a href="#basiccode">Hello World</a> example pass we illustrated how
830 pass registration works, and discussed some of the reasons that it is used and
831 what it does. Here we discuss how and why passes are registered.</p>
833 <p>Passes can be registered in several different ways. Depending on the general
834 classification of the pass, you should use one of the following templates to
835 register the pass:</p>
838 <li><b><tt>RegisterOpt</tt></b> - This template should be used when you are
839 registering a pass that logically should be available for use in the
840 '<tt>opt</tt>' utility.</li>
842 <li><b><tt>RegisterAnalysis</tt></b> - This template should be used when you are
843 registering a pass that logically should be available for use in the
844 '<tt>analyze</tt>' utility.</li>
846 <li><b><tt>RegisterPass</tt></b> - This is the generic form of the
847 <tt>Register*</tt> templates that should be used if you want your pass listed by
848 multiple or no utilities. This template takes an extra third argument that
849 specifies which tools it should be listed in. See the <a
850 href="http://llvm.cs.uiuc.edu/doxygen/PassSupport_8h-source.html">PassSupport.h</a>
851 file for more information.</li>
855 <p>Regardless of how you register your pass, you must specify at least two
856 parameters. The first parameter is the name of the pass that is to be used on
857 the command line to specify that the pass should be added to a program (for
858 example <tt>opt</tt> or <tt>analyze</tt>). The second argument is the name of
859 the pass, which is to be used for the <tt>--help</tt> output of programs, as
860 well as for debug output generated by the <tt>--debug-pass</tt> option.</p>
862 <p>If a pass is registered to be used by the <tt>analyze</tt> utility, you
863 should implement the virtual <tt>print</tt> method:</p>
867 <!-- _______________________________________________________________________ -->
868 <div class="doc_subsubsection">
869 <a name="print">The <tt>print</tt> method</a>
872 <div class="doc_text">
875 <b>virtual void</b> print(std::ostream &O, <b>const</b> Module *M) <b>const</b>;
878 <p>The <tt>print</tt> method must be implemented by "analyses" in order to print
879 a human readable version of the analysis results. This is useful for debugging
880 an analysis itself, as well as for other people to figure out how an analysis
881 works. The <tt>analyze</tt> tool uses this method to generate its output.</p>
883 <p>The <tt>ostream</tt> parameter specifies the stream to write the results on,
884 and the <tt>Module</tt> parameter gives a pointer to the top level module of the
885 program that has been analyzed. Note however that this pointer may be null in
886 certain circumstances (such as calling the <tt>Pass::dump()</tt> from a
887 debugger), so it should only be used to enhance debug output, it should not be
892 <!-- *********************************************************************** -->
893 <div class="doc_section">
894 <a name="interaction">Specifying interactions between passes</a>
896 <!-- *********************************************************************** -->
898 <div class="doc_text">
900 <p>One of the main responsibilities of the <tt>PassManager</tt> is the make sure
901 that passes interact with each other correctly. Because <tt>PassManager</tt>
902 tries to <a href="#passmanager">optimize the execution of passes</a> it must
903 know how the passes interact with each other and what dependencies exist between
904 the various passes. To track this, each pass can declare the set of passes that
905 are required to be executed before the current pass, and the passes which are
906 invalidated by the current pass.</p>
908 <p>Typically this functionality is used to require that analysis results are
909 computed before your pass is run. Running arbitrary transformation passes can
910 invalidate the computed analysis results, which is what the invalidation set
911 specifies. If a pass does not implement the <tt><a
912 href="#getAnalysisUsage">getAnalysisUsage</a></tt> method, it defaults to not
913 having any prerequisite passes, and invalidating <b>all</b> other passes.</p>
917 <!-- _______________________________________________________________________ -->
918 <div class="doc_subsubsection">
919 <a name="getAnalysisUsage">The <tt>getAnalysisUsage</tt> method</a>
922 <div class="doc_text">
925 <b>virtual void</b> getAnalysisUsage(AnalysisUsage &Info) <b>const</b>;
928 <p>By implementing the <tt>getAnalysisUsage</tt> method, the required and
929 invalidated sets may be specified for your transformation. The implementation
930 should fill in the <tt><a
931 href="http://llvm.cs.uiuc.edu/doxygen/classllvm_1_1AnalysisUsage.html">AnalysisUsage</a></tt>
932 object with information about which passes are required and not invalidated. To
933 do this, a pass may call any of the following methods on the AnalysisUsage
937 <!-- _______________________________________________________________________ -->
938 <div class="doc_subsubsection">
939 <a name="AU::addRequired">The <tt>AnalysisUsage::addRequired<></tt> and <tt>AnalysisUsage::addRequiredTransitive<></tt> methods</a>
942 <div class="doc_text">
944 If you pass requires a previous pass to be executed (an analysis for example),
945 it can use one of these methods to arrange for it to be run before your pass.
