1 <!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.01//EN"
2 "http://www.w3.org/TR/html4/strict.dtd">
5 <meta http-equiv="Content-Type" content="text/html; charset=utf-8">
6 <title>Writing an LLVM Pass</title>
7 <link rel="stylesheet" href="llvm.css" type="text/css">
11 <div class="doc_title">
16 <li><a href="#introduction">Introduction - What is a pass?</a></li>
17 <li><a href="#quickstart">Quick Start - Writing hello world</a>
19 <li><a href="#makefile">Setting up the build environment</a></li>
20 <li><a href="#basiccode">Basic code required</a></li>
21 <li><a href="#running">Running a pass with <tt>opt</tt></a></li>
23 <li><a href="#passtype">Pass classes and requirements</a>
25 <li><a href="#ImmutablePass">The <tt>ImmutablePass</tt> class</a></li>
26 <li><a href="#ModulePass">The <tt>ModulePass</tt> class</a>
28 <li><a href="#runOnModule">The <tt>runOnModule</tt> method</a></li>
30 <li><a href="#CallGraphSCCPass">The <tt>CallGraphSCCPass</tt> class</a>
32 <li><a href="#doInitialization_scc">The <tt>doInitialization(CallGraph
33 &)</tt> method</a></li>
34 <li><a href="#runOnSCC">The <tt>runOnSCC</tt> method</a></li>
35 <li><a href="#doFinalization_scc">The <tt>doFinalization(CallGraph
36 &)</tt> method</a></li>
38 <li><a href="#FunctionPass">The <tt>FunctionPass</tt> class</a>
40 <li><a href="#doInitialization_mod">The <tt>doInitialization(Module
41 &)</tt> method</a></li>
42 <li><a href="#runOnFunction">The <tt>runOnFunction</tt> method</a></li>
43 <li><a href="#doFinalization_mod">The <tt>doFinalization(Module
44 &)</tt> method</a></li>
46 <li><a href="#LoopPass">The <tt>LoopPass</tt> class</a>
48 <li><a href="#doInitialization_loop">The <tt>doInitialization(Loop *,
49 LPPassManager &)</tt> method</a></li>
50 <li><a href="#runOnLoop">The <tt>runOnLoop</tt> method</a></li>
51 <li><a href="#doFinalization_loop">The <tt>doFinalization()
54 <li><a href="#BasicBlockPass">The <tt>BasicBlockPass</tt> class</a>
56 <li><a href="#doInitialization_fn">The <tt>doInitialization(Function
57 &)</tt> method</a></li>
58 <li><a href="#runOnBasicBlock">The <tt>runOnBasicBlock</tt>
60 <li><a href="#doFinalization_fn">The <tt>doFinalization(Function
61 &)</tt> method</a></li>
63 <li><a href="#MachineFunctionPass">The <tt>MachineFunctionPass</tt>
66 <li><a href="#runOnMachineFunction">The
67 <tt>runOnMachineFunction(MachineFunction &)</tt> method</a></li>
70 <li><a href="#registration">Pass Registration</a>
72 <li><a href="#print">The <tt>print</tt> method</a></li>
74 <li><a href="#interaction">Specifying interactions between passes</a>
76 <li><a href="#getAnalysisUsage">The <tt>getAnalysisUsage</tt>
78 <li><a href="#AU::addRequired">The <tt>AnalysisUsage::addRequired<></tt> and <tt>AnalysisUsage::addRequiredTransitive<></tt> methods</a></li>
79 <li><a href="#AU::addPreserved">The <tt>AnalysisUsage::addPreserved<></tt> method</a></li>
80 <li><a href="#AU::examples">Example implementations of <tt>getAnalysisUsage</tt></a></li>
81 <li><a href="#getAnalysis">The <tt>getAnalysis<></tt> and
82 <tt>getAnalysisIfAvailable<></tt> methods</a></li>
84 <li><a href="#analysisgroup">Implementing Analysis Groups</a>
86 <li><a href="#agconcepts">Analysis Group Concepts</a></li>
87 <li><a href="#registerag">Using <tt>RegisterAnalysisGroup</tt></a></li>
89 <li><a href="#passStatistics">Pass Statistics</a>
90 <li><a href="#passmanager">What PassManager does</a>
92 <li><a href="#releaseMemory">The <tt>releaseMemory</tt> method</a></li>
94 <li><a href="#registering">Registering dynamically loaded passes</a>
96 <li><a href="#registering_existing">Using existing registries</a></li>
97 <li><a href="#registering_new">Creating new registries</a></li>
99 <li><a href="#debughints">Using GDB with dynamically loaded passes</a>
101 <li><a href="#breakpoint">Setting a breakpoint in your pass</a></li>
102 <li><a href="#debugmisc">Miscellaneous Problems</a></li>
104 <li><a href="#future">Future extensions planned</a>
106 <li><a href="#SMP">Multithreaded LLVM</a></li>
110 <div class="doc_author">
111 <p>Written by <a href="mailto:sabre@nondot.org">Chris Lattner</a> and
112 <a href="mailto:jlaskey@mac.com">Jim Laskey</a></p>
115 <!-- *********************************************************************** -->
116 <div class="doc_section">
117 <a name="introduction">Introduction - What is a pass?</a>
119 <!-- *********************************************************************** -->
121 <div class="doc_text">
123 <p>The LLVM Pass Framework is an important part of the LLVM system, because LLVM
124 passes are where most of the interesting parts of the compiler exist. Passes
125 perform the transformations and optimizations that make up the compiler, they
126 build the analysis results that are used by these transformations, and they are,
127 above all, a structuring technique for compiler code.</p>
129 <p>All LLVM passes are subclasses of the <tt><a
130 href="http://llvm.org/doxygen/classllvm_1_1Pass.html">Pass</a></tt>
131 class, which implement functionality by overriding virtual methods inherited
132 from <tt>Pass</tt>. Depending on how your pass works, you should inherit from
133 the <tt><a href="#ModulePass">ModulePass</a></tt>, <tt><a
134 href="#CallGraphSCCPass">CallGraphSCCPass</a></tt>, <tt><a
135 href="#FunctionPass">FunctionPass</a></tt>, or <tt><a
136 href="#LoopPass">LoopPass</a></tt>, or <tt><a
137 href="#BasicBlockPass">BasicBlockPass</a></tt> classes, which gives the system
138 more information about what your pass does, and how it can be combined with
139 other passes. One of the main features of the LLVM Pass Framework is that it
140 schedules passes to run in an efficient way based on the constraints that your
141 pass meets (which are indicated by which class they derive from).</p>
143 <p>We start by showing you how to construct a pass, everything from setting up
144 the code, to compiling, loading, and executing it. After the basics are down,
145 more advanced features are discussed.</p>
149 <!-- *********************************************************************** -->
150 <div class="doc_section">
151 <a name="quickstart">Quick Start - Writing hello world</a>
153 <!-- *********************************************************************** -->
155 <div class="doc_text">
157 <p>Here we describe how to write the "hello world" of passes. The "Hello" pass
158 is designed to simply print out the name of non-external functions that exist in
159 the program being compiled. It does not modify the program at all, it just
160 inspects it. The source code and files for this pass are available in the LLVM
161 source tree in the <tt>lib/Transforms/Hello</tt> directory.</p>
165 <!-- ======================================================================= -->
166 <div class="doc_subsection">
167 <a name="makefile">Setting up the build environment</a>
170 <div class="doc_text">
172 <p>First, you need to create a new directory somewhere in the LLVM source
173 base. For this example, we'll assume that you made
174 <tt>lib/Transforms/Hello</tt>. Next, you must set up a build script
175 (Makefile) that will compile the source code for the new pass. To do this,
176 copy the following into <tt>Makefile</tt>:</p>
179 <div class="doc_code"><pre>
180 # Makefile for hello pass
182 # Path to top level of LLVM hierarchy
185 # Name of the library to build
188 # Make the shared library become a loadable module so the tools can
189 # dlopen/dlsym on the resulting library.
192 # Tell the build system which LLVM libraries your pass needs. You'll probably
193 # need at least LLVMSystem.a, LLVMSupport.a, LLVMCore.a but possibly several
195 LLVMLIBS = LLVMCore.a LLVMSupport.a LLVMSystem.a
197 # Include the makefile implementation stuff
198 include $(LEVEL)/Makefile.common
201 <p>This makefile specifies that all of the <tt>.cpp</tt> files in the current
202 directory are to be compiled and linked together into a
203 <tt>Debug/lib/Hello.so</tt> shared object that can be dynamically loaded by
204 the <tt>opt</tt> or <tt>bugpoint</tt> tools via their <tt>-load</tt> options.
205 If your operating system uses a suffix other than .so (such as windows or
206 Mac OS/X), the appropriate extension will be used.</p>
208 <p>Now that we have the build scripts set up, we just need to write the code for
213 <!-- ======================================================================= -->
214 <div class="doc_subsection">
215 <a name="basiccode">Basic code required</a>
218 <div class="doc_text">
220 <p>Now that we have a way to compile our new pass, we just have to write it.
