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="#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(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="#passStatistics">Pass Statistics</a>
81 <li><a href="#passmanager">What PassManager does</a>
83 <li><a href="#releaseMemory">The <tt>releaseMemory</tt> method</a></li>
85 <li><a href="#debughints">Using GDB with dynamically loaded passes</a>
87 <li><a href="#breakpoint">Setting a breakpoint in your pass</a></li>
88 <li><a href="#debugmisc">Miscellaneous Problems</a></li>
90 <li><a href="#future">Future extensions planned</a>
92 <li><a href="#SMP">Multithreaded LLVM</a></li>
93 <li><a href="#PassFunctionPass"><tt>ModulePass</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 should inherit from
120 the <tt><a href="#ModulePass">ModulePass</a></tt>, <tt><a
121 href="#CallGraphSCCPass">CallGraphSCCPass</a></tt>, <tt><a
122 href="#FunctionPass">FunctionPass</a></tt>, or <tt><a
123 href="#BasicBlockPass">BasicBlockPass</a></tt> classes, which gives the system
124 more information about what your pass does, and how it can be combined with
125 other passes. One of the main features of the LLVM Pass Framework is that it
126 schedules passes to run in an efficient way based on the constraints that your
127 pass meets (which are indicated by which class they derive from).</p>
129 <p>We start by showing you how to construct a pass, everything from setting up
130 the code, to compiling, loading, and executing it. After the basics are down,
131 more advanced features are discussed.</p>
135 <!-- *********************************************************************** -->
136 <div class="doc_section">
137 <a name="quickstart">Quick Start - Writing hello world</a>
139 <!-- *********************************************************************** -->
141 <div class="doc_text">
143 <p>Here we describe how to write the "hello world" of passes. The "Hello" pass
144 is designed to simply print out the name of non-external functions that exist in
145 the program being compiled. It does not modify the program at all, it just
146 inspects it. The source code and files for this pass are available in the LLVM
147 source tree in the <tt>lib/Transforms/Hello</tt> directory.</p>
151 <!-- ======================================================================= -->
152 <div class="doc_subsection">
153 <a name="makefile">Setting up the build environment</a>
156 <div class="doc_text">
158 <p>First thing you need to do is create a new directory somewhere in the LLVM
159 source base. For this example, we'll assume that you made
160 "<tt>lib/Transforms/Hello</tt>". The first thing you must do is set up a build
161 script (Makefile) that will compile the source code for the new pass. To do
162 this, copy this into "<tt>Makefile</tt>":</p>
167 # Makefile for hello pass
169 # Path to top level of LLVM heirarchy
172 # Name of the library to build
175 # Build a dynamically loadable shared object
178 # Include the makefile implementation stuff
179 include $(LEVEL)/Makefile.common
182 <p>This makefile specifies that all of the <tt>.cpp</tt> files in the current
183 directory are to be compiled and linked together into a
184 <tt>lib/Debug/libhello.so</tt> shared object that can be dynamically loaded by
185 the <tt>opt</tt> or <tt>analyze</tt> tools. If your operating system uses a
186 suffix other than .so (such as windows or Mac OS/X), the appropriate extension
189 <p>Now that we have the build scripts set up, we just need to write the code for
194 <!-- ======================================================================= -->
195 <div class="doc_subsection">
196 <a name="basiccode">Basic code required</a>
199 <div class="doc_text">
201 <p>Now that we have a way to compile our new pass, we just have to write it.
205 <b>#include</b> "<a href="http://llvm.cs.uiuc.edu/doxygen/Pass_8h-source.html">llvm/Pass.h</a>"
206 <b>#include</b> "<a href="http://llvm.cs.uiuc.edu/doxygen/Function_8h-source.html">llvm/Function.h</a>"
209 <p>Which are needed because we are writing a <tt><a
210 href="http://llvm.cs.uiuc.edu/doxygen/classllvm_1_1Pass.html">Pass</a></tt>, and
211 we are operating on <tt><a
212 href="http://llvm.cs.uiuc.edu/doxygen/classllvm_1_1Function.html">Function</a></tt>'s.</p>
216 <b>using namespace llvm;</b>
218 <p>... which is required because the functions from the include files
219 live in the llvm namespace.
228 <p>... which starts out an anonymous namespace. Anonymous namespaces are to C++
229 what the "<tt>static</tt>" keyword is to C (at global scope). It makes the
230 things declared inside of the anonymous namespace only visible to the current
231 file. If you're not familiar with them, consult a decent C++ book for more
234 <p>Next, we declare our pass itself:</p>
237 <b>struct</b> Hello : <b>public</b> <a href="#FunctionPass">FunctionPass</a> {
240 <p>This declares a "<tt>Hello</tt>" class that is a subclass of <tt><a
241 href="http://llvm.cs.uiuc.edu/doxygen/structllvm_1_1FunctionPass.html">FunctionPass</a></tt>.
242 The different builtin pass subclasses are described in detail <a
243 href="#passtype">later</a>, but for now, know that <a
244 href="#FunctionPass"><tt>FunctionPass</tt></a>'s operate a function at a
248 <b>virtual bool</b> <a href="#runOnFunction">runOnFunction</a>(Function &F) {
249 std::cerr << "<i>Hello: </i>" << F.getName() << "\n";
252 }; <i>// end of struct Hello</i>
255 <p>We declare a "<a href="#runOnFunction"><tt>runOnFunction</tt></a>" method,
256 which overloads an abstract virtual method inherited from <a
257 href="#FunctionPass"><tt>FunctionPass</tt></a>. This is where we are supposed
258 to do our thing, so we just print out our message with the name of each
262 RegisterOpt<Hello> X("<i>hello</i>", "<i>Hello World Pass</i>");
263 } <i>// end of anonymous namespace</i>
266 <p>Lastly, we register our class <tt>Hello</tt>, giving it a command line
267 argument "<tt>hello</tt>", and a name "<tt>Hello World Pass</tt>". There are
268 several different ways of <a href="#registration">registering your pass</a>,
269 depending on what it is to be used for. For "optimizations" we use the
270 <tt>RegisterOpt</tt> template.</p>
272 <p>As a whole, the <tt>.cpp</tt> file looks like:</p>
275 <b>#include</b> "<a href="http://llvm.cs.uiuc.edu/doxygen/Pass_8h-source.html">llvm/Pass.h</a>"
276 <b>#include</b> "<a href="http://llvm.cs.uiuc.edu/doxygen/Function_8h-source.html">llvm/Function.h</a>"
278 <b>using namespace llvm;</b>
281 <b>struct Hello</b> : <b>public</b> <a href="#FunctionPass">FunctionPass</a> {
282 <b>virtual bool</b> <a href="#runOnFunction">runOnFunction</a>(Function &F) {
283 std::cerr << "<i>Hello: </i>" << F.getName() << "\n";
288 RegisterOpt<Hello> X("<i>hello</i>", "<i>Hello World Pass</i>");
292 <p>Now that it's all together, compile the file with a simple "<tt>gmake</tt>"
293 command in the local directory and you should get a new
294 "<tt>lib/Debug/libhello.so</tt> file. Note that everything in this file is
295 contained in an anonymous namespace: this reflects the fact that passes are self
296 contained units that do not need external interfaces (although they can have
297 them) to be useful.</p>
301 <!-- ======================================================================= -->
302 <div class="doc_subsection">
303 <a name="running">Running a pass with <tt>opt</tt> or <tt>analyze</tt></a>
306 <div class="doc_text">