946 LLVM has many different types of analyses and passes that can be required,
947 spaning the range from <tt>DominatorSet</tt> to <tt>BreakCriticalEdges</tt>.
948 requiring <tt>BreakCriticalEdges</tt>, for example, guarantees that there will
949 be no critical edges in the CFG when your pass has been run.
953 Some analyses chain to other analyses to do their job. For example, an <a
954 href="AliasAnalysis.html">AliasAnalysis</a> implementation is required to <a
955 href="AliasAnalysis.html#chaining">chain</a> to other alias analysis passes. In
956 cases where analyses chain, the <tt>addRequiredTransitive</tt> method should be
957 used instead of the <tt>addRequired</tt> method. This informs the PassManager
958 that the transitively required pass should be alive as long as the requiring
963 <!-- _______________________________________________________________________ -->
964 <div class="doc_subsubsection">
965 <a name="AU::addPreserved">The <tt>AnalysisUsage::addPreserved<></tt> method</a>
968 <div class="doc_text">
970 One of the jobs of the PassManager is to optimize how and when analyses are run.
971 In particular, it attempts to avoid recomputing data unless it needs to. For
972 this reason, passes are allowed to declare that they preserve (i.e., they don't
973 invalidate) an existing analysis if it's available. For example, a simple
974 constant folding pass would not modify the CFG, so it can't possible effect the
975 results of dominator analysis. By default, all passes are assumed to invalidate
980 The <tt>AnalysisUsage</tt> class provides several methods which are useful in
981 certain circumstances that are related to <tt>addPreserved</tt>. In particular,
982 the <tt>setPreservesAll</tt> method can be called to indicate that the pass does
983 not modify the LLVM program at all (which is true for analyses), and the
984 <tt>setPreservesCFG</tt> method can be used by transformations that change
985 instructions in the program but do not modify the CFG or terminator instructions
986 (note that this property is implicitly set for <a
987 href="#BasicBlockPass">BasicBlockPass</a>'s).
991 <tt>addPreserved</tt> is particularly useful for transformations like
992 <tt>BreakCriticalEdges</tt>. This pass knows how to update a small set of loop
993 and dominator related analyses if they exist, so it can preserve them, despite
994 the fact that it hacks on the CFG.
998 <!-- _______________________________________________________________________ -->
999 <div class="doc_subsubsection">
1000 <a name="AU::examples">Example implementations of <tt>getAnalysisUsage</tt></a>
1003 <div class="doc_text">
1006 <i>// This is an example implementation from an analysis, which does not modify
1007 // the program at all, yet has a prerequisite.</i>
1008 <b>void</b> <a href="http://llvm.cs.uiuc.edu/doxygen/structllvm_1_1PostDominanceFrontier.html">PostDominanceFrontier</a>::getAnalysisUsage(AnalysisUsage &AU) <b>const</b> {
1009 AU.setPreservesAll();
1010 AU.addRequired<<a href="http://llvm.cs.uiuc.edu/doxygen/structllvm_1_1PostDominatorTree.html">PostDominatorTree</a>>();
1017 <i>// This example modifies the program, but does not modify the CFG</i>
1018 <b>void</b> <a href="http://llvm.cs.uiuc.edu/doxygen/structLICM.html">LICM</a>::getAnalysisUsage(AnalysisUsage &AU) <b>const</b> {
1019 AU.setPreservesCFG();
1020 AU.addRequired<<a href="http://llvm.cs.uiuc.edu/doxygen/classllvm_1_1LoopInfo.html">LoopInfo</a>>();
1026 <!-- _______________________________________________________________________ -->
1027 <div class="doc_subsubsection">
1028 <a name="getAnalysis">The <tt>getAnalysis<></tt> and <tt>getAnalysisToUpdate<></tt> methods</a>
1031 <div class="doc_text">
1033 <p>The <tt>Pass::getAnalysis<></tt> method is automatically inherited by
1034 your class, providing you with access to the passes that you declared that you
1035 required with the <a href="#getAnalysisUsage"><tt>getAnalysisUsage</tt></a>
1036 method. It takes a single template argument that specifies which pass class you
1037 want, and returns a reference to that pass. For example:</p>
1040 bool LICM::runOnFunction(Function &F) {
1041 LoopInfo &LI = getAnalysis<LoopInfo>();