223 <div class="doc_code"><pre>
224 <b>#include</b> "<a href="http://llvm.org/doxygen/Pass_8h-source.html">llvm/Pass.h</a>"
225 <b>#include</b> "<a href="http://llvm.org/doxygen/Function_8h-source.html">llvm/Function.h</a>"
226 <b>#include</b> "<a href="http://llvm.org/doxygen/raw__ostream_8h.html">llvm/Support/raw_ostream.h</a>"
229 <p>Which are needed because we are writing a <tt><a
230 href="http://llvm.org/doxygen/classllvm_1_1Pass.html">Pass</a></tt>,
231 we are operating on <tt><a
232 href="http://llvm.org/doxygen/classllvm_1_1Function.html">Function</a></tt>'s,
233 and we will be doing some printing.</p>
236 <div class="doc_code"><pre>
237 <b>using namespace llvm;</b>
239 <p>... which is required because the functions from the include files
240 live in the llvm namespace.
245 <div class="doc_code"><pre>
249 <p>... which starts out an anonymous namespace. Anonymous namespaces are to C++
250 what the "<tt>static</tt>" keyword is to C (at global scope). It makes the
251 things declared inside of the anonymous namespace only visible to the current
252 file. If you're not familiar with them, consult a decent C++ book for more
255 <p>Next, we declare our pass itself:</p>
257 <div class="doc_code"><pre>
258 <b>struct</b> Hello : <b>public</b> <a href="#FunctionPass">FunctionPass</a> {
261 <p>This declares a "<tt>Hello</tt>" class that is a subclass of <tt><a
262 href="http://llvm.org/doxygen/classllvm_1_1FunctionPass.html">FunctionPass</a></tt>.
263 The different builtin pass subclasses are described in detail <a
264 href="#passtype">later</a>, but for now, know that <a
265 href="#FunctionPass"><tt>FunctionPass</tt></a>'s operate a function at a
268 <div class="doc_code"><pre>
270 Hello() : FunctionPass(&ID) {}
273 <p> This declares pass identifier used by LLVM to identify pass. This allows LLVM to
274 avoid using expensive C++ runtime information.</p>
276 <div class="doc_code"><pre>
277 <b>virtual bool</b> <a href="#runOnFunction">runOnFunction</a>(Function &F) {
278 errs() << "<i>Hello: </i>" << F.getName() << "\n";
281 }; <i>// end of struct Hello</i>
284 <p>We declare a "<a href="#runOnFunction"><tt>runOnFunction</tt></a>" method,
285 which overloads an abstract virtual method inherited from <a
286 href="#FunctionPass"><tt>FunctionPass</tt></a>. This is where we are supposed
287 to do our thing, so we just print out our message with the name of each
290 <div class="doc_code"><pre>
294 <p> We initialize pass ID here. LLVM uses ID's address to identify pass so
295 initialization value is not important.</p>
297 <div class="doc_code"><pre>
298 RegisterPass<Hello> X("<i>hello</i>", "<i>Hello World Pass</i>",
299 false /* Only looks at CFG */,
300 false /* Analysis Pass */);
301 } <i>// end of anonymous namespace</i>
304 <p>Lastly, we <a href="#registration">register our class</a> <tt>Hello</tt>,
305 giving it a command line
306 argument "<tt>hello</tt>", and a name "<tt>Hello World Pass</tt>".
307 Last two RegisterPass arguments are optional. Their default value is false.
308 If a pass walks CFG without modifying it then third argument is set to true.
309 If a pass is an analysis pass, for example dominator tree pass, then true
310 is supplied as fourth argument. </p>
312 <p>As a whole, the <tt>.cpp</tt> file looks like:</p>
314 <div class="doc_code"><pre>
315 <b>#include</b> "<a href="http://llvm.org/doxygen/Pass_8h-source.html">llvm/Pass.h</a>"
316 <b>#include</b> "<a href="http://llvm.org/doxygen/Function_8h-source.html">llvm/Function.h</a>"
317 <b>#include</b> "<a href="http://llvm.org/doxygen/raw__ostream_8h.html">llvm/Support/raw_ostream.h</a>"
319 <b>using namespace llvm;</b>
322 <b>struct Hello</b> : <b>public</b> <a href="#FunctionPass">FunctionPass</a> {
325 Hello() : FunctionPass(&ID) {}
327 <b>virtual bool</b> <a href="#runOnFunction">runOnFunction</a>(Function &F) {
328 errs() << "<i>Hello: </i>" << F.getName() << "\n";
334 RegisterPass<Hello> X("<i>hello</i>", "<i>Hello World Pass</i>");
338 <p>Now that it's all together, compile the file with a simple "<tt>gmake</tt>"
339 command in the local directory and you should get a new
340 "<tt>Debug/lib/Hello.so</tt> file. Note that everything in this file is
341 contained in an anonymous namespace: this reflects the fact that passes are self
342 contained units that do not need external interfaces (although they can have
343 them) to be useful.</p>
347 <!-- ======================================================================= -->
348 <div class="doc_subsection">
349 <a name="running">Running a pass with <tt>opt</tt></a>
352 <div class="doc_text">
354 <p>Now that you have a brand new shiny shared object file, we can use the
355 <tt>opt</tt> command to run an LLVM program through your pass. Because you
356 registered your pass with the <tt>RegisterPass</tt> template, you will be able to
357 use the <tt>opt</tt> tool to access it, once loaded.</p>
359 <p>To test it, follow the example at the end of the <a
360 href="GettingStarted.html">Getting Started Guide</a> to compile "Hello World" to
361 LLVM. We can now run the bitcode file (<tt>hello.bc</tt>) for the program
362 through our transformation like this (or course, any bitcode file will
365 <div class="doc_code"><pre>
366 $ opt -load ../../../Debug/lib/Hello.so -hello < hello.bc > /dev/null
372 <p>The '<tt>-load</tt>' option specifies that '<tt>opt</tt>' should load your
373 pass as a shared object, which makes '<tt>-hello</tt>' a valid command line
374 argument (which is one reason you need to <a href="#registration">register your
375 pass</a>). Because the hello pass does not modify the program in any
376 interesting way, we just throw away the result of <tt>opt</tt> (sending it to
377 <tt>/dev/null</tt>).</p>
379 <p>To see what happened to the other string you registered, try running
380 <tt>opt</tt> with the <tt>-help</tt> option:</p>
382 <div class="doc_code"><pre>
383 $ opt -load ../../../Debug/lib/Hello.so -help
384 OVERVIEW: llvm .bc -> .bc modular optimizer
386 USAGE: opt [options] <input bitcode>
389 Optimizations available:
391 -funcresolve - Resolve Functions
392 -gcse - Global Common Subexpression Elimination
393 -globaldce - Dead Global Elimination
394 <b>-hello - Hello World Pass</b>
395 -indvars - Canonicalize Induction Variables
396 -inline - Function Integration/Inlining
397 -instcombine - Combine redundant instructions
401 <p>The pass name get added as the information string for your pass, giving some
402 documentation to users of <tt>opt</tt>. Now that you have a working pass, you
403 would go ahead and make it do the cool transformations you want. Once you get
404 it all working and tested, it may become useful to find out how fast your pass
405 is. The <a href="#passManager"><tt>PassManager</tt></a> provides a nice command
406 line option (<tt>--time-passes</tt>) that allows you to get information about
407 the execution time of your pass along with the other passes you queue up. For
410 <div class="doc_code"><pre>
411 $ opt -load ../../../Debug/lib/Hello.so -hello -time-passes < hello.bc > /dev/null
415 ===============================================================================
416 ... Pass execution timing report ...