308 <p>Now that you have a brand new shiny shared object file, we can use the
309 <tt>opt</tt> command to run an LLVM program through your pass. Because you
310 registered your pass with the <tt>RegisterOpt</tt> template, you will be able to
311 use the <tt>opt</tt> tool to access it, once loaded.</p>
313 <p>To test it, follow the example at the end of the <a
314 href="GettingStarted.html">Getting Started Guide</a> to compile "Hello World" to
315 LLVM. We can now run the bytecode file (<tt>hello.bc</tt>) for the program
316 through our transformation like this (or course, any bytecode file will
320 $ opt -load ../../../lib/Debug/libhello.so -hello < hello.bc > /dev/null
326 <p>The '<tt>-load</tt>' option specifies that '<tt>opt</tt>' should load your
327 pass as a shared object, which makes '<tt>-hello</tt>' a valid command line
328 argument (which is one reason you need to <a href="#registration">register your
329 pass</a>). Because the hello pass does not modify the program in any
330 interesting way, we just throw away the result of <tt>opt</tt> (sending it to
331 <tt>/dev/null</tt>).</p>
333 <p>To see what happened to the other string you registered, try running
334 <tt>opt</tt> with the <tt>--help</tt> option:</p>
337 $ opt -load ../../../lib/Debug/libhello.so --help
338 OVERVIEW: llvm .bc -> .bc modular optimizer
340 USAGE: opt [options] <input bytecode>
343 Optimizations available:
345 -funcresolve - Resolve Functions
346 -gcse - Global Common Subexpression Elimination
347 -globaldce - Dead Global Elimination
348 <b>-hello - Hello World Pass</b>
349 -indvars - Canonicalize Induction Variables
350 -inline - Function Integration/Inlining
351 -instcombine - Combine redundant instructions
355 <p>The pass name get added as the information string for your pass, giving some
356 documentation to users of <tt>opt</tt>. Now that you have a working pass, you
357 would go ahead and make it do the cool transformations you want. Once you get
358 it all working and tested, it may become useful to find out how fast your pass
359 is. The <a href="#passManager"><tt>PassManager</tt></a> provides a nice command
360 line option (<tt>--time-passes</tt>) that allows you to get information about
361 the execution time of your pass along with the other passes you queue up. For
365 $ opt -load ../../../lib/Debug/libhello.so -hello -time-passes < hello.bc > /dev/null
369 ===============================================================================
370 ... Pass execution timing report ...
371 ===============================================================================
372 Total Execution Time: 0.02 seconds (0.0479059 wall clock)
374 ---User Time--- --System Time-- --User+System-- ---Wall Time--- --- Pass Name ---
375 0.0100 (100.0%) 0.0000 ( 0.0%) 0.0100 ( 50.0%) 0.0402 ( 84.0%) Bytecode Writer
376 0.0000 ( 0.0%) 0.0100 (100.0%) 0.0100 ( 50.0%) 0.0031 ( 6.4%) Dominator Set Construction
377 0.0000 ( 0.0%) 0.0000 ( 0.0%) 0.0000 ( 0.0%) 0.0013 ( 2.7%) Module Verifier
378 <b> 0.0000 ( 0.0%) 0.0000 ( 0.0%) 0.0000 ( 0.0%) 0.0033 ( 6.9%) Hello World Pass</b>
379 0.0100 (100.0%) 0.0100 (100.0%) 0.0200 (100.0%) 0.0479 (100.0%) TOTAL
382 <p>As you can see, our implementation above is pretty fast :). The additional
383 passes listed are automatically inserted by the '<tt>opt</tt>' tool to verify
384 that the LLVM emitted by your pass is still valid and well formed LLVM, which
385 hasn't been broken somehow.</p>
387 <p>Now that you have seen the basics of the mechanics behind passes, we can talk
388 about some more details of how they work and how to use them.</p>
392 <!-- *********************************************************************** -->
393 <div class="doc_section">
394 <a name="passtype">Pass classes and requirements</a>
396 <!-- *********************************************************************** -->
398 <div class="doc_text">
400 <p>One of the first things that you should do when designing a new pass is to
401 decide what class you should subclass for your pass. The <a
402 href="#basiccode">Hello World</a> example uses the <tt><a
403 href="#FunctionPass">FunctionPass</a></tt> class for its implementation, but we
404 did not discuss why or when this should occur. Here we talk about the classes
405 available, from the most general to the most specific.</p>
407 <p>When choosing a superclass for your Pass, you should choose the <b>most
408 specific</b> class possible, while still being able to meet the requirements
409 listed. This gives the LLVM Pass Infrastructure information necessary to
410 optimize how passes are run, so that the resultant compiler isn't unneccesarily
415 <!-- ======================================================================= -->
416 <div class="doc_subsection">
417 <a name="ImmutablePass">The <tt>ImmutablePass</tt> class</a>
420 <div class="doc_text">
422 <p>The most plain and boring type of pass is the "<tt><a
423 href="http://llvm.cs.uiuc.edu/doxygen/structllvm_1_1ImmutablePass.html">ImmutablePass</a></tt>"
424 class. This pass type is used for passes that do not have to be run, do not
425 change state, and never need to be updated. This is not a normal type of
426 transformation or analysis, but can provide information about the current
427 compiler configuration.</p>
429 <p>Although this pass class is very infrequently used, it is important for
430 providing information about the current target machine being compiled for, and
431 other static information that can affect the various transformations.</p>
433 <p><tt>ImmutablePass</tt>es never invalidate other transformations, are never
434 invalidated, and are never "run".</p>
438 <!-- ======================================================================= -->
439 <div class="doc_subsection">
440 <a name="ModulePass">The <tt>ModulePass</tt> class</a>
443 <div class="doc_text">
446 href="http://llvm.cs.uiuc.edu/doxygen/classllvm_1_1ModulePass.html">ModulePass</a></tt>"
447 class is the most general of all superclasses that you can use. Deriving from
448 <tt>ModulePass</tt> indicates that your pass uses the entire program as a unit,
449 refering to function bodies in no predictable order, or adding and removing
450 functions. Because nothing is known about the behavior of <tt>ModulePass</tt>
451 subclasses, no optimization can be done for their execution.</p>
453 <p>To write a correct <tt>ModulePass</tt> subclass, derive from
454 <tt>ModulePass</tt> and overload the <tt>runOnModule</tt> method with the
455 following signature:</p>
459 <!-- _______________________________________________________________________ -->
460 <div class="doc_subsubsection">
461 <a name="runOnModule">The <tt>runOnModule</tt> method</a>
464 <div class="doc_text">
467 <b>virtual bool</b> runOnModule(Module &M) = 0;
470 <p>The <tt>runOnModule</tt> method performs the interesting work of the pass,
471 and should return true if the module was modified by the transformation, false
476 <!-- ======================================================================= -->
477 <div class="doc_subsection">
478 <a name="CallGraphSCCPass">The <tt>CallGraphSCCPass</tt> class</a>
481 <div class="doc_text">
484 href="http://llvm.cs.uiuc.edu/doxygen/structllvm_1_1CallGraphSCCPass.html">CallGraphSCCPass</a></tt>"
485 is used by passes that need to traverse the program bottom-up on the call graph
486 (callees before callers). Deriving from CallGraphSCCPass provides some
487 mechanics for building and traversing the CallGraph, but also allows the system