1046 <p>This method call returns a reference to the pass desired. You may get a
1047 runtime assertion failure if you attempt to get an analysis that you did not
1048 declare as required in your <a
1049 href="#getAnalysisUsage"><tt>getAnalysisUsage</tt></a> implementation. This
1050 method can be called by your <tt>run*</tt> method implementation, or by any
1051 other local method invoked by your <tt>run*</tt> method.</p>
1054 If your pass is capable of updating analyses if they exist (e.g.,
1055 <tt>BreakCriticalEdges</tt>, as described above), you can use the
1056 <tt>getAnalysisToUpdate</tt> method, which returns a pointer to the analysis if
1057 it is active. For example:</p>
1061 if (DominatorSet *DS = getAnalysisToUpdate<DominatorSet>()) {
1062 <i>// A DominatorSet is active. This code will update it.</i>
1069 <!-- *********************************************************************** -->
1070 <div class="doc_section">
1071 <a name="analysisgroup">Implementing Analysis Groups</a>
1073 <!-- *********************************************************************** -->
1075 <div class="doc_text">
1077 <p>Now that we understand the basics of how passes are defined, how the are
1078 used, and how they are required from other passes, it's time to get a little bit
1079 fancier. All of the pass relationships that we have seen so far are very
1080 simple: one pass depends on one other specific pass to be run before it can run.
1081 For many applications, this is great, for others, more flexibility is
1084 <p>In particular, some analyses are defined such that there is a single simple
1085 interface to the analysis results, but multiple ways of calculating them.
1086 Consider alias analysis for example. The most trivial alias analysis returns
1087 "may alias" for any alias query. The most sophisticated analysis a
1088 flow-sensitive, context-sensitive interprocedural analysis that can take a
1089 significant amount of time to execute (and obviously, there is a lot of room
1090 between these two extremes for other implementations). To cleanly support
1091 situations like this, the LLVM Pass Infrastructure supports the notion of
1092 Analysis Groups.</p>
1096 <!-- _______________________________________________________________________ -->
1097 <div class="doc_subsubsection">
1098 <a name="agconcepts">Analysis Group Concepts</a>
1101 <div class="doc_text">
1103 <p>An Analysis Group is a single simple interface that may be implemented by
1104 multiple different passes. Analysis Groups can be given human readable names
1105 just like passes, but unlike passes, they need not derive from the <tt>Pass</tt>
1106 class. An analysis group may have one or more implementations, one of which is
1107 the "default" implementation.</p>
1109 <p>Analysis groups are used by client passes just like other passes are: the
1110 <tt>AnalysisUsage::addRequired()</tt> and <tt>Pass::getAnalysis()</tt> methods.
1111 In order to resolve this requirement, the <a href="#passmanager">PassManager</a>
1112 scans the available passes to see if any implementations of the analysis group
1113 are available. If none is available, the default implementation is created for
1114 the pass to use. All standard rules for <A href="#interaction">interaction
1115 between passes</a> still apply.</p>
1117 <p>Although <a href="#registration">Pass Registration</a> is optional for normal
1118 passes, all analysis group implementations must be registered, and must use the
1119 <A href="#registerag"><tt>RegisterAnalysisGroup</tt></a> template to join the
1120 implementation pool. Also, a default implementation of the interface
1121 <b>must</b> be registered with <A
1122 href="#registerag"><tt>RegisterAnalysisGroup</tt></a>.</p>
1124 <p>As a concrete example of an Analysis Group in action, consider the <a
1125 href="http://llvm.cs.uiuc.edu/doxygen/classllvm_1_1AliasAnalysis.html">AliasAnalysis</a>
1126 analysis group. The default implementation of the alias analysis interface (the
1128 href="http://llvm.cs.uiuc.edu/doxygen/structBasicAliasAnalysis.html">basicaa</a></tt>
1129 pass) just does a few simple checks that don't require significant analysis to
1130 compute (such as: two different globals can never alias each other, etc).
