Blame - docs/WritingAnLLVMPass.html - fp2-dev/platform/external/llvm

blob: 07e736da691e55ae8a2aa1325e72dc0c9e2205f1 [file] [log] [blame]

Dan Gohman	f17a25c	2007-07-18 16:29:46 +0000	[diff] [blame]	1	<!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.01//EN"
				2	"http://www.w3.org/TR/html4/strict.dtd">
				3	<html>
				4	<head>
				5	<meta http-equiv="Content-Type" content="text/html; charset=utf-8">
				6	<title>Writing an LLVM Pass</title>
				7	<link rel="stylesheet" href="llvm.css" type="text/css">
				8	</head>
				9	<body>
				10
				11	<div class="doc_title">
				12	Writing an LLVM Pass
				13	</div>
				14
				15	<ol>
				16	<li><a href="#introduction">Introduction - What is a pass?</a></li>
				17	<li><a href="#quickstart">Quick Start - Writing hello world</a>
				18	<ul>
				19	<li><a href="#makefile">Setting up the build environment</a></li>
				20	<li><a href="#basiccode">Basic code required</a></li>
				21	<li><a href="#running">Running a pass with <tt>opt</tt></a></li>
				22	</ul></li>
				23	<li><a href="#passtype">Pass classes and requirements</a>
				24	<ul>
				25	<li><a href="#ImmutablePass">The <tt>ImmutablePass</tt> class</a></li>
				26	<li><a href="#ModulePass">The <tt>ModulePass</tt> class</a>
				27	<ul>
				28	<li><a href="#runOnModule">The <tt>runOnModule</tt> method</a></li>
				29	</ul></li>
				30	<li><a href="#CallGraphSCCPass">The <tt>CallGraphSCCPass</tt> class</a>
				31	<ul>
				32	<li><a href="#doInitialization_scc">The <tt>doInitialization(CallGraph
				33	&)</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>
				37	</ul></li>
				38	<li><a href="#FunctionPass">The <tt>FunctionPass</tt> class</a>
				39	<ul>
				40	<li><a href="#doInitialization_mod">The <tt>doInitialization(Module
				41	&)</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>
				45	</ul></li>
				46	<li><a href="#LoopPass">The <tt>LoopPass</tt> class</a>
				47	<ul>
				48	<li><a href="#doInitialization_loop">The <tt>doInitialization(Loop *,
				49	LPPassManager &)</tt> method</a></li>
				50	<li><a href="#runOnLoop">The <tt>runOnLoop</tt> method</a></li>
				51	<li><a href="#doFinalization_loop">The <tt>doFinalization()
				52	</tt> method</a></li>
				53	</ul></li>
				54	<li><a href="#BasicBlockPass">The <tt>BasicBlockPass</tt> class</a>
				55	<ul>
				56	<li><a href="#doInitialization_fn">The <tt>doInitialization(Function
				57	&)</tt> method</a></li>
				58	<li><a href="#runOnBasicBlock">The <tt>runOnBasicBlock</tt>
				59	method</a></li>
				60	<li><a href="#doFinalization_fn">The <tt>doFinalization(Function
				61	&)</tt> method</a></li>
				62	</ul></li>
				63	<li><a href="#MachineFunctionPass">The <tt>MachineFunctionPass</tt>
				64	class</a>
				65	<ul>
				66	<li><a href="#runOnMachineFunction">The
				67	<tt>runOnMachineFunction(MachineFunction &)</tt> method</a></li>
				68	</ul></li>
				69	</ul>
				70	<li><a href="#registration">Pass Registration</a>
				71	<ul>
				72	<li><a href="#print">The <tt>print</tt> method</a></li>
				73	</ul></li>
				74	<li><a href="#interaction">Specifying interactions between passes</a>
				75	<ul>
				76	<li><a href="#getAnalysisUsage">The <tt>getAnalysisUsage</tt>
				77	method</a></li>
				78	<li><a href="#AU::addRequired">The <tt>AnalysisUsage::addRequired<></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 <tt>getAnalysisToUpdate<></tt> methods</a></li>
				82	</ul></li>
				83	<li><a href="#analysisgroup">Implementing Analysis Groups</a>
				84	<ul>
				85	<li><a href="#agconcepts">Analysis Group Concepts</a></li>
				86	<li><a href="#registerag">Using <tt>RegisterAnalysisGroup</tt></a></li>
				87	</ul></li>
				88	<li><a href="#passStatistics">Pass Statistics</a>
				89	<li><a href="#passmanager">What PassManager does</a>
				90	<ul>
				91	<li><a href="#releaseMemory">The <tt>releaseMemory</tt> method</a></li>
				92	</ul></li>
				93	<li><a href="#registering">Registering dynamically loaded passes</a>
				94	<ul>
				95	<li><a href="#registering_existing">Using existing registries</a></li>
				96	<li><a href="#registering_new">Creating new registries</a></li>
				97	</ul></li>
				98	<li><a href="#debughints">Using GDB with dynamically loaded passes</a>
				99	<ul>
				100	<li><a href="#breakpoint">Setting a breakpoint in your pass</a></li>
				101	<li><a href="#debugmisc">Miscellaneous Problems</a></li>
				102	</ul></li>
				103	<li><a href="#future">Future extensions planned</a>
				104	<ul>
				105	<li><a href="#SMP">Multithreaded LLVM</a></li>
				106	</ul></li>
				107	</ol>
				108
				109	<div class="doc_author">
				110	<p>Written by <a href="mailto:sabre@nondot.org">Chris Lattner</a> and
				111	<a href="mailto:jlaskey@mac.com">Jim Laskey</a></p>
				112	</div>
				113
				114	<!-- *********************************************************************** -->
				115	<div class="doc_section">
				116	<a name="introduction">Introduction - What is a pass?</a>
				117	</div>
				118	<!-- *********************************************************************** -->
				119
				120	<div class="doc_text">
				121
				122	<p>The LLVM Pass Framework is an important part of the LLVM system, because LLVM
				123	passes are where most of the interesting parts of the compiler exist. Passes
				124	perform the transformations and optimizations that make up the compiler, they
				125	build the analysis results that are used by these transformations, and they are,
				126	above all, a structuring technique for compiler code.</p>
				127
				128	<p>All LLVM passes are subclasses of the <tt><a
				129	href="http://llvm.org/doxygen/classllvm_1_1Pass.html">Pass</a></tt>
				130	class, which implement functionality by overriding virtual methods inherited
				131	from <tt>Pass</tt>. Depending on how your pass works, you should inherit from
				132	the <tt><a href="#ModulePass">ModulePass</a></tt>, <tt><a
				133	href="#CallGraphSCCPass">CallGraphSCCPass</a></tt>, <tt><a
				134	href="#FunctionPass">FunctionPass</a></tt>, or <tt><a
				135	href="#LoopPass">LoopPass</a></tt>, or <tt><a
				136	href="#BasicBlockPass">BasicBlockPass</a></tt> classes, which gives the system
				137	more information about what your pass does, and how it can be combined with
				138	other passes. One of the main features of the LLVM Pass Framework is that it
				139	schedules passes to run in an efficient way based on the constraints that your
				140	pass meets (which are indicated by which class they derive from).</p>
				141
				142	<p>We start by showing you how to construct a pass, everything from setting up
				143	the code, to compiling, loading, and executing it. After the basics are down,
				144	more advanced features are discussed.</p>
				145
				146	</div>
				147
				148	<!-- *********************************************************************** -->
				149	<div class="doc_section">
				150	<a name="quickstart">Quick Start - Writing hello world</a>
				151	</div>
				152	<!-- *********************************************************************** -->
				153
				154	<div class="doc_text">
				155
				156	<p>Here we describe how to write the "hello world" of passes. The "Hello" pass
				157	is designed to simply print out the name of non-external functions that exist in
				158	the program being compiled. It does not modify the program at all, it just
				159	inspects it. The source code and files for this pass are available in the LLVM
				160	source tree in the <tt>lib/Transforms/Hello</tt> directory.</p>
				161
				162	</div>
				163
				164	<!-- ======================================================================= -->
				165	<div class="doc_subsection">
				166	<a name="makefile">Setting up the build environment</a>
				167	</div>
				168
				169	<div class="doc_text">
				170
				171	<p>First, you need to create a new directory somewhere in the LLVM source
				172	base. For this example, we'll assume that you made
				173	<tt>lib/Transforms/Hello</tt>. Next, you must set up a build script
				174	(Makefile) that will compile the source code for the new pass. To do this,
				175	copy the following into <tt>Makefile</tt>:</p>
				176	<hr/>
				177
				178	<div class="doc_code"><pre>
				179	# Makefile for hello pass
				180
				181	# Path to top level of LLVM heirarchy
				182	LEVEL = ../../..
				183
				184	# Name of the library to build
				185	LIBRARYNAME = Hello
				186
				187	# Make the shared library become a loadable module so the tools can
				188	# dlopen/dlsym on the resulting library.