417 ===============================================================================
418 Total Execution Time: 0.02 seconds (0.0479059 wall clock)
420 ---User Time--- --System Time-- --User+System-- ---Wall Time--- --- Pass Name ---
421 0.0100 (100.0%) 0.0000 ( 0.0%) 0.0100 ( 50.0%) 0.0402 ( 84.0%) Bitcode Writer
422 0.0000 ( 0.0%) 0.0100 (100.0%) 0.0100 ( 50.0%) 0.0031 ( 6.4%) Dominator Set Construction
423 0.0000 ( 0.0%) 0.0000 ( 0.0%) 0.0000 ( 0.0%) 0.0013 ( 2.7%) Module Verifier
424 <b> 0.0000 ( 0.0%) 0.0000 ( 0.0%) 0.0000 ( 0.0%) 0.0033 ( 6.9%) Hello World Pass</b>
425 0.0100 (100.0%) 0.0100 (100.0%) 0.0200 (100.0%) 0.0479 (100.0%) TOTAL
428 <p>As you can see, our implementation above is pretty fast :). The additional
429 passes listed are automatically inserted by the '<tt>opt</tt>' tool to verify
430 that the LLVM emitted by your pass is still valid and well formed LLVM, which
431 hasn't been broken somehow.</p>
433 <p>Now that you have seen the basics of the mechanics behind passes, we can talk
434 about some more details of how they work and how to use them.</p>
438 <!-- *********************************************************************** -->
439 <div class="doc_section">
440 <a name="passtype">Pass classes and requirements</a>
442 <!-- *********************************************************************** -->
444 <div class="doc_text">
446 <p>One of the first things that you should do when designing a new pass is to
447 decide what class you should subclass for your pass. The <a
448 href="#basiccode">Hello World</a> example uses the <tt><a
449 href="#FunctionPass">FunctionPass</a></tt> class for its implementation, but we
450 did not discuss why or when this should occur. Here we talk about the classes
451 available, from the most general to the most specific.</p>
453 <p>When choosing a superclass for your Pass, you should choose the <b>most
454 specific</b> class possible, while still being able to meet the requirements
455 listed. This gives the LLVM Pass Infrastructure information necessary to
456 optimize how passes are run, so that the resultant compiler isn't unnecessarily
461 <!-- ======================================================================= -->
462 <div class="doc_subsection">
463 <a name="ImmutablePass">The <tt>ImmutablePass</tt> class</a>
466 <div class="doc_text">
468 <p>The most plain and boring type of pass is the "<tt><a
469 href="http://llvm.org/doxygen/classllvm_1_1ImmutablePass.html">ImmutablePass</a></tt>"
470 class. This pass type is used for passes that do not have to be run, do not
471 change state, and never need to be updated. This is not a normal type of
472 transformation or analysis, but can provide information about the current
473 compiler configuration.</p>
475 <p>Although this pass class is very infrequently used, it is important for
476 providing information about the current target machine being compiled for, and
477 other static information that can affect the various transformations.</p>
479 <p><tt>ImmutablePass</tt>es never invalidate other transformations, are never
480 invalidated, and are never "run".</p>
484 <!-- ======================================================================= -->
485 <div class="doc_subsection">
486 <a name="ModulePass">The <tt>ModulePass</tt> class</a>
489 <div class="doc_text">
492 href="http://llvm.org/doxygen/classllvm_1_1ModulePass.html">ModulePass</a></tt>"
493 class is the most general of all superclasses that you can use. Deriving from
494 <tt>ModulePass</tt> indicates that your pass uses the entire program as a unit,
495 referring to function bodies in no predictable order, or adding and removing
496 functions. Because nothing is known about the behavior of <tt>ModulePass</tt>
497 subclasses, no optimization can be done for their execution.</p>
499 <p>A module pass can use function level passes (e.g. dominators) using
500 the getAnalysis interface
501 <tt>getAnalysis<DominatorTree>(llvm::Function *)</tt> to provide the
502 function to retrieve analysis result for, if the function pass does not require
503 any module or immutable passes. Note that this can only be done for functions for which the
504 analysis ran, e.g. in the case of dominators you should only ask for the
505 DominatorTree for function definitions, not declarations.</p>
507 <p>To write a correct <tt>ModulePass</tt> subclass, derive from
508 <tt>ModulePass</tt> and overload the <tt>runOnModule</tt> method with the
509 following signature:</p>
513 <!-- _______________________________________________________________________ -->
514 <div class="doc_subsubsection">
515 <a name="runOnModule">The <tt>runOnModule</tt> method</a>
518 <div class="doc_text">
520 <div class="doc_code"><pre>
521 <b>virtual bool</b> runOnModule(Module &M) = 0;
524 <p>The <tt>runOnModule</tt> method performs the interesting work of the pass.
525 It should return true if the module was modified by the transformation and
530 <!-- ======================================================================= -->
531 <div class="doc_subsection">
532 <a name="CallGraphSCCPass">The <tt>CallGraphSCCPass</tt> class</a>
535 <div class="doc_text">
538 href="http://llvm.org/doxygen/classllvm_1_1CallGraphSCCPass.html">CallGraphSCCPass</a></tt>"
539 is used by passes that need to traverse the program bottom-up on the call graph
540 (callees before callers). Deriving from CallGraphSCCPass provides some
541 mechanics for building and traversing the CallGraph, but also allows the system
542 to optimize execution of CallGraphSCCPass's. If your pass meets the
543 requirements outlined below, and doesn't meet the requirements of a <tt><a
544 href="#FunctionPass">FunctionPass</a></tt> or <tt><a
545 href="#BasicBlockPass">BasicBlockPass</a></tt>, you should derive from
546 <tt>CallGraphSCCPass</tt>.</p>
548 <p><b>TODO</b>: explain briefly what SCC, Tarjan's algo, and B-U mean.</p>
550 <p>To be explicit, <tt>CallGraphSCCPass</tt> subclasses are:</p>
554 <li>... <em>not allowed</em> to modify any <tt>Function</tt>s that are not in
555 the current SCC.</li>
557 <li>... <em>not allowed</em> to inspect any Function's other than those in the
558 current SCC and the direct callees of the SCC.</li>
560 <li>... <em>required</em> to preserve the current CallGraph object, updating it
561 to reflect any changes made to the program.</li>
563 <li>... <em>not allowed</em> to add or remove SCC's from the current Module,
564 though they may change the contents of an SCC.</li>
566 <li>... <em>allowed</em> to add or remove global variables from the current
569 <li>... <em>allowed</em> to maintain state across invocations of
570 <a href="#runOnSCC"><tt>runOnSCC</tt></a> (including global data).</li>
573 <p>Implementing a <tt>CallGraphSCCPass</tt> is slightly tricky in some cases
574 because it has to handle SCCs with more than one node in it. All of the virtual
575 methods described below should return true if they modified the program, or
576 false if they didn't.</p>
580 <!-- _______________________________________________________________________ -->
581 <div class="doc_subsubsection">
582 <a name="doInitialization_scc">The <tt>doInitialization(CallGraph &)</tt>
586 <div class="doc_text">
588 <div class="doc_code"><pre>
589 <b>virtual bool</b> doInitialization(CallGraph &CG);
592 <p>The <tt>doIninitialize</tt> method is allowed to do most of the things that
593 <tt>CallGraphSCCPass</tt>'s are not allowed to do. They can add and remove
594 functions, get pointers to functions, etc. The <tt>doInitialization</tt> method
595 is designed to do simple initialization type of stuff that does not depend on
596 the SCCs being processed. The <tt>doInitialization</tt> method call is not
597 scheduled to overlap with any other pass executions (thus it should be very
602 <!-- _______________________________________________________________________ -->
603 <div class="doc_subsubsection">
604 <a name="runOnSCC">The <tt>runOnSCC</tt> method</a>
607 <div class="doc_text">
609 <div class="doc_code"><pre>
610 <b>virtual bool</b> runOnSCC(CallGraphSCC &SCC) = 0;
613 <p>The <tt>runOnSCC</tt> method performs the interesting work of the pass, and
614 should return true if the module was modified by the transformation, false
619 <!-- _______________________________________________________________________ -->
620 <div class="doc_subsubsection">
621 <a name="doFinalization_scc">The <tt>doFinalization(CallGraph
622 &)</tt> method</a>
625 <div class="doc_text">
627 <div class="doc_code"><pre>
628 <b>virtual bool</b> doFinalization(CallGraph &CG);
631 <p>The <tt>doFinalization</tt> method is an infrequently used method that is
632 called when the pass framework has finished calling <a
633 href="#runOnFunction"><tt>runOnFunction</tt></a> for every function in the
634 program being compiled.</p>
638 <!-- ======================================================================= -->
639 <div class="doc_subsection">
640 <a name="FunctionPass">The <tt>FunctionPass</tt> class</a>
643 <div class="doc_text">
645 <p>In contrast to <tt>ModulePass</tt> subclasses, <tt><a
646 href="http://llvm.org/doxygen/classllvm_1_1Pass.html">FunctionPass</a></tt>
647 subclasses do have a predictable, local behavior that can be expected by the
648 system. All <tt>FunctionPass</tt> execute on each function in the program
649 independent of all of the other functions in the program.