1131 Passes that use the <tt><a
1132 href="http://llvm.cs.uiuc.edu/doxygen/classllvm_1_1AliasAnalysis.html">AliasAnalysis</a></tt>
1133 interface (for example the <tt><a
1134 href="http://llvm.cs.uiuc.edu/doxygen/structGCSE.html">gcse</a></tt> pass), do
1135 not care which implementation of alias analysis is actually provided, they just
1136 use the designated interface.</p>
1138 <p>From the user's perspective, commands work just like normal. Issuing the
1139 command '<tt>opt -gcse ...</tt>' will cause the <tt>basicaa</tt> class to be
1140 instantiated and added to the pass sequence. Issuing the command '<tt>opt
1141 -somefancyaa -gcse ...</tt>' will cause the <tt>gcse</tt> pass to use the
1142 <tt>somefancyaa</tt> alias analysis (which doesn't actually exist, it's just a
1143 hypothetical example) instead.</p>
1147 <!-- _______________________________________________________________________ -->
1148 <div class="doc_subsubsection">
1149 <a name="registerag">Using <tt>RegisterAnalysisGroup</tt></a>
1152 <div class="doc_text">
1154 <p>The <tt>RegisterAnalysisGroup</tt> template is used to register the analysis
1155 group itself as well as add pass implementations to the analysis group. First,
1156 an analysis should be registered, with a human readable name provided for it.
1157 Unlike registration of passes, there is no command line argument to be specified
1158 for the Analysis Group Interface itself, because it is "abstract":</p>
1161 <b>static</b> RegisterAnalysisGroup<<a href="http://llvm.cs.uiuc.edu/doxygen/classllvm_1_1AliasAnalysis.html">AliasAnalysis</a>> A("<i>Alias Analysis</i>");
1164 <p>Once the analysis is registered, passes can declare that they are valid
1165 implementations of the interface by using the following code:</p>
1169 //<i> Analysis Group implementations <b>must</b> be registered normally...</i>
1170 RegisterOpt<FancyAA>
1171 B("<i>somefancyaa</i>", "<i>A more complex alias analysis implementation</i>");
1173 //<i> Declare that we implement the AliasAnalysis interface</i>
1174 RegisterAnalysisGroup<<a href="http://llvm.cs.uiuc.edu/doxygen/classllvm_1_1AliasAnalysis.html">AliasAnalysis</a>, FancyAA> C;
1178 <p>This just shows a class <tt>FancyAA</tt> that is registered normally, then
1179 uses the <tt>RegisterAnalysisGroup</tt> template to "join" the <tt><a
1180 href="http://llvm.cs.uiuc.edu/doxygen/classllvm_1_1AliasAnalysis.html">AliasAnalysis</a></tt>
1181 analysis group. Every implementation of an analysis group should join using
1182 this template. A single pass may join multiple different analysis groups with
1187 //<i> Analysis Group implementations <b>must</b> be registered normally...</i>
1188 RegisterOpt<<a href="http://llvm.cs.uiuc.edu/doxygen/structBasicAliasAnalysis.html">BasicAliasAnalysis</a>>
1189 D("<i>basicaa</i>", "<i>Basic Alias Analysis (default AA impl)</i>");
1191 //<i> Declare that we implement the AliasAnalysis interface</i>
1192 RegisterAnalysisGroup<<a href="http://llvm.cs.uiuc.edu/doxygen/classllvm_1_1AliasAnalysis.html">AliasAnalysis</a>, <a href="http://llvm.cs.uiuc.edu/doxygen/structBasicAliasAnalysis.html">BasicAliasAnalysis</a>, <b>true</b>> E;
1196 <p>Here we show how the default implementation is specified (using the extra
1197 argument to the <tt>RegisterAnalysisGroup</tt> template). There must be exactly
1198 one default implementation available at all times for an Analysis Group to be
1199 used. Here we declare that the <tt><a
1200 href="http://llvm.cs.uiuc.edu/doxygen/structBasicAliasAnalysis.html">BasicAliasAnalysis</a></tt>
1201 pass is the default implementation for the interface.</p>