488 to optimize execution of CallGraphSCCPass's. If your pass meets the
489 requirements outlined below, and doesn't meet the requirements of a <tt><a
490 href="#FunctionPass">FunctionPass</a></tt> or <tt><a
491 href="#BasicBlockPass">BasicBlockPass</a></tt>, you should derive from
492 <tt>CallGraphSCCPass</tt>.</p>
494 <p><b>TODO</b>: explain briefly what SCC, Tarjan's algo, and B-U mean.</p>
496 <p>To be explicit, <tt>CallGraphSCCPass</tt> subclasses are:</p>
500 <li>... <em>not allowed</em> to modify any <tt>Function</tt>s that are not in
501 the current SCC.</li>
503 <li>... <em>allowed</em> to inspect any Function's other than those in the
504 current SCC and the direct callees of the SCC.</li>
506 <li>... <em>required</em> to preserve the current CallGraph object, updating it
507 to reflect any changes made to the program.</li>
509 <li>... <em>not allowed</em> to add or remove SCC's from the current Module,
510 though they may change the contents of an SCC.</li>
512 <li>... <em>allowed</em> to add or remove global variables from the current
515 <li>... <em>allowed</em> to maintain state across invocations of
516 <a href="#runOnSCC"><tt>runOnSCC</tt></a> (including global data).</li>
519 <p>Implementing a <tt>CallGraphSCCPass</tt> is slightly tricky in some cases
520 because it has to handle SCCs with more than one node in it. All of the virtual
521 methods described below should return true if they modified the program, or
522 false if they didn't.</p>
526 <!-- _______________________________________________________________________ -->
527 <div class="doc_subsubsection">
528 <a name="doInitialization_scc">The <tt>doInitialization(Module &)</tt>
532 <div class="doc_text">
535 <b>virtual bool</b> doInitialization(Module &M);
538 <p>The <tt>doIninitialize</tt> method is allowed to do most of the things that
539 <tt>CallGraphSCCPass</tt>'s are not allowed to do. They can add and remove
540 functions, get pointers to functions, etc. The <tt>doInitialization</tt> method
541 is designed to do simple initialization type of stuff that does not depend on
542 the SCCs being processed. The <tt>doInitialization</tt> method call is not
543 scheduled to overlap with any other pass executions (thus it should be very
548 <!-- _______________________________________________________________________ -->
549 <div class="doc_subsubsection">
550 <a name="runOnSCC">The <tt>runOnSCC</tt> method</a>
553 <div class="doc_text">
556 <b>virtual bool</b> runOnSCC(const std::vector<CallGraphNode *> &SCCM) = 0;
559 <p>The <tt>runOnSCC</tt> method performs the interesting work of the pass, and
560 should return true if the module was modified by the transformation, false
565 <!-- _______________________________________________________________________ -->
566 <div class="doc_subsubsection">
567 <a name="doFinalization_scc">The <tt>doFinalization(Module
568 &)</tt> method</a>
571 <div class="doc_text">
574 <b>virtual bool</b> doFinalization(Module &M);
577 <p>The <tt>doFinalization</tt> method is an infrequently used method that is
578 called when the pass framework has finished calling <a
579 href="#runOnFunction"><tt>runOnFunction</tt></a> for every function in the
580 program being compiled.</p>
584 <!-- ======================================================================= -->
585 <div class="doc_subsection">
586 <a name="FunctionPass">The <tt>FunctionPass</tt> class</a>
589 <div class="doc_text">
591 <p>In contrast to <tt>ModulePass</tt> subclasses, <tt><a
592 href="http://llvm.cs.uiuc.edu/doxygen/classllvm_1_1Pass.html">FunctionPass</a></tt>
593 subclasses do have a predictable, local behavior that can be expected by the
594 system. All <tt>FunctionPass</tt> execute on each function in the program
595 independent of all of the other functions in the program.
596 <tt>FunctionPass</tt>'s do not require that they are executed in a particular
597 order, and <tt>FunctionPass</tt>'s do not modify external functions.</p>
599 <p>To be explicit, <tt>FunctionPass</tt> subclasses are not allowed to:</p>
602 <li>Modify a Function other than the one currently being processed.</li>
603 <li>Add or remove Function's from the current Module.</li>
604 <li>Add or remove global variables from the current Module.</li>
605 <li>Maintain state across invocations of
606 <a href="#runOnFunction"><tt>runOnFunction</tt></a> (including global data)</li>
609 <p>Implementing a <tt>FunctionPass</tt> is usually straightforward (See the <a
610 href="#basiccode">Hello World</a> pass for example). <tt>FunctionPass</tt>'s
611 may overload three virtual methods to do their work. All of these methods
612 should return true if they modified the program, or false if they didn't.</p>
616 <!-- _______________________________________________________________________ -->
617 <div class="doc_subsubsection">
618 <a name="doInitialization_mod">The <tt>doInitialization(Module &)</tt>
622 <div class="doc_text">
625 <b>virtual bool</b> doInitialization(Module &M);
628 <p>The <tt>doIninitialize</tt> method is allowed to do most of the things that
629 <tt>FunctionPass</tt>'s are not allowed to do. They can add and remove
630 functions, get pointers to functions, etc. The <tt>doInitialization</tt> method
631 is designed to do simple initialization type of stuff that does not depend on
632 the functions being processed. The <tt>doInitialization</tt> method call is not
633 scheduled to overlap with any other pass executions (thus it should be very
636 <p>A good example of how this method should be used is the <a
637 href="http://llvm.cs.uiuc.edu/doxygen/LowerAllocations_8cpp-source.html">LowerAllocations</a>
638 pass. This pass converts <tt>malloc</tt> and <tt>free</tt> instructions into
639 platform dependent <tt>malloc()</tt> and <tt>free()</tt> function calls. It
640 uses the <tt>doInitialization</tt> method to get a reference to the malloc and
641 free functions that it needs, adding prototypes to the module if necessary.</p>
645 <!-- _______________________________________________________________________ -->
646 <div class="doc_subsubsection">
647 <a name="runOnFunction">The <tt>runOnFunction</tt> method</a>
650 <div class="doc_text">
653 <b>virtual bool</b> runOnFunction(Function &F) = 0;
656 <p>The <tt>runOnFunction</tt> method must be implemented by your subclass to do
657 the transformation or analysis work of your pass. As usual, a true value should
658 be returned if the function is modified.</p>
662 <!-- _______________________________________________________________________ -->
663 <div class="doc_subsubsection">
664 <a name="doFinalization_mod">The <tt>doFinalization(Module
665 &)</tt> method</a>
668 <div class="doc_text">
671 <b>virtual bool</b> doFinalization(Module &M);
674 <p>The <tt>doFinalization</tt> method is an infrequently used method that is
675 called when the pass framework has finished calling <a
676 href="#runOnFunction"><tt>runOnFunction</tt></a> for every function in the
677 program being compiled.</p>
681 <!-- ======================================================================= -->
682 <div class="doc_subsection">
683 <a name="BasicBlockPass">The <tt>BasicBlockPass</tt> class</a>
686 <div class="doc_text">
688 <p><tt>BasicBlockPass</tt>'s are just like <a
689 href="#FunctionPass"><tt>FunctionPass</tt></a>'s, except that they must limit
690 their scope of inspection and modification to a single basic block at a time.