				189	LOADABLE_MODULE = 1
				190
				191	# Tell the build system which LLVM libraries your pass needs. You'll probably
				192	# need at least LLVMSystem.a, LLVMSupport.a, LLVMCore.a but possibly several
				193	# others too.
				194	LLVMLIBS = LLVMCore.a LLVMSupport.a LLVMSystem.a
				195
				196	# Include the makefile implementation stuff
				197	include $(LEVEL)/Makefile.common
				198	</pre></div>
				199
				200	<p>This makefile specifies that all of the <tt>.cpp</tt> files in the current
				201	directory are to be compiled and linked together into a
				202	<tt>Debug/lib/Hello.so</tt> shared object that can be dynamically loaded by
				203	the <tt>opt</tt> or <tt>bugpoint</tt> tools via their <tt>-load</tt> options.
				204	If your operating system uses a suffix other than .so (such as windows or
				205	Mac OS/X), the appropriate extension will be used.</p>
				206
				207	<p>Now that we have the build scripts set up, we just need to write the code for
				208	the pass itself.</p>
				209
				210	</div>
				211
				212	<!-- ======================================================================= -->
				213	<div class="doc_subsection">
				214	<a name="basiccode">Basic code required</a>
				215	</div>
				216
				217	<div class="doc_text">
				218
				219	<p>Now that we have a way to compile our new pass, we just have to write it.
				220	Start out with:</p>
				221
				222	<div class="doc_code"><pre>
				223	<b>#include</b> "<a href="http://llvm.org/doxygen/Pass_8h-source.html">llvm/Pass.h</a>"
				224	<b>#include</b> "<a href="http://llvm.org/doxygen/Function_8h-source.html">llvm/Function.h</a>"
				225	</pre></div>
				226
				227	<p>Which are needed because we are writing a <tt><a
				228	href="http://llvm.org/doxygen/classllvm_1_1Pass.html">Pass</a></tt>, and
				229	we are operating on <tt><a
				230	href="http://llvm.org/doxygen/classllvm_1_1Function.html">Function</a></tt>'s.</p>
				231
				232	<p>Next we have:</p>
				233	<div class="doc_code"><pre>
				234	<b>using namespace llvm;</b>
				235	</pre></div>
				236	<p>... which is required because the functions from the include files
				237	live in the llvm namespace.
				238	</p>
				239
				240	<p>Next we have:</p>
				241
				242	<div class="doc_code"><pre>
				243	<b>namespace</b> {
				244	</pre></div>
				245
				246	<p>... which starts out an anonymous namespace. Anonymous namespaces are to C++
				247	what the "<tt>static</tt>" keyword is to C (at global scope). It makes the
				248	things declared inside of the anonymous namespace only visible to the current
				249	file. If you're not familiar with them, consult a decent C++ book for more
				250	information.</p>
				251
				252	<p>Next, we declare our pass itself:</p>
				253
				254	<div class="doc_code"><pre>
				255	<b>struct</b> Hello : <b>public</b> <a href="#FunctionPass">FunctionPass</a> {
				256	</pre></div><p>
				257
				258	<p>This declares a "<tt>Hello</tt>" class that is a subclass of <tt><a
				259	href="http://llvm.org/doxygen/classllvm_1_1FunctionPass.html">FunctionPass</a></tt>.
				260	The different builtin pass subclasses are described in detail <a
				261	href="#passtype">later</a>, but for now, know that <a
				262	href="#FunctionPass"><tt>FunctionPass</tt></a>'s operate a function at a
				263	time.</p>
				264
				265	<div class="doc_code"><pre>
				266	static char ID;
				267	Hello() : FunctionPass((intptr_t)&ID) {}
				268	</pre></div><p>
				269
				270	<p> This declares pass identifier used by LLVM to identify pass. This allows LLVM to
				271	avoid using expensive C++ runtime information.</p>
				272
				273	<div class="doc_code"><pre>
				274	<b>virtual bool</b> <a href="#runOnFunction">runOnFunction</a>(Function &F) {
				275	llvm::cerr << "<i>Hello: </i>" << F.getName() << "\n";
				276	<b>return false</b>;
				277	}
				278	}; <i>// end of struct Hello</i>
				279	</pre></div>
				280
				281	<p>We declare a "<a href="#runOnFunction"><tt>runOnFunction</tt></a>" method,
				282	which overloads an abstract virtual method inherited from <a
				283	href="#FunctionPass"><tt>FunctionPass</tt></a>. This is where we are supposed
				284	to do our thing, so we just print out our message with the name of each
				285	function.</p>
				286
				287	<div class="doc_code"><pre>
				288	char Hello::ID = 0;
				289	</pre></div>
				290
				291	<p> We initialize pass ID here. LLVM uses ID's address to identify pass so
				292	initialization value is not important.</p>
				293
				294	<div class="doc_code"><pre>
Devang Patel	3aab76e	2008-03-19 21:56:59 +0000	[diff] [blame]	295	RegisterPass<Hello> X("<i>hello</i>", "<i>Hello World Pass</i>",
				296	false /* Only looks at CFG */,
				297	false /* Analysis Pass */);
Dan Gohman	f17a25c	2007-07-18 16:29:46 +0000	[diff] [blame]	298	} <i>// end of anonymous namespace</i>
				299	</pre></div>
				300
				301	<p>Lastly, we <a href="#registration">register our class</a> <tt>Hello</tt>,
				302	giving it a command line
Devang Patel	3aab76e	2008-03-19 21:56:59 +0000	[diff] [blame]	303	argument "<tt>hello</tt>", and a name "<tt>Hello World Pass</tt>".