1205 <!-- *********************************************************************** -->
1206 <div class="doc_section">
1207 <a name="passmanager">What PassManager does</a>
1209 <!-- *********************************************************************** -->
1211 <div class="doc_text">
1214 href="http://llvm.cs.uiuc.edu/doxygen/PassManager_8h-source.html"><tt>PassManager</tt></a>
1216 href="http://llvm.cs.uiuc.edu/doxygen/classllvm_1_1PassManager.html">class</a>
1217 takes a list of passes, ensures their <a href="#interaction">prerequisites</a>
1218 are set up correctly, and then schedules passes to run efficiently. All of the
1219 LLVM tools that run passes use the <tt>PassManager</tt> for execution of these
1222 <p>The <tt>PassManager</tt> does two main things to try to reduce the execution
1223 time of a series of passes:</p>
1226 <li><b>Share analysis results</b> - The PassManager attempts to avoid
1227 recomputing analysis results as much as possible. This means keeping track of
1228 which analyses are available already, which analyses get invalidated, and which
1229 analyses are needed to be run for a pass. An important part of work is that the
1230 <tt>PassManager</tt> tracks the exact lifetime of all analysis results, allowing
1231 it to <a href="#releaseMemory">free memory</a> allocated to holding analysis
1232 results as soon as they are no longer needed.</li>
1234 <li><b>Pipeline the execution of passes on the program</b> - The
1235 <tt>PassManager</tt> attempts to get better cache and memory usage behavior out
1236 of a series of passes by pipelining the passes together. This means that, given
1237 a series of consequtive <a href="#FunctionPass"><tt>FunctionPass</tt></a>'s, it
1238 will execute all of the <a href="#FunctionPass"><tt>FunctionPass</tt></a>'s on
1239 the first function, then all of the <a
1240 href="#FunctionPass"><tt>FunctionPass</tt></a>es on the second function,
1241 etc... until the entire program has been run through the passes.
1243 <p>This improves the cache behavior of the compiler, because it is only touching
1244 the LLVM program representation for a single function at a time, instead of
1245 traversing the entire program. It reduces the memory consumption of compiler,
1246 because, for example, only one <a
1247 href="http://llvm.cs.uiuc.edu/doxygen/structllvm_1_1DominatorSet.html"><tt>DominatorSet</tt></a>
1248 needs to be calculated at a time. This also makes it possible some <a
1249 href="#SMP">interesting enhancements</a> in the future.</p></li>
1253 <p>The effectiveness of the <tt>PassManager</tt> is influenced directly by how
1254 much information it has about the behaviors of the passes it is scheduling. For
1255 example, the "preserved" set is intentionally conservative in the face of an
1256 unimplemented <a href="#getAnalysisUsage"><tt>getAnalysisUsage</tt></a> method.
1257 Not implementing when it should be implemented will have the effect of not
1258 allowing any analysis results to live across the execution of your pass.</p>
1260 <p>The <tt>PassManager</tt> class exposes a <tt>--debug-pass</tt> command line
1261 options that is useful for debugging pass execution, seeing how things work, and
1262 diagnosing when you should be preserving more analyses than you currently are
1263 (To get information about all of the variants of the <tt>--debug-pass</tt>
1264 option, just type '<tt>opt --help-hidden</tt>').</p>
1266 <p>By using the <tt>--debug-pass=Structure</tt> option, for example, we can see
1267 how our <a href="#basiccode">Hello World</a> pass interacts with other passes.