650 <tt>FunctionPass</tt>'s do not require that they are executed in a particular
651 order, and <tt>FunctionPass</tt>'s do not modify external functions.</p>
653 <p>To be explicit, <tt>FunctionPass</tt> subclasses are not allowed to:</p>
656 <li>Modify a Function other than the one currently being processed.</li>
657 <li>Add or remove Function's from the current Module.</li>
658 <li>Add or remove global variables from the current Module.</li>
659 <li>Maintain state across invocations of
660 <a href="#runOnFunction"><tt>runOnFunction</tt></a> (including global data)</li>
663 <p>Implementing a <tt>FunctionPass</tt> is usually straightforward (See the <a
664 href="#basiccode">Hello World</a> pass for example). <tt>FunctionPass</tt>'s
665 may overload three virtual methods to do their work. All of these methods
666 should return true if they modified the program, or false if they didn't.</p>
670 <!-- _______________________________________________________________________ -->
671 <div class="doc_subsubsection">
672 <a name="doInitialization_mod">The <tt>doInitialization(Module &)</tt>
676 <div class="doc_text">
678 <div class="doc_code"><pre>
679 <b>virtual bool</b> doInitialization(Module &M);
682 <p>The <tt>doIninitialize</tt> method is allowed to do most of the things that
683 <tt>FunctionPass</tt>'s are not allowed to do. They can add and remove
684 functions, get pointers to functions, etc. The <tt>doInitialization</tt> method
685 is designed to do simple initialization type of stuff that does not depend on
686 the functions being processed. The <tt>doInitialization</tt> method call is not
687 scheduled to overlap with any other pass executions (thus it should be very
690 <p>A good example of how this method should be used is the <a
691 href="http://llvm.org/doxygen/LowerAllocations_8cpp-source.html">LowerAllocations</a>
692 pass. This pass converts <tt>malloc</tt> and <tt>free</tt> instructions into
693 platform dependent <tt>malloc()</tt> and <tt>free()</tt> function calls. It
694 uses the <tt>doInitialization</tt> method to get a reference to the malloc and
695 free functions that it needs, adding prototypes to the module if necessary.</p>
699 <!-- _______________________________________________________________________ -->
700 <div class="doc_subsubsection">
701 <a name="runOnFunction">The <tt>runOnFunction</tt> method</a>
704 <div class="doc_text">
706 <div class="doc_code"><pre>
707 <b>virtual bool</b> runOnFunction(Function &F) = 0;
710 <p>The <tt>runOnFunction</tt> method must be implemented by your subclass to do
711 the transformation or analysis work of your pass. As usual, a true value should
712 be returned if the function is modified.</p>
716 <!-- _______________________________________________________________________ -->
717 <div class="doc_subsubsection">
718 <a name="doFinalization_mod">The <tt>doFinalization(Module
719 &)</tt> method</a>
722 <div class="doc_text">
724 <div class="doc_code"><pre>
725 <b>virtual bool</b> doFinalization(Module &M);
728 <p>The <tt>doFinalization</tt> method is an infrequently used method that is
729 called when the pass framework has finished calling <a
730 href="#runOnFunction"><tt>runOnFunction</tt></a> for every function in the
731 program being compiled.</p>
735 <!-- ======================================================================= -->
736 <div class="doc_subsection">
737 <a name="LoopPass">The <tt>LoopPass</tt> class </a>
740 <div class="doc_text">
742 <p> All <tt>LoopPass</tt> execute on each loop in the function independent of
743 all of the other loops in the function. <tt>LoopPass</tt> processes loops in
744 loop nest order such that outer most loop is processed last. </p>
746 <p> <tt>LoopPass</tt> subclasses are allowed to update loop nest using
747 <tt>LPPassManager</tt> interface. Implementing a loop pass is usually
748 straightforward. <tt>Looppass</tt>'s may overload three virtual methods to
749 do their work. All these methods should return true if they modified the
750 program, or false if they didn't. </p>
753 <!-- _______________________________________________________________________ -->
754 <div class="doc_subsubsection">
755 <a name="doInitialization_loop">The <tt>doInitialization(Loop *,
756 LPPassManager &)</tt>
760 <div class="doc_text">
762 <div class="doc_code"><pre>
763 <b>virtual bool</b> doInitialization(Loop *, LPPassManager &LPM);
766 <p>The <tt>doInitialization</tt> method is designed to do simple initialization
767 type of stuff that does not depend on the functions being processed. The
768 <tt>doInitialization</tt> method call is not scheduled to overlap with any
769 other pass executions (thus it should be very fast). LPPassManager
770 interface should be used to access Function or Module level analysis
776 <!-- _______________________________________________________________________ -->
777 <div class="doc_subsubsection">
778 <a name="runOnLoop">The <tt>runOnLoop</tt> method</a>
781 <div class="doc_text">
783 <div class="doc_code"><pre>
784 <b>virtual bool</b> runOnLoop(Loop *, LPPassManager &LPM) = 0;
787 <p>The <tt>runOnLoop</tt> method must be implemented by your subclass to do
788 the transformation or analysis work of your pass. As usual, a true value should
789 be returned if the function is modified. <tt>LPPassManager</tt> interface
790 should be used to update loop nest.</p>
794 <!-- _______________________________________________________________________ -->
795 <div class="doc_subsubsection">
796 <a name="doFinalization_loop">The <tt>doFinalization()</tt> method</a>
799 <div class="doc_text">
801 <div class="doc_code"><pre>
802 <b>virtual bool</b> doFinalization();
805 <p>The <tt>doFinalization</tt> method is an infrequently used method that is
806 called when the pass framework has finished calling <a
807 href="#runOnLoop"><tt>runOnLoop</tt></a> for every loop in the
808 program being compiled. </p>
814 <!-- ======================================================================= -->
815 <div class="doc_subsection">
816 <a name="BasicBlockPass">The <tt>BasicBlockPass</tt> class</a>
819 <div class="doc_text">
821 <p><tt>BasicBlockPass</tt>'s are just like <a
822 href="#FunctionPass"><tt>FunctionPass</tt></a>'s, except that they must limit
823 their scope of inspection and modification to a single basic block at a time.
824 As such, they are <b>not</b> allowed to do any of the following:</p>
827 <li>Modify or inspect any basic blocks outside of the current one</li>
828 <li>Maintain state across invocations of
829 <a href="#runOnBasicBlock"><tt>runOnBasicBlock</tt></a></li>
830 <li>Modify the control flow graph (by altering terminator instructions)</li>
831 <li>Any of the things forbidden for
832 <a href="#FunctionPass"><tt>FunctionPass</tt></a>es.</li>
835 <p><tt>BasicBlockPass</tt>es are useful for traditional local and "peephole"
836 optimizations. They may override the same <a
837 href="#doInitialization_mod"><tt>doInitialization(Module &)</tt></a> and <a
838 href="#doFinalization_mod"><tt>doFinalization(Module &)</tt></a> methods that <a
839 href="#FunctionPass"><tt>FunctionPass</tt></a>'s have, but also have the following virtual methods that may also be implemented:</p>
843 <!-- _______________________________________________________________________ -->
844 <div class="doc_subsubsection">
845 <a name="doInitialization_fn">The <tt>doInitialization(Function
846 &)</tt> method</a>
849 <div class="doc_text">
851 <div class="doc_code"><pre>
852 <b>virtual bool</b> doInitialization(Function &F);
855 <p>The <tt>doIninitialize</tt> method is allowed to do most of the things that
856 <tt>BasicBlockPass</tt>'s are not allowed to do, but that
857 <tt>FunctionPass</tt>'s can. The <tt>doInitialization</tt> method is designed
858 to do simple initialization that does not depend on the
859 BasicBlocks being processed. The <tt>doInitialization</tt> method call is not
860 scheduled to overlap with any other pass executions (thus it should be very
865 <!-- _______________________________________________________________________ -->
866 <div class="doc_subsubsection">
867 <a name="runOnBasicBlock">The <tt>runOnBasicBlock</tt> method</a>
870 <div class="doc_text">
872 <div class="doc_code"><pre>
873 <b>virtual bool</b> runOnBasicBlock(BasicBlock &BB) = 0;
876 <p>Override this function to do the work of the <tt>BasicBlockPass</tt>. This
877 function is not allowed to inspect or modify basic blocks other than the
878 parameter, and are not allowed to modify the CFG. A true value must be returned
879 if the basic block is modified.</p>
883 <!-- _______________________________________________________________________ -->
884 <div class="doc_subsubsection">
885 <a name="doFinalization_fn">The <tt>doFinalization(Function &)</tt>
889 <div class="doc_text">
891 <div class="doc_code"><pre>
892 <b>virtual bool</b> doFinalization(Function &F);
895 <p>The <tt>doFinalization</tt> method is an infrequently used method that is
896 called when the pass framework has finished calling <a
897 href="#runOnBasicBlock"><tt>runOnBasicBlock</tt></a> for every BasicBlock in the
898 program being compiled. This can be used to perform per-function
903 <!-- ======================================================================= -->
904 <div class="doc_subsection">
905 <a name="MachineFunctionPass">The <tt>MachineFunctionPass</tt> class</a>
908 <div class="doc_text">
910 <p>A <tt>MachineFunctionPass</tt> is a part of the LLVM code generator that
911 executes on the machine-dependent representation of each LLVM function in the
914 <p>Code generator passes are registered and initialized specially by
915 <tt>TargetMachine::addPassesToEmitFile</tt> and similar routines, so they
916 cannot generally be run from the <tt>opt</tt> or <tt>bugpoint</tt>
919 <p>A <tt>MachineFunctionPass</tt> is also a <tt>FunctionPass</tt>, so all
920 the restrictions that apply to a <tt>FunctionPass</tt> also apply to it.