				304	Last two RegisterPass arguments are optional. Their default value is false.
				305	If a pass walks CFG without modifying it then third argument is set to true.
				306	If a pass is an analysis pass, for example dominator tree pass, then true
				307	is supplied as fourth argument. </p>
Dan Gohman	f17a25c	2007-07-18 16:29:46 +0000	[diff] [blame]	308
				309	<p>As a whole, the <tt>.cpp</tt> file looks like:</p>
				310
				311	<div class="doc_code"><pre>
				312	<b>#include</b> "<a href="http://llvm.org/doxygen/Pass_8h-source.html">llvm/Pass.h</a>"
				313	<b>#include</b> "<a href="http://llvm.org/doxygen/Function_8h-source.html">llvm/Function.h</a>"
				314
				315	<b>using namespace llvm;</b>
				316
				317	<b>namespace</b> {
				318	<b>struct Hello</b> : <b>public</b> <a href="#FunctionPass">FunctionPass</a> {
				319
				320	static char ID;
				321	Hello() : FunctionPass((intptr_t)&ID) {}
				322
				323	<b>virtual bool</b> <a href="#runOnFunction">runOnFunction</a>(Function &F) {
				324	llvm::cerr << "<i>Hello: </i>" << F.getName() << "\n";
				325	<b>return false</b>;
				326	}
				327	};
				328
Devang Patel	8e46f05	2007-07-25 21:05:39 +0000	[diff] [blame]	329	char Hello::ID = 0;
Dan Gohman	f17a25c	2007-07-18 16:29:46 +0000	[diff] [blame]	330	RegisterPass<Hello> X("<i>hello</i>", "<i>Hello World Pass</i>");
				331	}
				332	</pre></div>
				333
				334	<p>Now that it's all together, compile the file with a simple "<tt>gmake</tt>"
				335	command in the local directory and you should get a new
				336	"<tt>Debug/lib/Hello.so</tt> file. Note that everything in this file is
				337	contained in an anonymous namespace: this reflects the fact that passes are self
				338	contained units that do not need external interfaces (although they can have
				339	them) to be useful.</p>
				340
				341	</div>
				342
				343	<!-- ======================================================================= -->
				344	<div class="doc_subsection">
				345	<a name="running">Running a pass with <tt>opt</tt></a>
				346	</div>
				347
				348	<div class="doc_text">
				349
				350	<p>Now that you have a brand new shiny shared object file, we can use the
				351	<tt>opt</tt> command to run an LLVM program through your pass. Because you
				352	registered your pass with the <tt>RegisterPass</tt> template, you will be able to
				353	use the <tt>opt</tt> tool to access it, once loaded.</p>
				354
				355	<p>To test it, follow the example at the end of the <a
				356	href="GettingStarted.html">Getting Started Guide</a> to compile "Hello World" to
				357	LLVM. We can now run the bitcode file (<tt>hello.bc</tt>) for the program
				358	through our transformation like this (or course, any bitcode file will
				359	work):</p>
				360
				361	<div class="doc_code"><pre>
				362	$ opt -load ../../../Debug/lib/Hello.so -hello < hello.bc > /dev/null
				363	Hello: __main
				364	Hello: puts
				365	Hello: main
				366	</pre></div>
				367
				368	<p>The '<tt>-load</tt>' option specifies that '<tt>opt</tt>' should load your
				369	pass as a shared object, which makes '<tt>-hello</tt>' a valid command line
				370	argument (which is one reason you need to <a href="#registration">register your
				371	pass</a>). Because the hello pass does not modify the program in any
				372	interesting way, we just throw away the result of <tt>opt</tt> (sending it to
				373	<tt>/dev/null</tt>).</p>
				374
				375	<p>To see what happened to the other string you registered, try running
				376	<tt>opt</tt> with the <tt>--help</tt> option:</p>
				377
				378	<div class="doc_code"><pre>
				379	$ opt -load ../../../Debug/lib/Hello.so --help
				380	OVERVIEW: llvm .bc -> .bc modular optimizer
				381
				382	USAGE: opt [options] <input bitcode>
				383
				384	OPTIONS:
				385	Optimizations available:
				386	...
				387	-funcresolve - Resolve Functions
				388	-gcse - Global Common Subexpression Elimination
				389	-globaldce - Dead Global Elimination
				390	<b>-hello - Hello World Pass</b>
				391	-indvars - Canonicalize Induction Variables
				392	-inline - Function Integration/Inlining
				393	-instcombine - Combine redundant instructions
				394	...
				395	</pre></div>
				396
				397	<p>The pass name get added as the information string for your pass, giving some
				398	documentation to users of <tt>opt</tt>. Now that you have a working pass, you
				399	would go ahead and make it do the cool transformations you want. Once you get
				400	it all working and tested, it may become useful to find out how fast your pass
				401	is. The <a href="#passManager"><tt>PassManager</tt></a> provides a nice command
				402	line option (<tt>--time-passes</tt>) that allows you to get information about
				403	the execution time of your pass along with the other passes you queue up. For
				404	example:</p>
				405
				406	<div class="doc_code"><pre>
				407	$ opt -load ../../../Debug/lib/Hello.so -hello -time-passes < hello.bc > /dev/null
				408	Hello: __main
				409	Hello: puts
				410	Hello: main
				411	===============================================================================
				412	... Pass execution timing report ...
				413	===============================================================================
				414	Total Execution Time: 0.02 seconds (0.0479059 wall clock)
				415
				416	---User Time--- --System Time-- --User+System-- ---Wall Time--- --- Pass Name ---
				417	0.0100 (100.0%) 0.0000 ( 0.0%) 0.0100 ( 50.0%) 0.0402 ( 84.0%) Bitcode Writer
				418	0.0000 ( 0.0%) 0.0100 (100.0%) 0.0100 ( 50.0%) 0.0031 ( 6.4%) Dominator Set Construction
				419	0.0000 ( 0.0%) 0.0000 ( 0.0%) 0.0000 ( 0.0%) 0.0013 ( 2.7%) Module Verifier
				420	<b> 0.0000 ( 0.0%) 0.0000 ( 0.0%) 0.0000 ( 0.0%) 0.0033 ( 6.9%) Hello World Pass</b>
				421	0.0100 (100.0%) 0.0100 (100.0%) 0.0200 (100.0%) 0.0479 (100.0%) TOTAL
				422	</pre></div>
				423
				424	<p>As you can see, our implementation above is pretty fast :). The additional
				425	passes listed are automatically inserted by the '<tt>opt</tt>' tool to verify
				426	that the LLVM emitted by your pass is still valid and well formed LLVM, which
				427	hasn't been broken somehow.</p>
				428
				429	<p>Now that you have seen the basics of the mechanics behind passes, we can talk
				430	about some more details of how they work and how to use them.</p>
				431
				432	</div>
				433
				434	<!-- *********************************************************************** -->
				435	<div class="doc_section">
				436	<a name="passtype">Pass classes and requirements</a>
				437	</div>
				438	<!-- *********************************************************************** -->
				439
				440	<div class="doc_text">
				441
				442	<p>One of the first things that you should do when designing a new pass is to
				443	decide what class you should subclass for your pass. The <a
				444	href="#basiccode">Hello World</a> example uses the <tt><a
				445	href="#FunctionPass">FunctionPass</a></tt> class for its implementation, but we
				446	did not discuss why or when this should occur. Here we talk about the classes
				447	available, from the most general to the most specific.</p>
				448
				449	<p>When choosing a superclass for your Pass, you should choose the <b>most
				450	specific</b> class possible, while still being able to meet the requirements
				451	listed. This gives the LLVM Pass Infrastructure information necessary to
				452	optimize how passes are run, so that the resultant compiler isn't unneccesarily
				453	slow.</p>
				454
				455	</div>
				456
				457	<!-- ======================================================================= -->
				458	<div class="doc_subsection">
				459	<a name="ImmutablePass">The <tt>ImmutablePass</tt> class</a>
				460	</div>
				461
				462	<div class="doc_text">
				463
				464	<p>The most plain and boring type of pass is the "<tt><a
				465	href="http://llvm.org/doxygen/classllvm_1_1ImmutablePass.html">ImmutablePass</a></tt>"
				466	class. This pass type is used for passes that do not have to be run, do not
				467	change state, and never need to be updated. This is not a normal type of
				468	transformation or analysis, but can provide information about the current
				469	compiler configuration.</p>
				470
				471	<p>Although this pass class is very infrequently used, it is important for
				472	providing information about the current target machine being compiled for, and
				473	other static information that can affect the various transformations.</p>
				474
				475	<p><tt>ImmutablePass</tt>es never invalidate other transformations, are never
				476	invalidated, and are never "run".</p>
				477
				478	</div>
				479
				480	<!-- ======================================================================= -->
				481	<div class="doc_subsection">
				482	<a name="ModulePass">The <tt>ModulePass</tt> class</a>
				483	</div>
				484
				485	<div class="doc_text">
				486
				487	<p>The "<tt><a
				488	href="http://llvm.org/doxygen/classllvm_1_1ModulePass.html">ModulePass</a></tt>"
				489	class is the most general of all superclasses that you can use. Deriving from
				490	<tt>ModulePass</tt> indicates that your pass uses the entire program as a unit,
				491	refering to function bodies in no predictable order, or adding and removing
				492	functions. Because nothing is known about the behavior of <tt>ModulePass</tt>
				493	subclasses, no optimization can be done for their execution. A module pass
				494	can use function level passes (e.g. dominators) using getAnalysis interface
				495	<tt> getAnalysis<DominatorTree>(Function)</tt>. </p>
				496
				497	<p>To write a correct <tt>ModulePass</tt> subclass, derive from
				498	<tt>ModulePass</tt> and overload the <tt>runOnModule</tt> method with the
				499	following signature:</p>
				500
				501	</div>
				502
				503	<!-- _______________________________________________________________________ -->
				504	<div class="doc_subsubsection">
				505	<a name="runOnModule">The <tt>runOnModule</tt> method</a>
				506	</div>
				507
				508	<div class="doc_text">
				509
				510	<div class="doc_code"><pre>
				511	<b>virtual bool</b> runOnModule(Module &M) = 0;
				512	</pre></div>
				513
				514	<p>The <tt>runOnModule</tt> method performs the interesting work of the pass.