691 As such, they are <b>not</b> allowed to do any of the following:</p>
694 <li>Modify or inspect any basic blocks outside of the current one</li>
695 <li>Maintain state across invocations of
696 <a href="#runOnBasicBlock"><tt>runOnBasicBlock</tt></a></li>
697 <li>Modify the constrol flow graph (by altering terminator instructions)</li>
698 <li>Any of the things verboten for
699 <a href="#FunctionPass"><tt>FunctionPass</tt></a>es.</li>
702 <p><tt>BasicBlockPass</tt>es are useful for traditional local and "peephole"
703 optimizations. They may override the same <a
704 href="#doInitialization_mod"><tt>doInitialization(Module &)</tt></a> and <a
705 href="#doFinalization_mod"><tt>doFinalization(Module &)</tt></a> methods that <a
706 href="#FunctionPass"><tt>FunctionPass</tt></a>'s have, but also have the following virtual methods that may also be implemented:</p>
710 <!-- _______________________________________________________________________ -->
711 <div class="doc_subsubsection">
712 <a name="doInitialization_fn">The <tt>doInitialization(Function
713 &)</tt> method</a>
716 <div class="doc_text">
719 <b>virtual bool</b> doInitialization(Function &F);
722 <p>The <tt>doIninitialize</tt> method is allowed to do most of the things that
723 <tt>BasicBlockPass</tt>'s are not allowed to do, but that
724 <tt>FunctionPass</tt>'s can. The <tt>doInitialization</tt> method is designed
725 to do simple initialization type of stuff that does not depend on the
726 BasicBlocks being processed. The <tt>doInitialization</tt> method call is not
727 scheduled to overlap with any other pass executions (thus it should be very
732 <!-- _______________________________________________________________________ -->
733 <div class="doc_subsubsection">
734 <a name="runOnBasicBlock">The <tt>runOnBasicBlock</tt> method</a>
737 <div class="doc_text">
740 <b>virtual bool</b> runOnBasicBlock(BasicBlock &BB) = 0;
743 <p>Override this function to do the work of the <tt>BasicBlockPass</tt>. This
744 function is not allowed to inspect or modify basic blocks other than the
745 parameter, and are not allowed to modify the CFG. A true value must be returned
746 if the basic block is modified.</p>
750 <!-- _______________________________________________________________________ -->
751 <div class="doc_subsubsection">
752 <a name="doFinalization_fn">The <tt>doFinalization(Function &)</tt>
756 <div class="doc_text">
759 <b>virtual bool</b> doFinalization(Function &F);
762 <p>The <tt>doFinalization</tt> method is an infrequently used method that is
763 called when the pass framework has finished calling <a
764 href="#runOnBasicBlock"><tt>runOnBasicBlock</tt></a> for every BasicBlock in the
765 program being compiled. This can be used to perform per-function
770 <!-- ======================================================================= -->
771 <div class="doc_subsection">
772 <a name="MachineFunctionPass">The <tt>MachineFunctionPass</tt> class</a>
775 <div class="doc_text">
777 <p>A <tt>MachineFunctionPass</tt> is a part of the LLVM code generator that
778 executes on the machine-dependent representation of each LLVM function in the
779 program. A <tt>MachineFunctionPass</tt> is also a <tt>FunctionPass</tt>, so all
780 the restrictions that apply to a <tt>FunctionPass</tt> also apply to it.
781 <tt>MachineFunctionPass</tt>es also have additional restrictions. In particular,
782 <tt>MachineFunctionPass</tt>es are not allowed to do any of the following:</p>
785 <li>Modify any LLVM Instructions, BasicBlocks or Functions.</li>
786 <li>Modify a MachineFunction other than the one currently being processed.</li>
787 <li>Add or remove MachineFunctions from the current Module.</li>
788 <li>Add or remove global variables from the current Module.</li>
789 <li>Maintain state across invocations of <a
790 href="#runOnMachineFunction"><tt>runOnMachineFunction</tt></a> (including global
796 <!-- _______________________________________________________________________ -->
797 <div class="doc_subsubsection">
798 <a name="runOnMachineFunction">The <tt>runOnMachineFunction(MachineFunction
799 &MF)</tt> method</a>
802 <div class="doc_text">
805 <b>virtual bool</b> runOnMachineFunction(MachineFunction &MF) = 0;
808 <p><tt>runOnMachineFunction</tt> can be considered the main entry point of a
809 <tt>MachineFunctionPass</tt>; that is, you should override this method to do the
810 work of your <tt>MachineFunctionPass</tt>.</p>
812 <p>The <tt>runOnMachineFunction</tt> method is called on every
813 <tt>MachineFunction</tt> in a <tt>Module</tt>, so that the
814 <tt>MachineFunctionPass</tt> may perform optimizations on the machine-dependent
815 representation of the function. If you want to get at the LLVM <tt>Function</tt>
816 for the <tt>MachineFunction</tt> you're working on, use
817 <tt>MachineFunction</tt>'s <tt>getFunction()</tt> accessor method -- but
818 remember, you may not modify the LLVM <tt>Function</tt> or its contents from a
819 <tt>MachineFunctionPass</tt>.</p>
823 <!-- *********************************************************************** -->
824 <div class="doc_section">
825 <a name="registration">Pass registration</a>
827 <!-- *********************************************************************** -->
829 <div class="doc_text">
831 <p>In the <a href="#basiccode">Hello World</a> example pass we illustrated how
832 pass registration works, and discussed some of the reasons that it is used and
833 what it does. Here we discuss how and why passes are registered.</p>
835 <p>Passes can be registered in several different ways. Depending on the general
836 classification of the pass, you should use one of the following templates to
837 register the pass:</p>
840 <li><b><tt>RegisterOpt</tt></b> - This template should be used when you are
841 registering a pass that logically should be available for use in the
842 '<tt>opt</tt>' utility.</li>
844 <li><b><tt>RegisterAnalysis</tt></b> - This template should be used when you are
845 registering a pass that logically should be available for use in the
846 '<tt>analyze</tt>' utility.</li>
848 <li><b><tt>RegisterPass</tt></b> - This is the generic form of the
849 <tt>Register*</tt> templates that should be used if you want your pass listed by
850 multiple or no utilities. This template takes an extra third argument that
851 specifies which tools it should be listed in. See the <a
852 href="http://llvm.cs.uiuc.edu/doxygen/PassSupport_8h-source.html">PassSupport.h</a>
853 file for more information.</li>
857 <p>Regardless of how you register your pass, you must specify at least two
858 parameters. The first parameter is the name of the pass that is to be used on
859 the command line to specify that the pass should be added to a program (for
860 example <tt>opt</tt> or <tt>analyze</tt>). The second argument is the name of
861 the pass, which is to be used for the <tt>--help</tt> output of programs, as
862 well as for debug output generated by the <tt>--debug-pass</tt> option.</p>
864 <p>If a pass is registered to be used by the <tt>analyze</tt> utility, you
865 should implement the virtual <tt>print</tt> method:</p>
869 <!-- _______________________________________________________________________ -->
870 <div class="doc_subsubsection">
871 <a name="print">The <tt>print</tt> method</a>
874 <div class="doc_text">
877 <b>virtual void</b> print(std::ostream &O, <b>const</b> Module *M) <b>const</b>;
880 <p>The <tt>print</tt> method must be implemented by "analyses" in order to print
881 a human readable version of the analysis results. This is useful for debugging
882 an analysis itself, as well as for other people to figure out how an analysis
883 works. The <tt>analyze</tt> tool uses this method to generate its output.</p>
885 <p>The <tt>ostream</tt> parameter specifies the stream to write the results on,
886 and the <tt>Module</tt> parameter gives a pointer to the top level module of the
887 program that has been analyzed. Note however that this pointer may be null in
888 certain circumstances (such as calling the <tt>Pass::dump()</tt> from a
889 debugger), so it should only be used to enhance debug output, it should not be