1268 Lets try it out with the <tt>gcse</tt> and <tt>licm</tt> passes:</p>
1271 $ opt -load ../../../lib/Debug/libhello.so -gcse -licm --debug-pass=Structure < hello.bc > /dev/null
1273 Function Pass Manager
1274 Dominator Set Construction
1275 Immediate Dominators Construction
1276 Global Common Subexpression Elimination
1277 -- Immediate Dominators Construction
1278 -- Global Common Subexpression Elimination
1279 Natural Loop Construction
1280 Loop Invariant Code Motion
1281 -- Natural Loop Construction
1282 -- Loop Invariant Code Motion
1284 -- Dominator Set Construction
1290 <p>This output shows us when passes are constructed and when the analysis
1291 results are known to be dead (prefixed with '<tt>--</tt>'). Here we see that
1292 GCSE uses dominator and immediate dominator information to do its job. The LICM
1293 pass uses natural loop information, which uses dominator sets, but not immediate
1294 dominators. Because immediate dominators are no longer useful after the GCSE
1295 pass, it is immediately destroyed. The dominator sets are then reused to
1296 compute natural loop information, which is then used by the LICM pass.</p>
1298 <p>After the LICM pass, the module verifier runs (which is automatically added
1299 by the '<tt>opt</tt>' tool), which uses the dominator set to check that the
1300 resultant LLVM code is well formed. After it finishes, the dominator set
1301 information is destroyed, after being computed once, and shared by three
1304 <p>Lets see how this changes when we run the <a href="#basiccode">Hello
1305 World</a> pass in between the two passes:</p>
1308 $ opt -load ../../../lib/Debug/libhello.so -gcse -hello -licm --debug-pass=Structure < hello.bc > /dev/null
1310 Function Pass Manager
1311 Dominator Set Construction
1312 Immediate Dominators Construction
1313 Global Common Subexpression Elimination
1314 <b>-- Dominator Set Construction</b>
1315 -- Immediate Dominators Construction
1316 -- Global Common Subexpression Elimination
1317 <b> Hello World Pass
1319 Dominator Set Construction</b>
1320 Natural Loop Construction
1321 Loop Invariant Code Motion
1322 -- Natural Loop Construction
1323 -- Loop Invariant Code Motion
1325 -- Dominator Set Construction
1334 <p>Here we see that the <a href="#basiccode">Hello World</a> pass has killed the
1335 Dominator Set pass, even though it doesn't modify the code at all! To fix this,
1336 we need to add the following <a
1337 href="#getAnalysisUsage"><tt>getAnalysisUsage</tt></a> method to our pass:</p>
1340 <i>// We don't modify the program, so we preserve all analyses</i>
1341 <b>virtual void</b> getAnalysisUsage(AnalysisUsage &AU) <b>const</b> {
1342 AU.setPreservesAll();
1346 <p>Now when we run our pass, we get this output:</p>
1349 $ opt -load ../../../lib/Debug/libhello.so -gcse -hello -licm --debug-pass=Structure < hello.bc > /dev/null
1350 Pass Arguments: -gcse -hello -licm
1352 Function Pass Manager
1353 Dominator Set Construction
1354 Immediate Dominators Construction
1355 Global Common Subexpression Elimination
1356 -- Immediate Dominators Construction
1357 -- Global Common Subexpression Elimination
1360 Natural Loop Construction
1361 Loop Invariant Code Motion
1362 -- Loop Invariant Code Motion
1363 -- Natural Loop Construction
1365 -- Dominator Set Construction
1374 <p>Which shows that we don't accidentally invalidate dominator information
1375 anymore, and therefore do not have to compute it twice.</p>
1379 <!-- _______________________________________________________________________ -->
1380 <div class="doc_subsubsection">
1381 <a name="releaseMemory">The <tt>releaseMemory</tt> method</a>
1384 <div class="doc_text">
1387 <b>virtual void</b> releaseMemory();
1390 <p>The <tt>PassManager</tt> automatically determines when to compute analysis
1391 results, and how long to keep them around for. Because the lifetime of the pass
1392 object itself is effectively the entire duration of the compilation process, we
1393 need some way to free analysis results when they are no longer useful. The
1394 <tt>releaseMemory</tt> virtual method is the way to do this.</p>
1396 <p>If you are writing an analysis or any other pass that retains a significant
1397 amount of state (for use by another pass which "requires" your pass and uses the
1398 <a href="#getAnalysis">getAnalysis</a> method) you should implement
1399 <tt>releaseMEmory</tt> to, well, release the memory allocated to maintain this
1400 internal state. This method is called after the <tt>run*</tt> method for the
1401 class, before the next call of <tt>run*</tt> in your pass.</p>
1405 <!-- *********************************************************************** -->
1406 <div class="doc_section">
1407 <a name="debughints">Using GDB with dynamically loaded passes</a>
1409 <!-- *********************************************************************** -->
1411 <div class="doc_text">
1413 <p>Unfortunately, using GDB with dynamically loaded passes is not as easy as it
1414 should be. First of all, you can't set a breakpoint in a shared object that has
1415 not been loaded yet, and second of all there are problems with inlined functions
1416 in shared objects. Here are some suggestions to debugging your pass with
1419 <p>For sake of discussion, I'm going to assume that you are debugging a
1420 transformation invoked by <tt>opt</tt>, although nothing described here depends
1425 <!-- _______________________________________________________________________ -->
1426 <div class="doc_subsubsection">
1427 <a name="breakpoint">Setting a breakpoint in your pass</a>
1430 <div class="doc_text">
1432 <p>First thing you do is start <tt>gdb</tt> on the <tt>opt</tt> process:</p>