				515	It should return true if the module was modified by the transformation and
				516	false otherwise.</p>
				517
				518	</div>
				519
				520	<!-- ======================================================================= -->
				521	<div class="doc_subsection">
				522	<a name="CallGraphSCCPass">The <tt>CallGraphSCCPass</tt> class</a>
				523	</div>
				524
				525	<div class="doc_text">
				526
				527	<p>The "<tt><a
				528	href="http://llvm.org/doxygen/classllvm_1_1CallGraphSCCPass.html">CallGraphSCCPass</a></tt>"
				529	is used by passes that need to traverse the program bottom-up on the call graph
				530	(callees before callers). Deriving from CallGraphSCCPass provides some
				531	mechanics for building and traversing the CallGraph, but also allows the system
				532	to optimize execution of CallGraphSCCPass's. If your pass meets the
				533	requirements outlined below, and doesn't meet the requirements of a <tt><a
				534	href="#FunctionPass">FunctionPass</a></tt> or <tt><a
				535	href="#BasicBlockPass">BasicBlockPass</a></tt>, you should derive from
				536	<tt>CallGraphSCCPass</tt>.</p>
				537
				538	<p><b>TODO</b>: explain briefly what SCC, Tarjan's algo, and B-U mean.</p>
				539
				540	<p>To be explicit, <tt>CallGraphSCCPass</tt> subclasses are:</p>
				541
				542	<ol>
				543
				544	<li>... <em>not allowed</em> to modify any <tt>Function</tt>s that are not in
				545	the current SCC.</li>
				546
				547	<li>... <em>not allowed</em> to inspect any Function's other than those in the
				548	current SCC and the direct callees of the SCC.</li>
				549
				550	<li>... <em>required</em> to preserve the current CallGraph object, updating it
				551	to reflect any changes made to the program.</li>
				552
				553	<li>... <em>not allowed</em> to add or remove SCC's from the current Module,
				554	though they may change the contents of an SCC.</li>
				555
				556	<li>... <em>allowed</em> to add or remove global variables from the current
				557	Module.</li>
				558
				559	<li>... <em>allowed</em> to maintain state across invocations of
				560	<a href="#runOnSCC"><tt>runOnSCC</tt></a> (including global data).</li>
				561	</ol>
				562
				563	<p>Implementing a <tt>CallGraphSCCPass</tt> is slightly tricky in some cases
				564	because it has to handle SCCs with more than one node in it. All of the virtual
				565	methods described below should return true if they modified the program, or
				566	false if they didn't.</p>
				567
				568	</div>
				569
				570	<!-- _______________________________________________________________________ -->
				571	<div class="doc_subsubsection">
				572	<a name="doInitialization_scc">The <tt>doInitialization(CallGraph &)</tt>
				573	method</a>
				574	</div>
				575
				576	<div class="doc_text">
				577
				578	<div class="doc_code"><pre>
				579	<b>virtual bool</b> doInitialization(CallGraph &CG);
				580	</pre></div>
				581
				582	<p>The <tt>doIninitialize</tt> method is allowed to do most of the things that
				583	<tt>CallGraphSCCPass</tt>'s are not allowed to do. They can add and remove
				584	functions, get pointers to functions, etc. The <tt>doInitialization</tt> method
				585	is designed to do simple initialization type of stuff that does not depend on
				586	the SCCs being processed. The <tt>doInitialization</tt> method call is not
				587	scheduled to overlap with any other pass executions (thus it should be very
				588	fast).</p>
				589
				590	</div>
				591
				592	<!-- _______________________________________________________________________ -->
				593	<div class="doc_subsubsection">
				594	<a name="runOnSCC">The <tt>runOnSCC</tt> method</a>
				595	</div>
				596
				597	<div class="doc_text">
				598
				599	<div class="doc_code"><pre>
				600	<b>virtual bool</b> runOnSCC(const std::vector<CallGraphNode *> &SCCM) = 0;
				601	</pre></div>
				602
				603	<p>The <tt>runOnSCC</tt> method performs the interesting work of the pass, and
				604	should return true if the module was modified by the transformation, false
				605	otherwise.</p>
				606
				607	</div>
				608
				609	<!-- _______________________________________________________________________ -->
				610	<div class="doc_subsubsection">
				611	<a name="doFinalization_scc">The <tt>doFinalization(CallGraph
				612	&)</tt> method</a>
				613	</div>
				614
				615	<div class="doc_text">
				616
				617	<div class="doc_code"><pre>
				618	<b>virtual bool</b> doFinalization(CallGraph &CG);
				619	</pre></div>
				620
				621	<p>The <tt>doFinalization</tt> method is an infrequently used method that is
				622	called when the pass framework has finished calling <a
				623	href="#runOnFunction"><tt>runOnFunction</tt></a> for every function in the
				624	program being compiled.</p>
				625
				626	</div>
				627
				628	<!-- ======================================================================= -->
				629	<div class="doc_subsection">
				630	<a name="FunctionPass">The <tt>FunctionPass</tt> class</a>
				631	</div>
				632
				633	<div class="doc_text">
				634
				635	<p>In contrast to <tt>ModulePass</tt> subclasses, <tt><a
				636	href="http://llvm.org/doxygen/classllvm_1_1Pass.html">FunctionPass</a></tt>
				637	subclasses do have a predictable, local behavior that can be expected by the
				638	system. All <tt>FunctionPass</tt> execute on each function in the program
				639	independent of all of the other functions in the program.