894 <!-- *********************************************************************** -->
895 <div class="doc_section">
896 <a name="interaction">Specifying interactions between passes</a>
898 <!-- *********************************************************************** -->
900 <div class="doc_text">
902 <p>One of the main responsibilities of the <tt>PassManager</tt> is the make sure
903 that passes interact with each other correctly. Because <tt>PassManager</tt>
904 tries to <a href="#passmanager">optimize the execution of passes</a> it must
905 know how the passes interact with each other and what dependencies exist between
906 the various passes. To track this, each pass can declare the set of passes that
907 are required to be executed before the current pass, and the passes which are
908 invalidated by the current pass.</p>
910 <p>Typically this functionality is used to require that analysis results are
911 computed before your pass is run. Running arbitrary transformation passes can
912 invalidate the computed analysis results, which is what the invalidation set
913 specifies. If a pass does not implement the <tt><a
914 href="#getAnalysisUsage">getAnalysisUsage</a></tt> method, it defaults to not
915 having any prerequisite passes, and invalidating <b>all</b> other passes.</p>
919 <!-- _______________________________________________________________________ -->
920 <div class="doc_subsubsection">
921 <a name="getAnalysisUsage">The <tt>getAnalysisUsage</tt> method</a>
924 <div class="doc_text">
927 <b>virtual void</b> getAnalysisUsage(AnalysisUsage &Info) <b>const</b>;
930 <p>By implementing the <tt>getAnalysisUsage</tt> method, the required and
931 invalidated sets may be specified for your transformation. The implementation
932 should fill in the <tt><a
933 href="http://llvm.cs.uiuc.edu/doxygen/classllvm_1_1AnalysisUsage.html">AnalysisUsage</a></tt>
934 object with information about which passes are required and not invalidated. To
935 do this, a pass may call any of the following methods on the AnalysisUsage
939 <!-- _______________________________________________________________________ -->
940 <div class="doc_subsubsection">
941 <a name="AU::addRequired">The <tt>AnalysisUsage::addRequired<></tt> and <tt>AnalysisUsage::addRequiredTransitive<></tt> methods</a>
944 <div class="doc_text">
946 If you pass requires a previous pass to be executed (an analysis for example),
947 it can use one of these methods to arrange for it to be run before your pass.
948 LLVM has many different types of analyses and passes that can be required,
949 spaning the range from <tt>DominatorSet</tt> to <tt>BreakCriticalEdges</tt>.
950 requiring <tt>BreakCriticalEdges</tt>, for example, guarantees that there will
951 be no critical edges in the CFG when your pass has been run.
955 Some analyses chain to other analyses to do their job. For example, an <a
956 href="AliasAnalysis.html">AliasAnalysis</a> implementation is required to <a
957 href="AliasAnalysis.html#chaining">chain</a> to other alias analysis passes. In
958 cases where analyses chain, the <tt>addRequiredTransitive</tt> method should be
959 used instead of the <tt>addRequired</tt> method. This informs the PassManager
960 that the transitively required pass should be alive as long as the requiring
965 <!-- _______________________________________________________________________ -->
966 <div class="doc_subsubsection">
967 <a name="AU::addPreserved">The <tt>AnalysisUsage::addPreserved<></tt> method</a>
970 <div class="doc_text">
972 One of the jobs of the PassManager is to optimize how and when analyses are run.
973 In particular, it attempts to avoid recomputing data unless it needs to. For
974 this reason, passes are allowed to declare that they preserve (i.e., they don't
975 invalidate) an existing analysis if it's available. For example, a simple
976 constant folding pass would not modify the CFG, so it can't possible effect the
977 results of dominator analysis. By default, all passes are assumed to invalidate
982 The <tt>AnalysisUsage</tt> class provides several methods which are useful in
983 certain circumstances that are related to <tt>addPreserved</tt>. In particular,
984 the <tt>setPreservesAll</tt> method can be called to indicate that the pass does
985 not modify the LLVM program at all (which is true for analyses), and the
986 <tt>setPreservesCFG</tt> method can be used by transformations that change
987 instructions in the program but do not modify the CFG or terminator instructions
988 (note that this property is implicitly set for <a
989 href="#BasicBlockPass">BasicBlockPass</a>'s).
993 <tt>addPreserved</tt> is particularly useful for transformations like
994 <tt>BreakCriticalEdges</tt>. This pass knows how to update a small set of loop
995 and dominator related analyses if they exist, so it can preserve them, despite
996 the fact that it hacks on the CFG.
1000 <!-- _______________________________________________________________________ -->
1001 <div class="doc_subsubsection">
1002 <a name="AU::examples">Example implementations of <tt>getAnalysisUsage</tt></a>
1005 <div class="doc_text">
1008 <i>// This is an example implementation from an analysis, which does not modify
1009 // the program at all, yet has a prerequisite.</i>
1010 <b>void</b> <a href="http://llvm.cs.uiuc.edu/doxygen/structllvm_1_1PostDominanceFrontier.html">PostDominanceFrontier</a>::getAnalysisUsage(AnalysisUsage &AU) <b>const</b> {
1011 AU.setPreservesAll();
1012 AU.addRequired<<a href="http://llvm.cs.uiuc.edu/doxygen/structllvm_1_1PostDominatorTree.html">PostDominatorTree</a>>();
1019 <i>// This example modifies the program, but does not modify the CFG</i>
1020 <b>void</b> <a href="http://llvm.cs.uiuc.edu/doxygen/structLICM.html">LICM</a>::getAnalysisUsage(AnalysisUsage &AU) <b>const</b> {
1021 AU.setPreservesCFG();
1022 AU.addRequired<<a href="http://llvm.cs.uiuc.edu/doxygen/classllvm_1_1LoopInfo.html">LoopInfo</a>>();
1028 <!-- _______________________________________________________________________ -->
1029 <div class="doc_subsubsection">
1030 <a name="getAnalysis">The <tt>getAnalysis<></tt> and <tt>getAnalysisToUpdate<></tt> methods</a>
1033 <div class="doc_text">
1035 <p>The <tt>Pass::getAnalysis<></tt> method is automatically inherited by
1036 your class, providing you with access to the passes that you declared that you
1037 required with the <a href="#getAnalysisUsage"><tt>getAnalysisUsage</tt></a>
1038 method. It takes a single template argument that specifies which pass class you
1039 want, and returns a reference to that pass. For example:</p>
1042 bool LICM::runOnFunction(Function &F) {
1043 LoopInfo &LI = getAnalysis<LoopInfo>();
1048 <p>This method call returns a reference to the pass desired. You may get a
1049 runtime assertion failure if you attempt to get an analysis that you did not
1050 declare as required in your <a
1051 href="#getAnalysisUsage"><tt>getAnalysisUsage</tt></a> implementation. This
1052 method can be called by your <tt>run*</tt> method implementation, or by any
1053 other local method invoked by your <tt>run*</tt> method.</p>
1056 If your pass is capable of updating analyses if they exist (e.g.,
1057 <tt>BreakCriticalEdges</tt>, as described above), you can use the
1058 <tt>getAnalysisToUpdate</tt> method, which returns a pointer to the analysis if
1059 it is active. For example:</p>
1063 if (DominatorSet *DS = getAnalysisToUpdate<DominatorSet>()) {
1064 <i>// A DominatorSet is active. This code will update it.</i>
1071 <!-- *********************************************************************** -->
1072 <div class="doc_section">
1073 <a name="analysisgroup">Implementing Analysis Groups</a>
1075 <!-- *********************************************************************** -->
1077 <div class="doc_text">
1079 <p>Now that we understand the basics of how passes are defined, how the are
1080 used, and how they are required from other passes, it's time to get a little bit
1081 fancier. All of the pass relationships that we have seen so far are very
1082 simple: one pass depends on one other specific pass to be run before it can run.