1437 Copyright 2000 Free Software Foundation, Inc.
1438 GDB is free software, covered by the GNU General Public License, and you are
1439 welcome to change it and/or distribute copies of it under certain conditions.
1440 Type "show copying" to see the conditions.
1441 There is absolutely no warranty for GDB. Type "show warranty" for details.
1442 This GDB was configured as "sparc-sun-solaris2.6"...
1446 <p>Note that <tt>opt</tt> has a lot of debugging information in it, so it takes
1447 time to load. Be patient. Since we cannot set a breakpoint in our pass yet
1448 (the shared object isn't loaded until runtime), we must execute the process, and
1449 have it stop before it invokes our pass, but after it has loaded the shared
1450 object. The most foolproof way of doing this is to set a breakpoint in
1451 <tt>PassManager::run</tt> and then run the process with the arguments you
1455 (gdb) <b>break PassManager::run</b>
1456 Breakpoint 1 at 0x2413bc: file Pass.cpp, line 70.
1457 (gdb) <b>run test.bc -load $(LLVMTOP)/llvm/lib/Debug/[libname].so -[passoption]</b>
1458 Starting program: opt test.bc -load $(LLVMTOP)/llvm/lib/Debug/[libname].so -[passoption]
1459 Breakpoint 1, PassManager::run (this=0xffbef174, M=@0x70b298) at Pass.cpp:70
1460 70 bool PassManager::run(Module &M) { return PM->run(M); }
1464 <p>Once the <tt>opt</tt> stops in the <tt>PassManager::run</tt> method you are
1465 now free to set breakpoints in your pass so that you can trace through execution
1466 or do other standard debugging stuff.</p>
1470 <!-- _______________________________________________________________________ -->
1471 <div class="doc_subsubsection">
1472 <a name="debugmisc">Miscellaneous Problems</a>
1475 <div class="doc_text">
1477 <p>Once you have the basics down, there are a couple of problems that GDB has,
1478 some with solutions, some without.</p>
1481 <li>Inline functions have bogus stack information. In general, GDB does a
1482 pretty good job getting stack traces and stepping through inline functions.
1483 When a pass is dynamically loaded however, it somehow completely loses this
1484 capability. The only solution I know of is to de-inline a function (move it
1485 from the body of a class to a .cpp file).</li>
1487 <li>Restarting the program breaks breakpoints. After following the information
1488 above, you have succeeded in getting some breakpoints planted in your pass. Nex
1489 thing you know, you restart the program (i.e., you type '<tt>run</tt>' again),
1490 and you start getting errors about breakpoints being unsettable. The only way I
1491 have found to "fix" this problem is to <tt>delete</tt> the breakpoints that are
1492 already set in your pass, run the program, and re-set the breakpoints once
1493 execution stops in <tt>PassManager::run</tt>.</li>
1497 <p>Hopefully these tips will help with common case debugging situations. If
1498 you'd like to contribute some tips of your own, just contact <a
1499 href="mailto:sabre@nondot.org">Chris</a>.</p>
1503 <!-- *********************************************************************** -->
1504 <div class="doc_section">
1505 <a name="future">Future extensions planned</a>
1507 <!-- *********************************************************************** -->
1509 <div class="doc_text">
1511 <p>Although the LLVM Pass Infrastructure is very capable as it stands, and does
1512 some nifty stuff, there are things we'd like to add in the future. Here is
1513 where we are going:</p>
1517 <!-- _______________________________________________________________________ -->
1518 <div class="doc_subsubsection">
1519 <a name="SMP">Multithreaded LLVM</a>
1522 <div class="doc_text">
1524 <p>Multiple CPU machines are becoming more common and compilation can never be
1525 fast enough: obviously we should allow for a multithreaded compiler. Because of
1526 the semantics defined for passes above (specifically they cannot maintain state
1527 across invocations of their <tt>run*</tt> methods), a nice clean way to
1528 implement a multithreaded compiler would be for the <tt>PassManager</tt> class
1529 to create multiple instances of each pass object, and allow the separate
1530 instances to be hacking on different parts of the program at the same time.</p>
1532 <p>This implementation would prevent each of the passes from having to implement
1533 multithreaded constructs, requiring only the LLVM core to have locking in a few
1534 places (for global resources). Although this is a simple extension, we simply
1535 haven't had time (or multiprocessor machines, thus a reason) to implement this.