				640	<tt>FunctionPass</tt>'s do not require that they are executed in a particular
				641	order, and <tt>FunctionPass</tt>'s do not modify external functions.</p>
				642
				643	<p>To be explicit, <tt>FunctionPass</tt> subclasses are not allowed to:</p>
				644
				645	<ol>
				646	<li>Modify a Function other than the one currently being processed.</li>
				647	<li>Add or remove Function's from the current Module.</li>
				648	<li>Add or remove global variables from the current Module.</li>
				649	<li>Maintain state across invocations of
				650	<a href="#runOnFunction"><tt>runOnFunction</tt></a> (including global data)</li>
				651	</ol>
				652
				653	<p>Implementing a <tt>FunctionPass</tt> is usually straightforward (See the <a
				654	href="#basiccode">Hello World</a> pass for example). <tt>FunctionPass</tt>'s
				655	may overload three virtual methods to do their work. All of these methods
				656	should return true if they modified the program, or false if they didn't.</p>
				657
				658	</div>
				659
				660	<!-- _______________________________________________________________________ -->
				661	<div class="doc_subsubsection">
				662	<a name="doInitialization_mod">The <tt>doInitialization(Module &)</tt>
				663	method</a>
				664	</div>
				665
				666	<div class="doc_text">
				667
				668	<div class="doc_code"><pre>
				669	<b>virtual bool</b> doInitialization(Module &M);
				670	</pre></div>
				671
				672	<p>The <tt>doIninitialize</tt> method is allowed to do most of the things that
				673	<tt>FunctionPass</tt>'s are not allowed to do. They can add and remove
				674	functions, get pointers to functions, etc. The <tt>doInitialization</tt> method
				675	is designed to do simple initialization type of stuff that does not depend on
				676	the functions being processed. The <tt>doInitialization</tt> method call is not
				677	scheduled to overlap with any other pass executions (thus it should be very
				678	fast).</p>
				679
				680	<p>A good example of how this method should be used is the <a
				681	href="http://llvm.org/doxygen/LowerAllocations_8cpp-source.html">LowerAllocations</a>
				682	pass. This pass converts <tt>malloc</tt> and <tt>free</tt> instructions into
				683	platform dependent <tt>malloc()</tt> and <tt>free()</tt> function calls. It
				684	uses the <tt>doInitialization</tt> method to get a reference to the malloc and
				685	free functions that it needs, adding prototypes to the module if necessary.</p>
				686
				687	</div>
				688
				689	<!-- _______________________________________________________________________ -->
				690	<div class="doc_subsubsection">
				691	<a name="runOnFunction">The <tt>runOnFunction</tt> method</a>
				692	</div>
				693
				694	<div class="doc_text">
				695
				696	<div class="doc_code"><pre>
				697	<b>virtual bool</b> runOnFunction(Function &F) = 0;
				698	</pre></div><p>
				699
				700	<p>The <tt>runOnFunction</tt> method must be implemented by your subclass to do
				701	the transformation or analysis work of your pass. As usual, a true value should
				702	be returned if the function is modified.</p>
				703
				704	</div>
				705
				706	<!-- _______________________________________________________________________ -->
				707	<div class="doc_subsubsection">
				708	<a name="doFinalization_mod">The <tt>doFinalization(Module
				709	&)</tt> method</a>
				710	</div>
				711
				712	<div class="doc_text">
				713
				714	<div class="doc_code"><pre>
				715	<b>virtual bool</b> doFinalization(Module &M);
				716	</pre></div>
				717
				718	<p>The <tt>doFinalization</tt> method is an infrequently used method that is
				719	called when the pass framework has finished calling <a
				720	href="#runOnFunction"><tt>runOnFunction</tt></a> for every function in the
				721	program being compiled.</p>
				722
				723	</div>
				724
				725	<!-- ======================================================================= -->
				726	<div class="doc_subsection">
				727	<a name="LoopPass">The <tt>LoopPass</tt> class </a>
				728	</div>
				729
				730	<div class="doc_text">
				731
				732	<p> All <tt>LoopPass</tt> execute on each loop in the function independent of
				733	all of the other loops in the function. <tt>LoopPass</tt> processes loops in
				734	loop nest order such that outer most loop is processed last. </p>
				735
				736	<p> <tt>LoopPass</tt> subclasses are allowed to update loop nest using
				737	<tt>LPPassManager</tt> interface. Implementing a loop pass is usually
				738	straightforward. <tt>Looppass</tt>'s may overload three virtual methods to
				739	do their work. All these methods should return true if they modified the
				740	program, or false if they didn't. </p>
				741	</div>
				742
				743	<!-- _______________________________________________________________________ -->
				744	<div class="doc_subsubsection">
				745	<a name="doInitialization_loop">The <tt>doInitialization(Loop *,
				746	LPPassManager &)</tt>
				747	method</a>
				748	</div>
				749
				750	<div class="doc_text">
				751
				752	<div class="doc_code"><pre>
				753	<b>virtual bool</b> doInitialization(Loop *, LPPassManager &LPM);
				754	</pre></div>
				755
				756	<p>The <tt>doInitialization</tt> method is designed to do simple initialization
				757	type of stuff that does not depend on the functions being processed. The
				758	<tt>doInitialization</tt> method call is not scheduled to overlap with any
				759	other pass executions (thus it should be very fast). LPPassManager
				760	interface should be used to access Function or Module level analysis
				761	information.</p>
				762
				763	</div>
				764
				765
				766	<!-- _______________________________________________________________________ -->
				767	<div class="doc_subsubsection">
				768	<a name="runOnLoop">The <tt>runOnLoop</tt> method</a>
				769	</div>
				770
				771	<div class="doc_text">
				772
				773	<div class="doc_code"><pre>
				774	<b>virtual bool</b> runOnLoop(Loop *, LPPassManager &LPM) = 0;
				775	</pre></div><p>
				776
				777	<p>The <tt>runOnLoop</tt> method must be implemented by your subclass to do
				778	the transformation or analysis work of your pass. As usual, a true value should
				779	be returned if the function is modified. <tt>LPPassManager</tt> interface
				780	should be used to update loop nest.</p>
				781
				782	</div>
				783
				784	<!-- _______________________________________________________________________ -->
				785	<div class="doc_subsubsection">
				786	<a name="doFinalization_loop">The <tt>doFinalization()</tt> method</a>
				787	</div>
				788
				789	<div class="doc_text">
				790
				791	<div class="doc_code"><pre>
				792	<b>virtual bool</b> doFinalization();
				793	</pre></div>
				794
				795	<p>The <tt>doFinalization</tt> method is an infrequently used method that is
				796	called when the pass framework has finished calling <a
				797	href="#runOnLoop"><tt>runOnLoop</tt></a> for every loop in the
				798	program being compiled. </p>
				799
				800	</div>
				801
				802
				803
				804	<!-- ======================================================================= -->
				805	<div class="doc_subsection">
				806	<a name="BasicBlockPass">The <tt>BasicBlockPass</tt> class</a>
				807	</div>
				808
				809	<div class="doc_text">
				810
				811	<p><tt>BasicBlockPass</tt>'s are just like <a
				812	href="#FunctionPass"><tt>FunctionPass</tt></a>'s, except that they must limit
				813	their scope of inspection and modification to a single basic block at a time.
				814	As such, they are <b>not</b> allowed to do any of the following:</p>
				815
				816	<ol>
				817	<li>Modify or inspect any basic blocks outside of the current one</li>
				818	<li>Maintain state across invocations of
				819	<a href="#runOnBasicBlock"><tt>runOnBasicBlock</tt></a></li>
				820	<li>Modify the control flow graph (by altering terminator instructions)</li>
				821	<li>Any of the things forbidden for
				822	<a href="#FunctionPass"><tt>FunctionPass</tt></a>es.</li>
				823	</ol>
				824
				825	<p><tt>BasicBlockPass</tt>es are useful for traditional local and "peephole"
				826	optimizations. They may override the same <a
				827	href="#doInitialization_mod"><tt>doInitialization(Module &)</tt></a> and <a
				828	href="#doFinalization_mod"><tt>doFinalization(Module &)</tt></a> methods that <a
				829	href="#FunctionPass"><tt>FunctionPass</tt></a>'s have, but also have the following virtual methods that may also be implemented:</p>
				830
				831	</div>
				832
				833	<!-- _______________________________________________________________________ -->
				834	<div class="doc_subsubsection">
				835	<a name="doInitialization_fn">The <tt>doInitialization(Function
				836	&)</tt> method</a>
				837	</div>
				838
				839	<div class="doc_text">
				840
				841	<div class="doc_code"><pre>
				842	<b>virtual bool</b> doInitialization(Function &F);
				843	</pre></div>
				844
				845	<p>The <tt>doIninitialize</tt> method is allowed to do most of the things that
				846	<tt>BasicBlockPass</tt>'s are not allowed to do, but that
				847	<tt>FunctionPass</tt>'s can. The <tt>doInitialization</tt> method is designed
				848	to do simple initialization that does not depend on the
				849	BasicBlocks being processed. The <tt>doInitialization</tt> method call is not
				850	scheduled to overlap with any other pass executions (thus it should be very
				851	fast).</p>
				852
				853	</div>
				854
				855	<!-- _______________________________________________________________________ -->
				856	<div class="doc_subsubsection">
				857	<a name="runOnBasicBlock">The <tt>runOnBasicBlock</tt> method</a>
				858	</div>
				859
				860	<div class="doc_text">
				861
				862	<div class="doc_code"><pre>
				863	<b>virtual bool</b> runOnBasicBlock(BasicBlock &BB) = 0;
				864	</pre></div>
				865
				866	<p>Override this function to do the work of the <tt>BasicBlockPass</tt>. This
				867	function is not allowed to inspect or modify basic blocks other than the
				868	parameter, and are not allowed to modify the CFG. A true value must be returned
				869	if the basic block is modified.</p>
				870
				871	</div>
				872
				873	<!-- _______________________________________________________________________ -->
				874	<div class="doc_subsubsection">
				875	<a name="doFinalization_fn">The <tt>doFinalization(Function &)</tt>
				876	method</a>
				877	</div>
				878
				879	<div class="doc_text">
				880
				881	<div class="doc_code"><pre>
				882	<b>virtual bool</b> doFinalization(Function &F);
				883	</pre></div>
				884
				885	<p>The <tt>doFinalization</tt> method is an infrequently used method that is
				886	called when the pass framework has finished calling <a
				887	href="#runOnBasicBlock"><tt>runOnBasicBlock</tt></a> for every BasicBlock in the
				888	program being compiled. This can be used to perform per-function
				889	finalization.</p>
				890
				891	</div>
				892
				893	<!-- ======================================================================= -->
				894	<div class="doc_subsection">
				895	<a name="MachineFunctionPass">The <tt>MachineFunctionPass</tt> class</a>
				896	</div>
				897
				898	<div class="doc_text">
				899
				900	<p>A <tt>MachineFunctionPass</tt> is a part of the LLVM code generator that
				901	executes on the machine-dependent representation of each LLVM function in the
				902	program. A <tt>MachineFunctionPass</tt> is also a <tt>FunctionPass</tt>, so all
				903	the restrictions that apply to a <tt>FunctionPass</tt> also apply to it.