1083 For many applications, this is great, for others, more flexibility is
1086 <p>In particular, some analyses are defined such that there is a single simple
1087 interface to the analysis results, but multiple ways of calculating them.
1088 Consider alias analysis for example. The most trivial alias analysis returns
1089 "may alias" for any alias query. The most sophisticated analysis a
1090 flow-sensitive, context-sensitive interprocedural analysis that can take a
1091 significant amount of time to execute (and obviously, there is a lot of room
1092 between these two extremes for other implementations). To cleanly support
1093 situations like this, the LLVM Pass Infrastructure supports the notion of
1094 Analysis Groups.</p>
1098 <!-- _______________________________________________________________________ -->
1099 <div class="doc_subsubsection">
1100 <a name="agconcepts">Analysis Group Concepts</a>
1103 <div class="doc_text">
1105 <p>An Analysis Group is a single simple interface that may be implemented by
1106 multiple different passes. Analysis Groups can be given human readable names
1107 just like passes, but unlike passes, they need not derive from the <tt>Pass</tt>
1108 class. An analysis group may have one or more implementations, one of which is
1109 the "default" implementation.</p>
1111 <p>Analysis groups are used by client passes just like other passes are: the
1112 <tt>AnalysisUsage::addRequired()</tt> and <tt>Pass::getAnalysis()</tt> methods.
1113 In order to resolve this requirement, the <a href="#passmanager">PassManager</a>
1114 scans the available passes to see if any implementations of the analysis group
1115 are available. If none is available, the default implementation is created for
1116 the pass to use. All standard rules for <A href="#interaction">interaction
1117 between passes</a> still apply.</p>
1119 <p>Although <a href="#registration">Pass Registration</a> is optional for normal
1120 passes, all analysis group implementations must be registered, and must use the
1121 <A href="#registerag"><tt>RegisterAnalysisGroup</tt></a> template to join the
1122 implementation pool. Also, a default implementation of the interface
1123 <b>must</b> be registered with <A
1124 href="#registerag"><tt>RegisterAnalysisGroup</tt></a>.</p>
1126 <p>As a concrete example of an Analysis Group in action, consider the <a
1127 href="http://llvm.cs.uiuc.edu/doxygen/classllvm_1_1AliasAnalysis.html">AliasAnalysis</a>
1128 analysis group. The default implementation of the alias analysis interface (the
1130 href="http://llvm.cs.uiuc.edu/doxygen/structBasicAliasAnalysis.html">basicaa</a></tt>
1131 pass) just does a few simple checks that don't require significant analysis to
1132 compute (such as: two different globals can never alias each other, etc).
1133 Passes that use the <tt><a
1134 href="http://llvm.cs.uiuc.edu/doxygen/classllvm_1_1AliasAnalysis.html">AliasAnalysis</a></tt>
1135 interface (for example the <tt><a
1136 href="http://llvm.cs.uiuc.edu/doxygen/structGCSE.html">gcse</a></tt> pass), do
1137 not care which implementation of alias analysis is actually provided, they just
1138 use the designated interface.</p>
1140 <p>From the user's perspective, commands work just like normal. Issuing the
1141 command '<tt>opt -gcse ...</tt>' will cause the <tt>basicaa</tt> class to be
1142 instantiated and added to the pass sequence. Issuing the command '<tt>opt
1143 -somefancyaa -gcse ...</tt>' will cause the <tt>gcse</tt> pass to use the
1144 <tt>somefancyaa</tt> alias analysis (which doesn't actually exist, it's just a
1145 hypothetical example) instead.</p>
1149 <!-- _______________________________________________________________________ -->
1150 <div class="doc_subsubsection">
1151 <a name="registerag">Using <tt>RegisterAnalysisGroup</tt></a>
1154 <div class="doc_text">
1156 <p>The <tt>RegisterAnalysisGroup</tt> template is used to register the analysis
1157 group itself as well as add pass implementations to the analysis group. First,
1158 an analysis should be registered, with a human readable name provided for it.
1159 Unlike registration of passes, there is no command line argument to be specified
1160 for the Analysis Group Interface itself, because it is "abstract":</p>
1163 <b>static</b> RegisterAnalysisGroup<<a href="http://llvm.cs.uiuc.edu/doxygen/classllvm_1_1AliasAnalysis.html">AliasAnalysis</a>> A("<i>Alias Analysis</i>");
1166 <p>Once the analysis is registered, passes can declare that they are valid
1167 implementations of the interface by using the following code:</p>
1171 //<i> Analysis Group implementations <b>must</b> be registered normally...</i>
1172 RegisterOpt<FancyAA>
1173 B("<i>somefancyaa</i>", "<i>A more complex alias analysis implementation</i>");
1175 //<i> Declare that we implement the AliasAnalysis interface</i>
1176 RegisterAnalysisGroup<<a href="http://llvm.cs.uiuc.edu/doxygen/classllvm_1_1AliasAnalysis.html">AliasAnalysis</a>, FancyAA> C;
1180 <p>This just shows a class <tt>FancyAA</tt> that is registered normally, then
1181 uses the <tt>RegisterAnalysisGroup</tt> template to "join" the <tt><a
1182 href="http://llvm.cs.uiuc.edu/doxygen/classllvm_1_1AliasAnalysis.html">AliasAnalysis</a></tt>
1183 analysis group. Every implementation of an analysis group should join using
1184 this template. A single pass may join multiple different analysis groups with
1189 //<i> Analysis Group implementations <b>must</b> be registered normally...</i>
1190 RegisterOpt<<a href="http://llvm.cs.uiuc.edu/doxygen/structBasicAliasAnalysis.html">BasicAliasAnalysis</a>>
1191 D("<i>basicaa</i>", "<i>Basic Alias Analysis (default AA impl)</i>");
1193 //<i> Declare that we implement the AliasAnalysis interface</i>
1194 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;
1198 <p>Here we show how the default implementation is specified (using the extra
1199 argument to the <tt>RegisterAnalysisGroup</tt> template). There must be exactly
1200 one default implementation available at all times for an Analysis Group to be
1201 used. Here we declare that the <tt><a
1202 href="http://llvm.cs.uiuc.edu/doxygen/structBasicAliasAnalysis.html">BasicAliasAnalysis</a></tt>
1203 pass is the default implementation for the interface.</p>
1207 <!-- *********************************************************************** -->
1208 <div class="doc_section">
1209 <a name="passStatistics">Pass Statistics</a>
1211 <!-- *********************************************************************** -->
1213 <div class="doc_text">
1215 href="http:://llvm.cs.uiuc.edu/doxygen/Statistic_8h-source.html"><tt>Statistic</tt></a>
1216 class, is designed to be an easy way to expose various success
1217 metrics from passes. These statistics are printed at the end of a
1218 run, when the -stats command line option is enabled on the command
1219 line. See the <a href="http://llvm.org/docs/ProgrammersManual.html#Statistic">Statistics section</a> in the Programmer's Manual for details.