1536 Despite that, we have kept the LLVM passes SMP ready, and you should too.</p>
1540 <!-- _______________________________________________________________________ -->
1541 <div class="doc_subsubsection">
1542 <a name="ModuleSource">A new <tt>ModuleSource</tt> interface</a>
1545 <div class="doc_text">
1547 <p>Currently, the <tt>PassManager</tt>'s <tt>run</tt> method takes a <tt><a
1548 href="http://llvm.cs.uiuc.edu/doxygen/classllvm_1_1Module.html">Module</a></tt>
1549 as input, and runs all of the passes on this module. The problem with this
1550 approach is that none of the <tt>PassManager</tt> features can be used for
1551 timing and debugging the actual <b>loading</b> of the module from disk or
1554 <p>To solve this problem, eventually the <tt>PassManager</tt> class will accept
1555 a <tt>ModuleSource</tt> object instead of a Module itself. When complete, this
1556 will also allow for streaming of functions out of the bytecode representation,
1557 allowing us to avoid holding the entire program in memory at once if we only are
1558 dealing with <a href="#FunctionPass">FunctionPass</a>es.</p>
1560 <p>As part of a different issue, eventually the bytecode loader will be extended
1561 to allow on-demand loading of functions from the bytecode representation, in
1562 order to better support the runtime reoptimizer. The bytecode format is already
1563 capable of this, the loader just needs to be reworked a bit.</p>
1567 <!-- _______________________________________________________________________ -->
1568 <div class="doc_subsubsection">
1569 <a name="PassFunctionPass"><tt>Pass</tt>es requiring <tt>FunctionPass</tt>es</a>
1572 <div class="doc_text">
1574 <p>Currently it is illegal for a <a href="#Pass"><tt>Pass</tt></a> to require a
1575 <a href="#FunctionPass"><tt>FunctionPass</tt></a>. This is because there is
1576 only one instance of the <a href="#FunctionPass"><tt>FunctionPass</tt></a>
1577 object ever created, thus nowhere to store information for all of the functions
1578 in the program at the same time. Although this has come up a couple of times
1579 before, this has always been worked around by factoring one big complicated pass
1580 into a global and an interprocedural part, both of which are distinct. In the
1581 future, it would be nice to have this though.</p>
1583 <p>Note that it is no problem for a <a
1584 href="#FunctionPass"><tt>FunctionPass</tt></a> to require the results of a <a
1585 href="#Pass"><tt>Pass</tt></a>, only the other way around.</p>
1589 <!-- *********************************************************************** -->
1592 <a href="http://jigsaw.w3.org/css-validator/check/referer"><img
1593 src="http://jigsaw.w3.org/css-validator/images/vcss" alt="Valid CSS!"></a>
1594 <a href="http://validator.w3.org/check/referer"><img
1595 src="http://www.w3.org/Icons/valid-html401" alt="Valid HTML 4.01!" /></a>
1597 <a href="mailto:sabre@nondot.org">Chris Lattner</a><br>
1598 <a href="http://llvm.cs.uiuc.edu">The LLVM Compiler Infrastructure</a><br>
1599 Last modified: $Date$