				904	<tt>MachineFunctionPass</tt>es also have additional restrictions. In particular,
				905	<tt>MachineFunctionPass</tt>es are not allowed to do any of the following:</p>
				906
				907	<ol>
				908	<li>Modify any LLVM Instructions, BasicBlocks or Functions.</li>
				909	<li>Modify a MachineFunction other than the one currently being processed.</li>
				910	<li>Add or remove MachineFunctions from the current Module.</li>
				911	<li>Add or remove global variables from the current Module.</li>
				912	<li>Maintain state across invocations of <a
				913	href="#runOnMachineFunction"><tt>runOnMachineFunction</tt></a> (including global
				914	data)</li>
				915	</ol>
				916
				917	</div>
				918
				919	<!-- _______________________________________________________________________ -->
				920	<div class="doc_subsubsection">
				921	<a name="runOnMachineFunction">The <tt>runOnMachineFunction(MachineFunction
				922	&MF)</tt> method</a>
				923	</div>
				924
				925	<div class="doc_text">
				926
				927	<div class="doc_code"><pre>
				928	<b>virtual bool</b> runOnMachineFunction(MachineFunction &MF) = 0;
				929	</pre></div>
				930
				931	<p><tt>runOnMachineFunction</tt> can be considered the main entry point of a
				932	<tt>MachineFunctionPass</tt>; that is, you should override this method to do the
				933	work of your <tt>MachineFunctionPass</tt>.</p>
				934
				935	<p>The <tt>runOnMachineFunction</tt> method is called on every
				936	<tt>MachineFunction</tt> in a <tt>Module</tt>, so that the
				937	<tt>MachineFunctionPass</tt> may perform optimizations on the machine-dependent
				938	representation of the function. If you want to get at the LLVM <tt>Function</tt>
				939	for the <tt>MachineFunction</tt> you're working on, use
				940	<tt>MachineFunction</tt>'s <tt>getFunction()</tt> accessor method -- but
				941	remember, you may not modify the LLVM <tt>Function</tt> or its contents from a
				942	<tt>MachineFunctionPass</tt>.</p>
				943
				944	</div>
				945
				946	<!-- *********************************************************************** -->
				947	<div class="doc_section">
				948	<a name="registration">Pass registration</a>
				949	</div>
				950	<!-- *********************************************************************** -->
				951
				952	<div class="doc_text">
				953
				954	<p>In the <a href="#basiccode">Hello World</a> example pass we illustrated how
				955	pass registration works, and discussed some of the reasons that it is used and
				956	what it does. Here we discuss how and why passes are registered.</p>
				957
				958	<p>As we saw above, passes are registered with the <b><tt>RegisterPass</tt></b>
				959	template, which requires you to pass at least two
				960	parameters. The first parameter is the name of the pass that is to be used on
				961	the command line to specify that the pass should be added to a program (for
				962	example, with <tt>opt</tt> or <tt>bugpoint</tt>). The second argument is the
				963	name of the pass, which is to be used for the <tt>--help</tt> output of
				964	programs, as
				965	well as for debug output generated by the <tt>--debug-pass</tt> option.</p>
				966
				967	<p>If you want your pass to be easily dumpable, you should
				968	implement the virtual <tt>print</tt> method:</p>
				969
				970	</div>
				971
				972	<!-- _______________________________________________________________________ -->
				973	<div class="doc_subsubsection">
				974	<a name="print">The <tt>print</tt> method</a>
				975	</div>
				976
				977	<div class="doc_text">
				978
				979	<div class="doc_code"><pre>
				980	<b>virtual void</b> print(llvm::OStream &O, <b>const</b> Module *M) <b>const</b>;
				981	</pre></div>
				982
				983	<p>The <tt>print</tt> method must be implemented by "analyses" in order to print
				984	a human readable version of the analysis results. This is useful for debugging
				985	an analysis itself, as well as for other people to figure out how an analysis
				986	works. Use the <tt>opt -analyze</tt> argument to invoke this method.</p>
				987
				988	<p>The <tt>llvm::OStream</tt> parameter specifies the stream to write the results on,
				989	and the <tt>Module</tt> parameter gives a pointer to the top level module of the
				990	program that has been analyzed. Note however that this pointer may be null in
				991	certain circumstances (such as calling the <tt>Pass::dump()</tt> from a
				992	debugger), so it should only be used to enhance debug output, it should not be
				993	depended on.</p>
				994
				995	</div>
				996
				997	<!-- *********************************************************************** -->
				998	<div class="doc_section">
				999	<a name="interaction">Specifying interactions between passes</a>
				1000	</div>
				1001	<!-- *********************************************************************** -->
				1002
				1003	<div class="doc_text">
				1004
John Criswell	a99e43f	2007-12-03 19:34:25 +0000	[diff] [blame]	1005	<p>One of the main responsibilities of the <tt>PassManager</tt> is to make sure
Dan Gohman	f17a25c	2007-07-18 16:29:46 +0000	[diff] [blame]	1006	that passes interact with each other correctly. Because <tt>PassManager</tt>
				1007	tries to <a href="#passmanager">optimize the execution of passes</a> it must
				1008	know how the passes interact with each other and what dependencies exist between
				1009	the various passes. To track this, each pass can declare the set of passes that
				1010	are required to be executed before the current pass, and the passes which are
				1011	invalidated by the current pass.</p>
				1012
				1013	<p>Typically this functionality is used to require that analysis results are
				1014	computed before your pass is run. Running arbitrary transformation passes can
				1015	invalidate the computed analysis results, which is what the invalidation set
				1016	specifies. If a pass does not implement the <tt><a
				1017	href="#getAnalysisUsage">getAnalysisUsage</a></tt> method, it defaults to not
				1018	having any prerequisite passes, and invalidating <b>all</b> other passes.</p>
				1019
				1020	</div>
				1021
				1022	<!-- _______________________________________________________________________ -->
				1023	<div class="doc_subsubsection">
				1024	<a name="getAnalysisUsage">The <tt>getAnalysisUsage</tt> method</a>
				1025	</div>
				1026
				1027	<div class="doc_text">
				1028
				1029	<div class="doc_code"><pre>
				1030	<b>virtual void</b> getAnalysisUsage(AnalysisUsage &Info) <b>const</b>;
				1031	</pre></div>
				1032
				1033	<p>By implementing the <tt>getAnalysisUsage</tt> method, the required and
				1034	invalidated sets may be specified for your transformation. The implementation
				1035	should fill in the <tt><a
				1036	href="http://llvm.org/doxygen/classllvm_1_1AnalysisUsage.html">AnalysisUsage</a></tt>
				1037	object with information about which passes are required and not invalidated. To
				1038	do this, a pass may call any of the following methods on the AnalysisUsage
				1039	object:</p>
				1040	</div>
				1041
				1042	<!-- _______________________________________________________________________ -->
				1043	<div class="doc_subsubsection">
				1044	<a name="AU::addRequired">The <tt>AnalysisUsage::addRequired<></tt> and <tt>AnalysisUsage::addRequiredTransitive<></tt> methods</a>
				1045	</div>
				1046
				1047	<div class="doc_text">
				1048	<p>
				1049	If your pass requires a previous pass to be executed (an analysis for example),
				1050	it can use one of these methods to arrange for it to be run before your pass.