921 <tt>MachineFunctionPass</tt>es also have additional restrictions. In particular,
922 <tt>MachineFunctionPass</tt>es are not allowed to do any of the following:</p>
925 <li>Modify or create any LLVM IR Instructions, BasicBlocks, Arguments,
926 Functions, GlobalVariables, GlobalAliases, or Modules.</li>
927 <li>Modify a MachineFunction other than the one currently being processed.</li>
928 <li>Maintain state across invocations of <a
929 href="#runOnMachineFunction"><tt>runOnMachineFunction</tt></a> (including global
935 <!-- _______________________________________________________________________ -->
936 <div class="doc_subsubsection">
937 <a name="runOnMachineFunction">The <tt>runOnMachineFunction(MachineFunction
938 &MF)</tt> method</a>
941 <div class="doc_text">
943 <div class="doc_code"><pre>
944 <b>virtual bool</b> runOnMachineFunction(MachineFunction &MF) = 0;
947 <p><tt>runOnMachineFunction</tt> can be considered the main entry point of a
948 <tt>MachineFunctionPass</tt>; that is, you should override this method to do the
949 work of your <tt>MachineFunctionPass</tt>.</p>
951 <p>The <tt>runOnMachineFunction</tt> method is called on every
952 <tt>MachineFunction</tt> in a <tt>Module</tt>, so that the
953 <tt>MachineFunctionPass</tt> may perform optimizations on the machine-dependent
954 representation of the function. If you want to get at the LLVM <tt>Function</tt>
955 for the <tt>MachineFunction</tt> you're working on, use
956 <tt>MachineFunction</tt>'s <tt>getFunction()</tt> accessor method -- but
957 remember, you may not modify the LLVM <tt>Function</tt> or its contents from a
958 <tt>MachineFunctionPass</tt>.</p>
962 <!-- *********************************************************************** -->
963 <div class="doc_section">
964 <a name="registration">Pass registration</a>
966 <!-- *********************************************************************** -->
968 <div class="doc_text">
970 <p>In the <a href="#basiccode">Hello World</a> example pass we illustrated how
971 pass registration works, and discussed some of the reasons that it is used and
972 what it does. Here we discuss how and why passes are registered.</p>
974 <p>As we saw above, passes are registered with the <b><tt>RegisterPass</tt></b>
975 template, which requires you to pass at least two
976 parameters. The first parameter is the name of the pass that is to be used on
977 the command line to specify that the pass should be added to a program (for
978 example, with <tt>opt</tt> or <tt>bugpoint</tt>). The second argument is the
979 name of the pass, which is to be used for the <tt>-help</tt> output of
981 well as for debug output generated by the <tt>--debug-pass</tt> option.</p>
983 <p>If you want your pass to be easily dumpable, you should
984 implement the virtual <tt>print</tt> method:</p>
988 <!-- _______________________________________________________________________ -->
989 <div class="doc_subsubsection">
990 <a name="print">The <tt>print</tt> method</a>
993 <div class="doc_text">
995 <div class="doc_code"><pre>
996 <b>virtual void</b> print(std::ostream &O, <b>const</b> Module *M) <b>const</b>;
999 <p>The <tt>print</tt> method must be implemented by "analyses" in order to print
1000 a human readable version of the analysis results. This is useful for debugging
1001 an analysis itself, as well as for other people to figure out how an analysis
1002 works. Use the <tt>opt -analyze</tt> argument to invoke this method.</p>
1004 <p>The <tt>llvm::OStream</tt> parameter specifies the stream to write the results on,
1005 and the <tt>Module</tt> parameter gives a pointer to the top level module of the
1006 program that has been analyzed. Note however that this pointer may be null in
1007 certain circumstances (such as calling the <tt>Pass::dump()</tt> from a
1008 debugger), so it should only be used to enhance debug output, it should not be
1013 <!-- *********************************************************************** -->
1014 <div class="doc_section">
1015 <a name="interaction">Specifying interactions between passes</a>
1017 <!-- *********************************************************************** -->
1019 <div class="doc_text">
1021 <p>One of the main responsibilities of the <tt>PassManager</tt> is to make sure
1022 that passes interact with each other correctly. Because <tt>PassManager</tt>
1023 tries to <a href="#passmanager">optimize the execution of passes</a> it must
1024 know how the passes interact with each other and what dependencies exist between
1025 the various passes. To track this, each pass can declare the set of passes that
1026 are required to be executed before the current pass, and the passes which are
1027 invalidated by the current pass.</p>
1029 <p>Typically this functionality is used to require that analysis results are
1030 computed before your pass is run. Running arbitrary transformation passes can
1031 invalidate the computed analysis results, which is what the invalidation set
1032 specifies. If a pass does not implement the <tt><a
1033 href="#getAnalysisUsage">getAnalysisUsage</a></tt> method, it defaults to not
1034 having any prerequisite passes, and invalidating <b>all</b> other passes.</p>
1038 <!-- _______________________________________________________________________ -->
1039 <div class="doc_subsubsection">
1040 <a name="getAnalysisUsage">The <tt>getAnalysisUsage</tt> method</a>
1043 <div class="doc_text">
1045 <div class="doc_code"><pre>
1046 <b>virtual void</b> getAnalysisUsage(AnalysisUsage &Info) <b>const</b>;
1049 <p>By implementing the <tt>getAnalysisUsage</tt> method, the required and
1050 invalidated sets may be specified for your transformation. The implementation
1051 should fill in the <tt><a
1052 href="http://llvm.org/doxygen/classllvm_1_1AnalysisUsage.html">AnalysisUsage</a></tt>
1053 object with information about which passes are required and not invalidated. To
1054 do this, a pass may call any of the following methods on the AnalysisUsage
1058 <!-- _______________________________________________________________________ -->
1059 <div class="doc_subsubsection">
1060 <a name="AU::addRequired">The <tt>AnalysisUsage::addRequired<></tt> and <tt>AnalysisUsage::addRequiredTransitive<></tt> methods</a>
1063 <div class="doc_text">
1065 If your pass requires a previous pass to be executed (an analysis for example),
1066 it can use one of these methods to arrange for it to be run before your pass.
1067 LLVM has many different types of analyses and passes that can be required,
1068 spanning the range from <tt>DominatorSet</tt> to <tt>BreakCriticalEdges</tt>.
1069 Requiring <tt>BreakCriticalEdges</tt>, for example, guarantees that there will
1070 be no critical edges in the CFG when your pass has been run.
1074 Some analyses chain to other analyses to do their job. For example, an <a
1075 href="AliasAnalysis.html">AliasAnalysis</a> implementation is required to <a
1076 href="AliasAnalysis.html#chaining">chain</a> to other alias analysis passes. In
1077 cases where analyses chain, the <tt>addRequiredTransitive</tt> method should be
1078 used instead of the <tt>addRequired</tt> method. This informs the PassManager
1079 that the transitively required pass should be alive as long as the requiring
1084 <!-- _______________________________________________________________________ -->
1085 <div class="doc_subsubsection">
1086 <a name="AU::addPreserved">The <tt>AnalysisUsage::addPreserved<></tt> method</a>
1089 <div class="doc_text">
1091 One of the jobs of the PassManager is to optimize how and when analyses are run.
1092 In particular, it attempts to avoid recomputing data unless it needs to. For
1093 this reason, passes are allowed to declare that they preserve (i.e., they don't
1094 invalidate) an existing analysis if it's available. For example, a simple
1095 constant folding pass would not modify the CFG, so it can't possibly affect the
1096 results of dominator analysis. By default, all passes are assumed to invalidate
1101 The <tt>AnalysisUsage</tt> class provides several methods which are useful in
1102 certain circumstances that are related to <tt>addPreserved</tt>. In particular,
1103 the <tt>setPreservesAll</tt> method can be called to indicate that the pass does
1104 not modify the LLVM program at all (which is true for analyses), and the
1105 <tt>setPreservesCFG</tt> method can be used by transformations that change
1106 instructions in the program but do not modify the CFG or terminator instructions
1107 (note that this property is implicitly set for <a
1108 href="#BasicBlockPass">BasicBlockPass</a>'s).
1112 <tt>addPreserved</tt> is particularly useful for transformations like
1113 <tt>BreakCriticalEdges</tt>. This pass knows how to update a small set of loop
1114 and dominator related analyses if they exist, so it can preserve them, despite
1115 the fact that it hacks on the CFG.
1119 <!-- _______________________________________________________________________ -->
1120 <div class="doc_subsubsection">
1121 <a name="AU::examples">Example implementations of <tt>getAnalysisUsage</tt></a>
1124 <div class="doc_text">
1126 <div class="doc_code"><pre>
1127 <i>// This is an example implementation from an analysis, which does not modify
1128 // the program at all, yet has a prerequisite.</i>
1129 <b>void</b> <a href="http://llvm.org/doxygen/classllvm_1_1PostDominanceFrontier.html">PostDominanceFrontier</a>::getAnalysisUsage(AnalysisUsage &AU) <b>const</b> {
1130 AU.setPreservesAll();
1131 AU.addRequired<<a href="http://llvm.org/doxygen/classllvm_1_1PostDominatorTree.html">PostDominatorTree</a>>();
1137 <div class="doc_code"><pre>
1138 <i>// This example modifies the program, but does not modify the CFG</i>
1139 <b>void</b> <a href="http://llvm.org/doxygen/structLICM.html">LICM</a>::getAnalysisUsage(AnalysisUsage &AU) <b>const</b> {
1140 AU.setPreservesCFG();
1141 AU.addRequired<<a href="http://llvm.org/doxygen/classllvm_1_1LoopInfo.html">LoopInfo</a>>();
1147 <!-- _______________________________________________________________________ -->
1148 <div class="doc_subsubsection">
1149 <a name="getAnalysis">The <tt>getAnalysis<></tt> and
1150 <tt>getAnalysisIfAvailable<></tt> methods</a>
1153 <div class="doc_text">
1155 <p>The <tt>Pass::getAnalysis<></tt> method is automatically inherited by
1156 your class, providing you with access to the passes that you declared that you
1157 required with the <a href="#getAnalysisUsage"><tt>getAnalysisUsage</tt></a>
1158 method. It takes a single template argument that specifies which pass class you
1159 want, and returns a reference to that pass. For example:</p>
1161 <div class="doc_code"><pre>
1162 bool LICM::runOnFunction(Function &F) {
1163 LoopInfo &LI = getAnalysis<LoopInfo>();
1168 <p>This method call returns a reference to the pass desired. You may get a
1169 runtime assertion failure if you attempt to get an analysis that you did not
1170 declare as required in your <a
1171 href="#getAnalysisUsage"><tt>getAnalysisUsage</tt></a> implementation. This
1172 method can be called by your <tt>run*</tt> method implementation, or by any
1173 other local method invoked by your <tt>run*</tt> method.