1224 <!-- *********************************************************************** -->
1225 <div class="doc_section">
1226 <a name="passmanager">What PassManager does</a>
1228 <!-- *********************************************************************** -->
1230 <div class="doc_text">
1233 href="http://llvm.cs.uiuc.edu/doxygen/PassManager_8h-source.html"><tt>PassManager</tt></a>
1235 href="http://llvm.cs.uiuc.edu/doxygen/classllvm_1_1PassManager.html">class</a>
1236 takes a list of passes, ensures their <a href="#interaction">prerequisites</a>
1237 are set up correctly, and then schedules passes to run efficiently. All of the
1238 LLVM tools that run passes use the <tt>PassManager</tt> for execution of these
1241 <p>The <tt>PassManager</tt> does two main things to try to reduce the execution
1242 time of a series of passes:</p>
1245 <li><b>Share analysis results</b> - The PassManager attempts to avoid
1246 recomputing analysis results as much as possible. This means keeping track of
1247 which analyses are available already, which analyses get invalidated, and which
1248 analyses are needed to be run for a pass. An important part of work is that the
1249 <tt>PassManager</tt> tracks the exact lifetime of all analysis results, allowing
1250 it to <a href="#releaseMemory">free memory</a> allocated to holding analysis
1251 results as soon as they are no longer needed.</li>
1253 <li><b>Pipeline the execution of passes on the program</b> - The
1254 <tt>PassManager</tt> attempts to get better cache and memory usage behavior out
1255 of a series of passes by pipelining the passes together. This means that, given
1256 a series of consequtive <a href="#FunctionPass"><tt>FunctionPass</tt></a>'s, it
1257 will execute all of the <a href="#FunctionPass"><tt>FunctionPass</tt></a>'s on
1258 the first function, then all of the <a
1259 href="#FunctionPass"><tt>FunctionPass</tt></a>es on the second function,
1260 etc... until the entire program has been run through the passes.
1262 <p>This improves the cache behavior of the compiler, because it is only touching
1263 the LLVM program representation for a single function at a time, instead of
1264 traversing the entire program. It reduces the memory consumption of compiler,
1265 because, for example, only one <a
1266 href="http://llvm.cs.uiuc.edu/doxygen/structllvm_1_1DominatorSet.html"><tt>DominatorSet</tt></a>
1267 needs to be calculated at a time. This also makes it possible some <a
1268 href="#SMP">interesting enhancements</a> in the future.</p></li>
1272 <p>The effectiveness of the <tt>PassManager</tt> is influenced directly by how
1273 much information it has about the behaviors of the passes it is scheduling. For
1274 example, the "preserved" set is intentionally conservative in the face of an
1275 unimplemented <a href="#getAnalysisUsage"><tt>getAnalysisUsage</tt></a> method.
1276 Not implementing when it should be implemented will have the effect of not
1277 allowing any analysis results to live across the execution of your pass.</p>
1279 <p>The <tt>PassManager</tt> class exposes a <tt>--debug-pass</tt> command line
1280 options that is useful for debugging pass execution, seeing how things work, and
1281 diagnosing when you should be preserving more analyses than you currently are
1282 (To get information about all of the variants of the <tt>--debug-pass</tt>
1283 option, just type '<tt>opt --help-hidden</tt>').</p>
1285 <p>By using the <tt>--debug-pass=Structure</tt> option, for example, we can see
1286 how our <a href="#basiccode">Hello World</a> pass interacts with other passes.
1287 Lets try it out with the <tt>gcse</tt> and <tt>licm</tt> passes:</p>
1290 $ opt -load ../../../lib/Debug/libhello.so -gcse -licm --debug-pass=Structure < hello.bc > /dev/null
1292 Function Pass Manager
1293 Dominator Set Construction
1294 Immediate Dominators Construction
1295 Global Common Subexpression Elimination
1296 -- Immediate Dominators Construction
1297 -- Global Common Subexpression Elimination
1298 Natural Loop Construction
1299 Loop Invariant Code Motion
1300 -- Natural Loop Construction
1301 -- Loop Invariant Code Motion
1303 -- Dominator Set Construction
1309 <p>This output shows us when passes are constructed and when the analysis
1310 results are known to be dead (prefixed with '<tt>--</tt>'). Here we see that
1311 GCSE uses dominator and immediate dominator information to do its job. The LICM
1312 pass uses natural loop information, which uses dominator sets, but not immediate
1313 dominators. Because immediate dominators are no longer useful after the GCSE
1314 pass, it is immediately destroyed. The dominator sets are then reused to
1315 compute natural loop information, which is then used by the LICM pass.</p>
1317 <p>After the LICM pass, the module verifier runs (which is automatically added
1318 by the '<tt>opt</tt>' tool), which uses the dominator set to check that the
1319 resultant LLVM code is well formed. After it finishes, the dominator set
1320 information is destroyed, after being computed once, and shared by three
1323 <p>Lets see how this changes when we run the <a href="#basiccode">Hello
1324 World</a> pass in between the two passes:</p>
1327 $ opt -load ../../../lib/Debug/libhello.so -gcse -hello -licm --debug-pass=Structure < hello.bc > /dev/null
1329 Function Pass Manager
1330 Dominator Set Construction
1331 Immediate Dominators Construction
1332 Global Common Subexpression Elimination
1333 <b>-- Dominator Set Construction</b>
1334 -- Immediate Dominators Construction
1335 -- Global Common Subexpression Elimination
1336 <b> Hello World Pass
1338 Dominator Set Construction</b>
1339 Natural Loop Construction
1340 Loop Invariant Code Motion
1341 -- Natural Loop Construction
1342 -- Loop Invariant Code Motion
1344 -- Dominator Set Construction
1353 <p>Here we see that the <a href="#basiccode">Hello World</a> pass has killed the
1354 Dominator Set pass, even though it doesn't modify the code at all! To fix this,
1355 we need to add the following <a
1356 href="#getAnalysisUsage"><tt>getAnalysisUsage</tt></a> method to our pass:</p>
1359 <i>// We don't modify the program, so we preserve all analyses</i>
1360 <b>virtual void</b> getAnalysisUsage(AnalysisUsage &AU) <b>const</b> {
1361 AU.setPreservesAll();
1365 <p>Now when we run our pass, we get this output:</p>
1368 $ opt -load ../../../lib/Debug/libhello.so -gcse -hello -licm --debug-pass=Structure < hello.bc > /dev/null
1369 Pass Arguments: -gcse -hello -licm
1371 Function Pass Manager
1372 Dominator Set Construction
1373 Immediate Dominators Construction
1374 Global Common Subexpression Elimination
1375 -- Immediate Dominators Construction
1376 -- Global Common Subexpression Elimination
1379 Natural Loop Construction
1380 Loop Invariant Code Motion
1381 -- Loop Invariant Code Motion
1382 -- Natural Loop Construction
1384 -- Dominator Set Construction
1393 <p>Which shows that we don't accidentally invalidate dominator information
1394 anymore, and therefore do not have to compute it twice.</p>
1398 <!-- _______________________________________________________________________ -->
1399 <div class="doc_subsubsection">
1400 <a name="releaseMemory">The <tt>releaseMemory</tt> method</a>
1403 <div class="doc_text">
1406 <b>virtual void</b> releaseMemory();
1409 <p>The <tt>PassManager</tt> automatically determines when to compute analysis
1410 results, and how long to keep them around for. Because the lifetime of the pass
1411 object itself is effectively the entire duration of the compilation process, we
1412 need some way to free analysis results when they are no longer useful. The
1413 <tt>releaseMemory</tt> virtual method is the way to do this.</p>
1415 <p>If you are writing an analysis or any other pass that retains a significant
1416 amount of state (for use by another pass which "requires" your pass and uses the
1417 <a href="#getAnalysis">getAnalysis</a> method) you should implement
1418 <tt>releaseMEmory</tt> to, well, release the memory allocated to maintain this
1419 internal state. This method is called after the <tt>run*</tt> method for the
1420 class, before the next call of <tt>run*</tt> in your pass.</p>
1424 <!-- *********************************************************************** -->
1425 <div class="doc_section">
1426 <a name="debughints">Using GDB with dynamically loaded passes</a>
1428 <!-- *********************************************************************** -->
1430 <div class="doc_text">
1432 <p>Unfortunately, using GDB with dynamically loaded passes is not as easy as it
1433 should be. First of all, you can't set a breakpoint in a shared object that has
1434 not been loaded yet, and second of all there are problems with inlined functions
1435 in shared objects. Here are some suggestions to debugging your pass with
1438 <p>For sake of discussion, I'm going to assume that you are debugging a
1439 transformation invoked by <tt>opt</tt>, although nothing described here depends
1444 <!-- _______________________________________________________________________ -->
1445 <div class="doc_subsubsection">
1446 <a name="breakpoint">Setting a breakpoint in your pass</a>
1449 <div class="doc_text">
1451 <p>First thing you do is start <tt>gdb</tt> on the <tt>opt</tt> process:</p>