				1051	LLVM has many different types of analyses and passes that can be required,
				1052	spanning the range from <tt>DominatorSet</tt> to <tt>BreakCriticalEdges</tt>.
				1053	Requiring <tt>BreakCriticalEdges</tt>, for example, guarantees that there will
				1054	be no critical edges in the CFG when your pass has been run.
				1055	</p>
				1056
				1057	<p>
				1058	Some analyses chain to other analyses to do their job. For example, an <a
				1059	href="AliasAnalysis.html">AliasAnalysis</a> implementation is required to <a
				1060	href="AliasAnalysis.html#chaining">chain</a> to other alias analysis passes. In
				1061	cases where analyses chain, the <tt>addRequiredTransitive</tt> method should be
				1062	used instead of the <tt>addRequired</tt> method. This informs the PassManager
				1063	that the transitively required pass should be alive as long as the requiring
				1064	pass is.
				1065	</p>
				1066	</div>
				1067
				1068	<!-- _______________________________________________________________________ -->
				1069	<div class="doc_subsubsection">
				1070	<a name="AU::addPreserved">The <tt>AnalysisUsage::addPreserved<></tt> method</a>
				1071	</div>
				1072
				1073	<div class="doc_text">
				1074	<p>
				1075	One of the jobs of the PassManager is to optimize how and when analyses are run.
				1076	In particular, it attempts to avoid recomputing data unless it needs to. For
				1077	this reason, passes are allowed to declare that they preserve (i.e., they don't
				1078	invalidate) an existing analysis if it's available. For example, a simple
				1079	constant folding pass would not modify the CFG, so it can't possibly affect the
				1080	results of dominator analysis. By default, all passes are assumed to invalidate
				1081	all others.
				1082	</p>
				1083
				1084	<p>
				1085	The <tt>AnalysisUsage</tt> class provides several methods which are useful in
				1086	certain circumstances that are related to <tt>addPreserved</tt>. In particular,
				1087	the <tt>setPreservesAll</tt> method can be called to indicate that the pass does
				1088	not modify the LLVM program at all (which is true for analyses), and the
				1089	<tt>setPreservesCFG</tt> method can be used by transformations that change
				1090	instructions in the program but do not modify the CFG or terminator instructions
				1091	(note that this property is implicitly set for <a
				1092	href="#BasicBlockPass">BasicBlockPass</a>'s).
				1093	</p>
				1094
				1095	<p>
				1096	<tt>addPreserved</tt> is particularly useful for transformations like
				1097	<tt>BreakCriticalEdges</tt>. This pass knows how to update a small set of loop
				1098	and dominator related analyses if they exist, so it can preserve them, despite
				1099	the fact that it hacks on the CFG.
				1100	</p>
				1101	</div>
				1102
				1103	<!-- _______________________________________________________________________ -->
				1104	<div class="doc_subsubsection">
				1105	<a name="AU::examples">Example implementations of <tt>getAnalysisUsage</tt></a>
				1106	</div>
				1107
				1108	<div class="doc_text">
				1109
				1110	<div class="doc_code"><pre>
				1111	<i>// This is an example implementation from an analysis, which does not modify
				1112	// the program at all, yet has a prerequisite.</i>
				1113	<b>void</b> <a href="http://llvm.org/doxygen/classllvm_1_1PostDominanceFrontier.html">PostDominanceFrontier</a>::getAnalysisUsage(AnalysisUsage &AU) <b>const</b> {
				1114	AU.setPreservesAll();
				1115	AU.addRequired<<a href="http://llvm.org/doxygen/classllvm_1_1PostDominatorTree.html">PostDominatorTree</a>>();
				1116	}
				1117	</pre></div>
				1118
				1119	<p>and:</p>
				1120
				1121	<div class="doc_code"><pre>
				1122	<i>// This example modifies the program, but does not modify the CFG</i>
				1123	<b>void</b> <a href="http://llvm.org/doxygen/structLICM.html">LICM</a>::getAnalysisUsage(AnalysisUsage &AU) <b>const</b> {
				1124	AU.setPreservesCFG();
				1125	AU.addRequired<<a href="http://llvm.org/doxygen/classllvm_1_1LoopInfo.html">LoopInfo</a>>();
				1126	}
				1127	</pre></div>
				1128
				1129	</div>
				1130
				1131	<!-- _______________________________________________________________________ -->
				1132	<div class="doc_subsubsection">
				1133	<a name="getAnalysis">The <tt>getAnalysis<></tt> and <tt>getAnalysisToUpdate<></tt> methods</a>
				1134	</div>
				1135
				1136	<div class="doc_text">
				1137
				1138	<p>The <tt>Pass::getAnalysis<></tt> method is automatically inherited by
				1139	your class, providing you with access to the passes that you declared that you
				1140	required with the <a href="#getAnalysisUsage"><tt>getAnalysisUsage</tt></a>
				1141	method. It takes a single template argument that specifies which pass class you
				1142	want, and returns a reference to that pass. For example:</p>
				1143
				1144	<div class="doc_code"><pre>
				1145	bool LICM::runOnFunction(Function &F) {
				1146	LoopInfo &LI = getAnalysis<LoopInfo>();
				1147	...
				1148	}
				1149	</pre></div>
				1150
				1151	<p>This method call returns a reference to the pass desired. You may get a
				1152	runtime assertion failure if you attempt to get an analysis that you did not
				1153	declare as required in your <a
				1154	href="#getAnalysisUsage"><tt>getAnalysisUsage</tt></a> implementation. This
				1155	method can be called by your <tt>run*</tt> method implementation, or by any
				1156	other local method invoked by your <tt>run*</tt> method.
				1157
				1158	A module level pass can use function level analysis info using this interface.
				1159	For example:</p>
				1160
				1161	<div class="doc_code"><pre>
				1162	bool ModuleLevelPass::runOnModule(Module &M) {
				1163	...
				1164	DominatorTree &DT = getAnalysis<DominatorTree>(Func);
				1165	...
				1166	}
				1167	</pre></div>
				1168
				1169	<p>In above example, runOnFunction for DominatorTree is called by pass manager
				1170	before returning a reference to the desired pass.</p>
				1171
				1172	<p>
				1173	If your pass is capable of updating analyses if they exist (e.g.,
				1174	<tt>BreakCriticalEdges</tt>, as described above), you can use the
				1175	<tt>getAnalysisToUpdate</tt> method, which returns a pointer to the analysis if
				1176	it is active. For example:</p>
				1177
				1178	<div class="doc_code"><pre>
				1179	...
				1180	if (DominatorSet *DS = getAnalysisToUpdate<DominatorSet>()) {
				1181	<i>// A DominatorSet is active. This code will update it.</i>
				1182	}
				1183	...
				1184	</pre></div>
				1185
				1186	</div>
				1187
				1188	<!-- *********************************************************************** -->
				1189	<div class="doc_section">
				1190	<a name="analysisgroup">Implementing Analysis Groups</a>
				1191	</div>
				1192	<!-- *********************************************************************** -->
				1193
				1194	<div class="doc_text">
				1195
Chris Lattner	942c395	2007-11-16 05:32:05 +0000	[diff] [blame]	1196	<p>Now that we understand the basics of how passes are defined, how they are
Dan Gohman	f17a25c	2007-07-18 16:29:46 +0000	[diff] [blame]	1197	used, and how they are required from other passes, it's time to get a little bit
				1198	fancier. All of the pass relationships that we have seen so far are very
				1199	simple: one pass depends on one other specific pass to be run before it can run.