1175 A module level pass can use function level analysis info using this interface.
1178 <div class="doc_code"><pre>
1179 bool ModuleLevelPass::runOnModule(Module &M) {
1181 DominatorTree &DT = getAnalysis<DominatorTree>(Func);
1186 <p>In above example, runOnFunction for DominatorTree is called by pass manager
1187 before returning a reference to the desired pass.</p>
1190 If your pass is capable of updating analyses if they exist (e.g.,
1191 <tt>BreakCriticalEdges</tt>, as described above), you can use the
1192 <tt>getAnalysisIfAvailable</tt> method, which returns a pointer to the analysis
1193 if it is active. For example:</p>
1195 <div class="doc_code"><pre>
1197 if (DominatorSet *DS = getAnalysisIfAvailable<DominatorSet>()) {
1198 <i>// A DominatorSet is active. This code will update it.</i>
1205 <!-- *********************************************************************** -->
1206 <div class="doc_section">
1207 <a name="analysisgroup">Implementing Analysis Groups</a>
1209 <!-- *********************************************************************** -->
1211 <div class="doc_text">
1213 <p>Now that we understand the basics of how passes are defined, how they are
1214 used, and how they are required from other passes, it's time to get a little bit
1215 fancier. All of the pass relationships that we have seen so far are very
1216 simple: one pass depends on one other specific pass to be run before it can run.
1217 For many applications, this is great, for others, more flexibility is
1220 <p>In particular, some analyses are defined such that there is a single simple
1221 interface to the analysis results, but multiple ways of calculating them.
1222 Consider alias analysis for example. The most trivial alias analysis returns
1223 "may alias" for any alias query. The most sophisticated analysis a
1224 flow-sensitive, context-sensitive interprocedural analysis that can take a
1225 significant amount of time to execute (and obviously, there is a lot of room
1226 between these two extremes for other implementations). To cleanly support
1227 situations like this, the LLVM Pass Infrastructure supports the notion of
1228 Analysis Groups.</p>
1232 <!-- _______________________________________________________________________ -->
1233 <div class="doc_subsubsection">
1234 <a name="agconcepts">Analysis Group Concepts</a>
1237 <div class="doc_text">
1239 <p>An Analysis Group is a single simple interface that may be implemented by
1240 multiple different passes. Analysis Groups can be given human readable names
1241 just like passes, but unlike passes, they need not derive from the <tt>Pass</tt>
1242 class. An analysis group may have one or more implementations, one of which is
1243 the "default" implementation.</p>
1245 <p>Analysis groups are used by client passes just like other passes are: the
1246 <tt>AnalysisUsage::addRequired()</tt> and <tt>Pass::getAnalysis()</tt> methods.
1247 In order to resolve this requirement, the <a href="#passmanager">PassManager</a>
1248 scans the available passes to see if any implementations of the analysis group
1249 are available. If none is available, the default implementation is created for
1250 the pass to use. All standard rules for <A href="#interaction">interaction
1251 between passes</a> still apply.</p>
1253 <p>Although <a href="#registration">Pass Registration</a> is optional for normal
1254 passes, all analysis group implementations must be registered, and must use the
1255 <A href="#registerag"><tt>RegisterAnalysisGroup</tt></a> template to join the
1256 implementation pool. Also, a default implementation of the interface
1257 <b>must</b> be registered with <A
1258 href="#registerag"><tt>RegisterAnalysisGroup</tt></a>.</p>
1260 <p>As a concrete example of an Analysis Group in action, consider the <a
1261 href="http://llvm.org/doxygen/classllvm_1_1AliasAnalysis.html">AliasAnalysis</a>
1262 analysis group. The default implementation of the alias analysis interface (the
1264 href="http://llvm.org/doxygen/structBasicAliasAnalysis.html">basicaa</a></tt>
1265 pass) just does a few simple checks that don't require significant analysis to
1266 compute (such as: two different globals can never alias each other, etc).
1267 Passes that use the <tt><a
1268 href="http://llvm.org/doxygen/classllvm_1_1AliasAnalysis.html">AliasAnalysis</a></tt>
1269 interface (for example the <tt><a
1270 href="http://llvm.org/doxygen/structGCSE.html">gcse</a></tt> pass), do
1271 not care which implementation of alias analysis is actually provided, they just
1272 use the designated interface.</p>
1274 <p>From the user's perspective, commands work just like normal. Issuing the
1275 command '<tt>opt -gcse ...</tt>' will cause the <tt>basicaa</tt> class to be
1276 instantiated and added to the pass sequence. Issuing the command '<tt>opt
1277 -somefancyaa -gcse ...</tt>' will cause the <tt>gcse</tt> pass to use the
1278 <tt>somefancyaa</tt> alias analysis (which doesn't actually exist, it's just a
1279 hypothetical example) instead.</p>
1283 <!-- _______________________________________________________________________ -->
1284 <div class="doc_subsubsection">
1285 <a name="registerag">Using <tt>RegisterAnalysisGroup</tt></a>
1288 <div class="doc_text">
1290 <p>The <tt>RegisterAnalysisGroup</tt> template is used to register the analysis
1291 group itself as well as add pass implementations to the analysis group. First,
1292 an analysis should be registered, with a human readable name provided for it.
1293 Unlike registration of passes, there is no command line argument to be specified
1294 for the Analysis Group Interface itself, because it is "abstract":</p>
1296 <div class="doc_code"><pre>
1297 <b>static</b> RegisterAnalysisGroup<<a href="http://llvm.org/doxygen/classllvm_1_1AliasAnalysis.html">AliasAnalysis</a>> A("<i>Alias Analysis</i>");
1300 <p>Once the analysis is registered, passes can declare that they are valid
1301 implementations of the interface by using the following code:</p>
1303 <div class="doc_code"><pre>
1305 //<i> Analysis Group implementations <b>must</b> be registered normally...</i>
1306 RegisterPass<FancyAA>
1307 B("<i>somefancyaa</i>", "<i>A more complex alias analysis implementation</i>");
1309 //<i> Declare that we implement the AliasAnalysis interface</i>
1310 RegisterAnalysisGroup<<a href="http://llvm.org/doxygen/classllvm_1_1AliasAnalysis.html">AliasAnalysis</a>> C(B);
1314 <p>This just shows a class <tt>FancyAA</tt> that is registered normally, then
1315 uses the <tt>RegisterAnalysisGroup</tt> template to "join" the <tt><a
1316 href="http://llvm.org/doxygen/classllvm_1_1AliasAnalysis.html">AliasAnalysis</a></tt>
1317 analysis group. Every implementation of an analysis group should join using
1318 this template. A single pass may join multiple different analysis groups with
1321 <div class="doc_code"><pre>
1323 //<i> Analysis Group implementations <b>must</b> be registered normally...</i>
1324 RegisterPass<<a href="http://llvm.org/doxygen/structBasicAliasAnalysis.html">BasicAliasAnalysis</a>>
1325 D("<i>basicaa</i>", "<i>Basic Alias Analysis (default AA impl)</i>");
1327 //<i> Declare that we implement the AliasAnalysis interface</i>
1328 RegisterAnalysisGroup<<a href="http://llvm.org/doxygen/classllvm_1_1AliasAnalysis.html">AliasAnalysis</a>, <b>true</b>> E(D);
1332 <p>Here we show how the default implementation is specified (using the extra
1333 argument to the <tt>RegisterAnalysisGroup</tt> template). There must be exactly
1334 one default implementation available at all times for an Analysis Group to be
1335 used. Only default implementation can derive from <tt>ImmutablePass</tt>.
1336 Here we declare that the
1337 <tt><a href="http://llvm.org/doxygen/structBasicAliasAnalysis.html">BasicAliasAnalysis</a></tt>
1338 pass is the default implementation for the interface.</p>
1342 <!-- *********************************************************************** -->
1343 <div class="doc_section">
1344 <a name="passStatistics">Pass Statistics</a>
1346 <!-- *********************************************************************** -->
1348 <div class="doc_text">
1350 href="http://llvm.org/doxygen/Statistic_8h-source.html"><tt>Statistic</tt></a>
1351 class is designed to be an easy way to expose various success
1352 metrics from passes. These statistics are printed at the end of a
1353 run, when the -stats command line option is enabled on the command
1354 line. See the <a href="http://llvm.org/docs/ProgrammersManual.html#Statistic">Statistics section</a> in the Programmer's Manual for details.