1456 Copyright 2000 Free Software Foundation, Inc.
1457 GDB is free software, covered by the GNU General Public License, and you are
1458 welcome to change it and/or distribute copies of it under certain conditions.
1459 Type "show copying" to see the conditions.
1460 There is absolutely no warranty for GDB. Type "show warranty" for details.
1461 This GDB was configured as "sparc-sun-solaris2.6"...
1465 <p>Note that <tt>opt</tt> has a lot of debugging information in it, so it takes
1466 time to load. Be patient. Since we cannot set a breakpoint in our pass yet
1467 (the shared object isn't loaded until runtime), we must execute the process, and
1468 have it stop before it invokes our pass, but after it has loaded the shared
1469 object. The most foolproof way of doing this is to set a breakpoint in
1470 <tt>PassManager::run</tt> and then run the process with the arguments you
1474 (gdb) <b>break PassManager::run</b>
1475 Breakpoint 1 at 0x2413bc: file Pass.cpp, line 70.
1476 (gdb) <b>run test.bc -load $(LLVMTOP)/llvm/lib/Debug/[libname].so -[passoption]</b>
1477 Starting program: opt test.bc -load $(LLVMTOP)/llvm/lib/Debug/[libname].so -[passoption]
1478 Breakpoint 1, PassManager::run (this=0xffbef174, M=@0x70b298) at Pass.cpp:70
1479 70 bool PassManager::run(Module &M) { return PM->run(M); }
1483 <p>Once the <tt>opt</tt> stops in the <tt>PassManager::run</tt> method you are
1484 now free to set breakpoints in your pass so that you can trace through execution
1485 or do other standard debugging stuff.</p>
1489 <!-- _______________________________________________________________________ -->
1490 <div class="doc_subsubsection">
1491 <a name="debugmisc">Miscellaneous Problems</a>
1494 <div class="doc_text">
1496 <p>Once you have the basics down, there are a couple of problems that GDB has,
1497 some with solutions, some without.</p>
1500 <li>Inline functions have bogus stack information. In general, GDB does a
1501 pretty good job getting stack traces and stepping through inline functions.
1502 When a pass is dynamically loaded however, it somehow completely loses this
1503 capability. The only solution I know of is to de-inline a function (move it
1504 from the body of a class to a .cpp file).</li>
1506 <li>Restarting the program breaks breakpoints. After following the information
1507 above, you have succeeded in getting some breakpoints planted in your pass. Nex
1508 thing you know, you restart the program (i.e., you type '<tt>run</tt>' again),
1509 and you start getting errors about breakpoints being unsettable. The only way I
1510 have found to "fix" this problem is to <tt>delete</tt> the breakpoints that are
1511 already set in your pass, run the program, and re-set the breakpoints once
1512 execution stops in <tt>PassManager::run</tt>.</li>
1516 <p>Hopefully these tips will help with common case debugging situations. If
1517 you'd like to contribute some tips of your own, just contact <a
1518 href="mailto:sabre@nondot.org">Chris</a>.</p>
1522 <!-- *********************************************************************** -->
1523 <div class="doc_section">
1524 <a name="future">Future extensions planned</a>
1526 <!-- *********************************************************************** -->
1528 <div class="doc_text">
1530 <p>Although the LLVM Pass Infrastructure is very capable as it stands, and does
1531 some nifty stuff, there are things we'd like to add in the future. Here is
1532 where we are going:</p>
1536 <!-- _______________________________________________________________________ -->
1537 <div class="doc_subsubsection">
1538 <a name="SMP">Multithreaded LLVM</a>
1541 <div class="doc_text">
1543 <p>Multiple CPU machines are becoming more common and compilation can never be
1544 fast enough: obviously we should allow for a multithreaded compiler. Because of
1545 the semantics defined for passes above (specifically they cannot maintain state
1546 across invocations of their <tt>run*</tt> methods), a nice clean way to
1547 implement a multithreaded compiler would be for the <tt>PassManager</tt> class
1548 to create multiple instances of each pass object, and allow the separate
1549 instances to be hacking on different parts of the program at the same time.</p>
1551 <p>This implementation would prevent each of the passes from having to implement
1552 multithreaded constructs, requiring only the LLVM core to have locking in a few
1553 places (for global resources). Although this is a simple extension, we simply
1554 haven't had time (or multiprocessor machines, thus a reason) to implement this.
1555 Despite that, we have kept the LLVM passes SMP ready, and you should too.</p>
1559 <!-- _______________________________________________________________________ -->
1560 <div class="doc_subsubsection">
1561 <a name="PassFunctionPass"><tt>ModulePass</tt>es requiring <tt>FunctionPass</tt>es</a>
1564 <div class="doc_text">
1566 <p>Currently it is illegal for a <a href="#ModulePass"><tt>ModulePass</tt></a>
1567 to require a <a href="#FunctionPass"><tt>FunctionPass</tt></a>. This is because
1568 there is only one instance of the <a
1569 href="#FunctionPass"><tt>FunctionPass</tt></a> object ever created, thus nowhere
1570 to store information for all of the functions in the program at the same time.
1571 Although this has come up a couple of times before, this has always been worked
1572 around by factoring one big complicated pass into a global and an
1573 interprocedural part, both of which are distinct. In the future, it would be
1574 nice to have this though.</p>
1576 <p>Note that it is no problem for a <a
1577 href="#FunctionPass"><tt>FunctionPass</tt></a> to require the results of a <a
1578 href="#ModulePass"><tt>ModulePass</tt></a>, only the other way around.</p>
1582 <!-- *********************************************************************** -->
1585 <a href="http://jigsaw.w3.org/css-validator/check/referer"><img
1586 src="http://jigsaw.w3.org/css-validator/images/vcss" alt="Valid CSS!"></a>
1587 <a href="http://validator.w3.org/check/referer"><img
1588 src="http://www.w3.org/Icons/valid-html401" alt="Valid HTML 4.01!" /></a>
1590 <a href="mailto:sabre@nondot.org">Chris Lattner</a><br>
1591 <a href="http://llvm.cs.uiuc.edu">The LLVM Compiler Infrastructure</a><br>
1592 Last modified: $Date$