1359 <!-- *********************************************************************** -->
1360 <div class="doc_section">
1361 <a name="passmanager">What PassManager does</a>
1363 <!-- *********************************************************************** -->
1365 <div class="doc_text">
1368 href="http://llvm.org/doxygen/PassManager_8h-source.html"><tt>PassManager</tt></a>
1370 href="http://llvm.org/doxygen/classllvm_1_1PassManager.html">class</a>
1371 takes a list of passes, ensures their <a href="#interaction">prerequisites</a>
1372 are set up correctly, and then schedules passes to run efficiently. All of the
1373 LLVM tools that run passes use the <tt>PassManager</tt> for execution of these
1376 <p>The <tt>PassManager</tt> does two main things to try to reduce the execution
1377 time of a series of passes:</p>
1380 <li><b>Share analysis results</b> - The PassManager attempts to avoid
1381 recomputing analysis results as much as possible. This means keeping track of
1382 which analyses are available already, which analyses get invalidated, and which
1383 analyses are needed to be run for a pass. An important part of work is that the
1384 <tt>PassManager</tt> tracks the exact lifetime of all analysis results, allowing
1385 it to <a href="#releaseMemory">free memory</a> allocated to holding analysis
1386 results as soon as they are no longer needed.</li>
1388 <li><b>Pipeline the execution of passes on the program</b> - The
1389 <tt>PassManager</tt> attempts to get better cache and memory usage behavior out
1390 of a series of passes by pipelining the passes together. This means that, given
1391 a series of consequtive <a href="#FunctionPass"><tt>FunctionPass</tt></a>'s, it
1392 will execute all of the <a href="#FunctionPass"><tt>FunctionPass</tt></a>'s on
1393 the first function, then all of the <a
1394 href="#FunctionPass"><tt>FunctionPass</tt></a>es on the second function,
1395 etc... until the entire program has been run through the passes.
1397 <p>This improves the cache behavior of the compiler, because it is only touching
1398 the LLVM program representation for a single function at a time, instead of
1399 traversing the entire program. It reduces the memory consumption of compiler,
1400 because, for example, only one <a
1401 href="http://llvm.org/doxygen/classllvm_1_1DominatorSet.html"><tt>DominatorSet</tt></a>
1402 needs to be calculated at a time. This also makes it possible to implement
1404 href="#SMP">interesting enhancements</a> in the future.</p></li>
1408 <p>The effectiveness of the <tt>PassManager</tt> is influenced directly by how
1409 much information it has about the behaviors of the passes it is scheduling. For
1410 example, the "preserved" set is intentionally conservative in the face of an
1411 unimplemented <a href="#getAnalysisUsage"><tt>getAnalysisUsage</tt></a> method.
1412 Not implementing when it should be implemented will have the effect of not
1413 allowing any analysis results to live across the execution of your pass.</p>
1415 <p>The <tt>PassManager</tt> class exposes a <tt>--debug-pass</tt> command line
1416 options that is useful for debugging pass execution, seeing how things work, and
1417 diagnosing when you should be preserving more analyses than you currently are
1418 (To get information about all of the variants of the <tt>--debug-pass</tt>
1419 option, just type '<tt>opt -help-hidden</tt>').</p>
1421 <p>By using the <tt>--debug-pass=Structure</tt> option, for example, we can see
1422 how our <a href="#basiccode">Hello World</a> pass interacts with other passes.
1423 Lets try it out with the <tt>gcse</tt> and <tt>licm</tt> passes:</p>
1425 <div class="doc_code"><pre>
1426 $ opt -load ../../../Debug/lib/Hello.so -gcse -licm --debug-pass=Structure < hello.bc > /dev/null
1428 Function Pass Manager
1429 Dominator Set Construction
1430 Immediate Dominators Construction
1431 Global Common Subexpression Elimination
1432 -- Immediate Dominators Construction
1433 -- Global Common Subexpression Elimination
1434 Natural Loop Construction
1435 Loop Invariant Code Motion
1436 -- Natural Loop Construction
1437 -- Loop Invariant Code Motion
1439 -- Dominator Set Construction
1445 <p>This output shows us when passes are constructed and when the analysis
1446 results are known to be dead (prefixed with '<tt>--</tt>'). Here we see that
1447 GCSE uses dominator and immediate dominator information to do its job. The LICM
1448 pass uses natural loop information, which uses dominator sets, but not immediate
1449 dominators. Because immediate dominators are no longer useful after the GCSE
1450 pass, it is immediately destroyed. The dominator sets are then reused to
1451 compute natural loop information, which is then used by the LICM pass.</p>
1453 <p>After the LICM pass, the module verifier runs (which is automatically added
1454 by the '<tt>opt</tt>' tool), which uses the dominator set to check that the
1455 resultant LLVM code is well formed. After it finishes, the dominator set
1456 information is destroyed, after being computed once, and shared by three
1459 <p>Lets see how this changes when we run the <a href="#basiccode">Hello
1460 World</a> pass in between the two passes:</p>
1462 <div class="doc_code"><pre>
1463 $ opt -load ../../../Debug/lib/Hello.so -gcse -hello -licm --debug-pass=Structure < hello.bc > /dev/null
1465 Function Pass Manager
1466 Dominator Set Construction
1467 Immediate Dominators Construction
1468 Global Common Subexpression Elimination
1469 <b>-- Dominator Set Construction</b>
1470 -- Immediate Dominators Construction
1471 -- Global Common Subexpression Elimination
1472 <b> Hello World Pass
1474 Dominator Set Construction</b>
1475 Natural Loop Construction
1476 Loop Invariant Code Motion
1477 -- Natural Loop Construction
1478 -- Loop Invariant Code Motion
1480 -- Dominator Set Construction
1489 <p>Here we see that the <a href="#basiccode">Hello World</a> pass has killed the
1490 Dominator Set pass, even though it doesn't modify the code at all! To fix this,
1491 we need to add the following <a
1492 href="#getAnalysisUsage"><tt>getAnalysisUsage</tt></a> method to our pass:</p>
1494 <div class="doc_code"><pre>
1495 <i>// We don't modify the program, so we preserve all analyses</i>
1496 <b>virtual void</b> getAnalysisUsage(AnalysisUsage &AU) <b>const</b> {
1497 AU.setPreservesAll();
1501 <p>Now when we run our pass, we get this output:</p>
1503 <div class="doc_code"><pre>
1504 $ opt -load ../../../Debug/lib/Hello.so -gcse -hello -licm --debug-pass=Structure < hello.bc > /dev/null
1505 Pass Arguments: -gcse -hello -licm
1507 Function Pass Manager
1508 Dominator Set Construction
1509 Immediate Dominators Construction
1510 Global Common Subexpression Elimination
1511 -- Immediate Dominators Construction
1512 -- Global Common Subexpression Elimination
1515 Natural Loop Construction
1516 Loop Invariant Code Motion
1517 -- Loop Invariant Code Motion
1518 -- Natural Loop Construction
1520 -- Dominator Set Construction
1529 <p>Which shows that we don't accidentally invalidate dominator information
1530 anymore, and therefore do not have to compute it twice.</p>
1534 <!-- _______________________________________________________________________ -->
1535 <div class="doc_subsubsection">
1536 <a name="releaseMemory">The <tt>releaseMemory</tt> method</a>
1539 <div class="doc_text">
1541 <div class="doc_code"><pre>
1542 <b>virtual void</b> releaseMemory();
1545 <p>The <tt>PassManager</tt> automatically determines when to compute analysis
1546 results, and how long to keep them around for. Because the lifetime of the pass
1547 object itself is effectively the entire duration of the compilation process, we
1548 need some way to free analysis results when they are no longer useful. The
1549 <tt>releaseMemory</tt> virtual method is the way to do this.</p>
1551 <p>If you are writing an analysis or any other pass that retains a significant
1552 amount of state (for use by another pass which "requires" your pass and uses the
1553 <a href="#getAnalysis">getAnalysis</a> method) you should implement
1554 <tt>releaseMemory</tt> to, well, release the memory allocated to maintain this
1555 internal state. This method is called after the <tt>run*</tt> method for the
1556 class, before the next call of <tt>run*</tt> in your pass.</p>
1560 <!-- *********************************************************************** -->
1561 <div class="doc_section">
1562 <a name="registering">Registering dynamically loaded passes</a>
1564 <!-- *********************************************************************** -->
1566 <div class="doc_text">
1568 <p><i>Size matters</i> when constructing production quality tools using llvm,
1569 both for the purposes of distribution, and for regulating the resident code size
1570 when running on the target system. Therefore, it becomes desirable to
1571 selectively use some passes, while omitting others and maintain the flexibility
1572 to change configurations later on. You want to be able to do all this, and,
1573 provide feedback to the user. This is where pass registration comes into
1576 <p>The fundamental mechanisms for pass registration are the
1577 <tt>MachinePassRegistry</tt> class and subclasses of
1578 <tt>MachinePassRegistryNode</tt>.</p>
1580 <p>An instance of <tt>MachinePassRegistry</tt> is used to maintain a list of
1581 <tt>MachinePassRegistryNode</tt> objects. This instance maintains the list and
1582 communicates additions and deletions to the command line interface.</p>
1584 <p>An instance of <tt>MachinePassRegistryNode</tt> subclass is used to maintain
1585 information provided about a particular pass. This information includes the
1586 command line name, the command help string and the address of the function used
1587 to create an instance of the pass. A global static constructor of one of these
1588 instances <i>registers</i> with a corresponding <tt>MachinePassRegistry</tt>,
1589 the static destructor <i>unregisters</i>. Thus a pass that is statically linked
1590 in the tool will be registered at start up. A dynamically loaded pass will
1591 register on load and unregister at unload.</p>
1595 <!-- _______________________________________________________________________ -->
1596 <div class="doc_subsection">
1597 <a name="registering_existing">Using existing registries</a>
1600 <div class="doc_text">
1602 <p>There are predefined registries to track instruction scheduling
1603 (<tt>RegisterScheduler</tt>) and register allocation (<tt>RegisterRegAlloc</tt>)
1604 machine passes. Here we will describe how to <i>register</i> a register
1605 allocator machine pass.</p>
1607 <p>Implement your register allocator machine pass. In your register allocator
1608 .cpp file add the following include;</p>
1610 <div class="doc_code"><pre>
1611 #include "llvm/CodeGen/RegAllocRegistry.h"
1614 <p>Also in your register allocator .cpp file, define a creator function in the
1617 <div class="doc_code"><pre>
1618 FunctionPass *createMyRegisterAllocator() {
1619 return new MyRegisterAllocator();
1623 <p>Note that the signature of this function should match the type of
1624 <tt>RegisterRegAlloc::FunctionPassCtor</tt>. In the same file add the
1625 "installing" declaration, in the form;</p>
1627 <div class="doc_code"><pre>
1628 static RegisterRegAlloc myRegAlloc("myregalloc",
1629 " my register allocator help string",
1630 createMyRegisterAllocator);
1633 <p>Note the two spaces prior to the help string produces a tidy result on the
1636 <div class="doc_code"><pre>
1639 -regalloc - Register allocator to use (default=linearscan)
1640 =linearscan - linear scan register allocator
1641 =local - local register allocator
1642 =simple - simple register allocator
1643 =myregalloc - my register allocator help string
1647 <p>And that's it. The user is now free to use <tt>-regalloc=myregalloc</tt> as
1648 an option. Registering instruction schedulers is similar except use the
1649 <tt>RegisterScheduler</tt> class. Note that the
1650 <tt>RegisterScheduler::FunctionPassCtor</tt> is significantly different from
1651 <tt>RegisterRegAlloc::FunctionPassCtor</tt>.</p>
1653 <p>To force the load/linking of your register allocator into the llc/lli tools,
1654 add your creator function's global declaration to "Passes.h" and add a "pseudo"
1655 call line to <tt>llvm/Codegen/LinkAllCodegenComponents.h</tt>.</p>
1660 <!-- _______________________________________________________________________ -->
1661 <div class="doc_subsection">
1662 <a name="registering_new">Creating new registries</a>
1665 <div class="doc_text">
1667 <p>The easiest way to get started is to clone one of the existing registries; we
1668 recommend <tt>llvm/CodeGen/RegAllocRegistry.h</tt>. The key things to modify
1669 are the class name and the <tt>FunctionPassCtor</tt> type.</p>
1671 <p>Then you need to declare the registry. Example: if your pass registry is
1672 <tt>RegisterMyPasses</tt> then define;</p>
1674 <div class="doc_code"><pre>
1675 MachinePassRegistry RegisterMyPasses::Registry;
1678 <p>And finally, declare the command line option for your passes. Example:</p>
1680 <div class="doc_code"><pre>
1681 cl::opt<RegisterMyPasses::FunctionPassCtor, false,
1682 RegisterPassParser<RegisterMyPasses> >
1684 cl::init(&createDefaultMyPass),
1685 cl::desc("my pass option help"));
1688 <p>Here the command option is "mypass", with createDefaultMyPass as the default
1693 <!-- *********************************************************************** -->
1694 <div class="doc_section">
1695 <a name="debughints">Using GDB with dynamically loaded passes</a>
1697 <!-- *********************************************************************** -->
1699 <div class="doc_text">
1701 <p>Unfortunately, using GDB with dynamically loaded passes is not as easy as it
1702 should be. First of all, you can't set a breakpoint in a shared object that has
1703 not been loaded yet, and second of all there are problems with inlined functions
1704 in shared objects. Here are some suggestions to debugging your pass with
1707 <p>For sake of discussion, I'm going to assume that you are debugging a
1708 transformation invoked by <tt>opt</tt>, although nothing described here depends
1713 <!-- _______________________________________________________________________ -->
1714 <div class="doc_subsubsection">
1715 <a name="breakpoint">Setting a breakpoint in your pass</a>
1718 <div class="doc_text">
1720 <p>First thing you do is start <tt>gdb</tt> on the <tt>opt</tt> process:</p>
1722 <div class="doc_code"><pre>
1725 Copyright 2000 Free Software Foundation, Inc.
1726 GDB is free software, covered by the GNU General Public License, and you are
1727 welcome to change it and/or distribute copies of it under certain conditions.
1728 Type "show copying" to see the conditions.
1729 There is absolutely no warranty for GDB. Type "show warranty" for details.
1730 This GDB was configured as "sparc-sun-solaris2.6"...
1734 <p>Note that <tt>opt</tt> has a lot of debugging information in it, so it takes
1735 time to load. Be patient. Since we cannot set a breakpoint in our pass yet
1736 (the shared object isn't loaded until runtime), we must execute the process, and
1737 have it stop before it invokes our pass, but after it has loaded the shared
1738 object. The most foolproof way of doing this is to set a breakpoint in
1739 <tt>PassManager::run</tt> and then run the process with the arguments you
1742 <div class="doc_code"><pre>
1743 (gdb) <b>break llvm::PassManager::run</b>
1744 Breakpoint 1 at 0x2413bc: file Pass.cpp, line 70.
1745 (gdb) <b>run test.bc -load $(LLVMTOP)/llvm/Debug/lib/[libname].so -[passoption]</b>
1746 Starting program: opt test.bc -load $(LLVMTOP)/llvm/Debug/lib/[libname].so -[passoption]
1747 Breakpoint 1, PassManager::run (this=0xffbef174, M=@0x70b298) at Pass.cpp:70
1748 70 bool PassManager::run(Module &M) { return PM->run(M); }
1752 <p>Once the <tt>opt</tt> stops in the <tt>PassManager::run</tt> method you are
1753 now free to set breakpoints in your pass so that you can trace through execution
1754 or do other standard debugging stuff.</p>
1758 <!-- _______________________________________________________________________ -->
1759 <div class="doc_subsubsection">
1760 <a name="debugmisc">Miscellaneous Problems</a>
1763 <div class="doc_text">
1765 <p>Once you have the basics down, there are a couple of problems that GDB has,
1766 some with solutions, some without.</p>
1769 <li>Inline functions have bogus stack information. In general, GDB does a
1770 pretty good job getting stack traces and stepping through inline functions.
1771 When a pass is dynamically loaded however, it somehow completely loses this
1772 capability. The only solution I know of is to de-inline a function (move it
1773 from the body of a class to a .cpp file).</li>
1775 <li>Restarting the program breaks breakpoints. After following the information
1776 above, you have succeeded in getting some breakpoints planted in your pass. Nex
1777 thing you know, you restart the program (i.e., you type '<tt>run</tt>' again),
1778 and you start getting errors about breakpoints being unsettable. The only way I
1779 have found to "fix" this problem is to <tt>delete</tt> the breakpoints that are
1780 already set in your pass, run the program, and re-set the breakpoints once
1781 execution stops in <tt>PassManager::run</tt>.</li>
1785 <p>Hopefully these tips will help with common case debugging situations. If
1786 you'd like to contribute some tips of your own, just contact <a
1787 href="mailto:sabre@nondot.org">Chris</a>.</p>
1791 <!-- *********************************************************************** -->
1792 <div class="doc_section">
1793 <a name="future">Future extensions planned</a>
1795 <!-- *********************************************************************** -->
1797 <div class="doc_text">
1799 <p>Although the LLVM Pass Infrastructure is very capable as it stands, and does
1800 some nifty stuff, there are things we'd like to add in the future. Here is
1801 where we are going:</p>
1805 <!-- _______________________________________________________________________ -->
1806 <div class="doc_subsubsection">
1807 <a name="SMP">Multithreaded LLVM</a>
1810 <div class="doc_text">
1812 <p>Multiple CPU machines are becoming more common and compilation can never be
1813 fast enough: obviously we should allow for a multithreaded compiler. Because of
1814 the semantics defined for passes above (specifically they cannot maintain state
1815 across invocations of their <tt>run*</tt> methods), a nice clean way to
1816 implement a multithreaded compiler would be for the <tt>PassManager</tt> class
1817 to create multiple instances of each pass object, and allow the separate
1818 instances to be hacking on different parts of the program at the same time.</p>
1820 <p>This implementation would prevent each of the passes from having to implement
1821 multithreaded constructs, requiring only the LLVM core to have locking in a few
1822 places (for global resources). Although this is a simple extension, we simply
1823 haven't had time (or multiprocessor machines, thus a reason) to implement this.
1824 Despite that, we have kept the LLVM passes SMP ready, and you should too.</p>
1828 <!-- *********************************************************************** -->
1831 <a href="http://jigsaw.w3.org/css-validator/check/referer"><img
1832 src="http://jigsaw.w3.org/css-validator/images/vcss-blue" alt="Valid CSS"></a>
1833 <a href="http://validator.w3.org/check/referer"><img
1834 src="http://www.w3.org/Icons/valid-html401-blue" alt="Valid HTML 4.01"></a>
1836 <a href="mailto:sabre@nondot.org">Chris Lattner</a><br>
1837 <a href="http://llvm.org">The LLVM Compiler Infrastructure</a><br>
1838 Last modified: $Date$