				1200	For many applications, this is great, for others, more flexibility is
				1201	required.</p>
				1202
				1203	<p>In particular, some analyses are defined such that there is a single simple
				1204	interface to the analysis results, but multiple ways of calculating them.
				1205	Consider alias analysis for example. The most trivial alias analysis returns
				1206	"may alias" for any alias query. The most sophisticated analysis a
				1207	flow-sensitive, context-sensitive interprocedural analysis that can take a
				1208	significant amount of time to execute (and obviously, there is a lot of room
				1209	between these two extremes for other implementations). To cleanly support
				1210	situations like this, the LLVM Pass Infrastructure supports the notion of
				1211	Analysis Groups.</p>
				1212
				1213	</div>
				1214
				1215	<!-- _______________________________________________________________________ -->
				1216	<div class="doc_subsubsection">
				1217	<a name="agconcepts">Analysis Group Concepts</a>
				1218	</div>
				1219
				1220	<div class="doc_text">
				1221
				1222	<p>An Analysis Group is a single simple interface that may be implemented by
				1223	multiple different passes. Analysis Groups can be given human readable names
				1224	just like passes, but unlike passes, they need not derive from the <tt>Pass</tt>
				1225	class. An analysis group may have one or more implementations, one of which is
				1226	the "default" implementation.</p>
				1227
				1228	<p>Analysis groups are used by client passes just like other passes are: the
				1229	<tt>AnalysisUsage::addRequired()</tt> and <tt>Pass::getAnalysis()</tt> methods.
				1230	In order to resolve this requirement, the <a href="#passmanager">PassManager</a>
				1231	scans the available passes to see if any implementations of the analysis group
				1232	are available. If none is available, the default implementation is created for
				1233	the pass to use. All standard rules for <A href="#interaction">interaction
				1234	between passes</a> still apply.</p>
				1235
				1236	<p>Although <a href="#registration">Pass Registration</a> is optional for normal
				1237	passes, all analysis group implementations must be registered, and must use the
				1238	<A href="#registerag"><tt>RegisterAnalysisGroup</tt></a> template to join the
				1239	implementation pool. Also, a default implementation of the interface
				1240	<b>must</b> be registered with <A
				1241	href="#registerag"><tt>RegisterAnalysisGroup</tt></a>.</p>
				1242
				1243	<p>As a concrete example of an Analysis Group in action, consider the <a
				1244	href="http://llvm.org/doxygen/classllvm_1_1AliasAnalysis.html">AliasAnalysis</a>
				1245	analysis group. The default implementation of the alias analysis interface (the
				1246	<tt><a
				1247	href="http://llvm.org/doxygen/structBasicAliasAnalysis.html">basicaa</a></tt>
				1248	pass) just does a few simple checks that don't require significant analysis to
				1249	compute (such as: two different globals can never alias each other, etc).
				1250	Passes that use the <tt><a
				1251	href="http://llvm.org/doxygen/classllvm_1_1AliasAnalysis.html">AliasAnalysis</a></tt>
				1252	interface (for example the <tt><a
				1253	href="http://llvm.org/doxygen/structGCSE.html">gcse</a></tt> pass), do
				1254	not care which implementation of alias analysis is actually provided, they just
				1255	use the designated interface.</p>
				1256
				1257	<p>From the user's perspective, commands work just like normal. Issuing the
				1258	command '<tt>opt -gcse ...</tt>' will cause the <tt>basicaa</tt> class to be
				1259	instantiated and added to the pass sequence. Issuing the command '<tt>opt
				1260	-somefancyaa -gcse ...</tt>' will cause the <tt>gcse</tt> pass to use the
				1261	<tt>somefancyaa</tt> alias analysis (which doesn't actually exist, it's just a
				1262	hypothetical example) instead.</p>
				1263
				1264	</div>
				1265
				1266	<!-- _______________________________________________________________________ -->
				1267	<div class="doc_subsubsection">
				1268	<a name="registerag">Using <tt>RegisterAnalysisGroup</tt></a>
				1269	</div>
				1270
				1271	<div class="doc_text">
				1272
				1273	<p>The <tt>RegisterAnalysisGroup</tt> template is used to register the analysis
				1274	group itself as well as add pass implementations to the analysis group. First,
				1275	an analysis should be registered, with a human readable name provided for it.
				1276	Unlike registration of passes, there is no command line argument to be specified
				1277	for the Analysis Group Interface itself, because it is "abstract":</p>
				1278
				1279	<div class="doc_code"><pre>
				1280	<b>static</b> RegisterAnalysisGroup<<a href="http://llvm.org/doxygen/classllvm_1_1AliasAnalysis.html">AliasAnalysis</a>> A("<i>Alias Analysis</i>");
				1281	</pre></div>
				1282
				1283	<p>Once the analysis is registered, passes can declare that they are valid
				1284	implementations of the interface by using the following code:</p>
				1285
				1286	<div class="doc_code"><pre>
				1287	<b>namespace</b> {
				1288	//<i> Analysis Group implementations <b>must</b> be registered normally...</i>
				1289	RegisterPass<FancyAA>
				1290	B("<i>somefancyaa</i>", "<i>A more complex alias analysis implementation</i>");
				1291
				1292	//<i> Declare that we implement the AliasAnalysis interface</i>
				1293	RegisterAnalysisGroup<<a href="http://llvm.org/doxygen/classllvm_1_1AliasAnalysis.html">AliasAnalysis</a>> C(B);
				1294	}
				1295	</pre></div>
				1296
				1297	<p>This just shows a class <tt>FancyAA</tt> that is registered normally, then
				1298	uses the <tt>RegisterAnalysisGroup</tt> template to "join" the <tt><a
				1299	href="http://llvm.org/doxygen/classllvm_1_1AliasAnalysis.html">AliasAnalysis</a></tt>
				1300	analysis group. Every implementation of an analysis group should join using
				1301	this template. A single pass may join multiple different analysis groups with
				1302	no problem.</p>
				1303
				1304	<div class="doc_code"><pre>
				1305	<b>namespace</b> {
				1306	//<i> Analysis Group implementations <b>must</b> be registered normally...</i>
				1307	RegisterPass<<a href="http://llvm.org/doxygen/structBasicAliasAnalysis.html">BasicAliasAnalysis</a>>
				1308	D("<i>basicaa</i>", "<i>Basic Alias Analysis (default AA impl)</i>");
				1309
				1310	//<i> Declare that we implement the AliasAnalysis interface</i>
				1311	RegisterAnalysisGroup<<a href="http://llvm.org/doxygen/classllvm_1_1AliasAnalysis.html">AliasAnalysis</a>, <b>true</b>> E(D);
				1312	}
				1313	</pre></div>
				1314
				1315	<p>Here we show how the default implementation is specified (using the extra
				1316	argument to the <tt>RegisterAnalysisGroup</tt> template). There must be exactly
				1317	one default implementation available at all times for an Analysis Group to be
				1318	used. Only default implementation can derive from <tt>ImmutablePass</tt>.
				1319	Here we declare that the
				1320	<tt><a href="http://llvm.org/doxygen/structBasicAliasAnalysis.html">BasicAliasAnalysis</a></tt>
				1321	pass is the default implementation for the interface.</p>
				1322
				1323	</div>
				1324
				1325	<!-- *********************************************************************** -->
				1326	<div class="doc_section">
				1327	<a name="passStatistics">Pass Statistics</a>
				1328	</div>
				1329	<!-- *********************************************************************** -->
				1330
				1331	<div class="doc_text">
				1332	<p>The <a
				1333	href="http://llvm.org/doxygen/Statistic_8h-source.html"><tt>Statistic</tt></a>
				1334	class is designed to be an easy way to expose various success
				1335	metrics from passes. These statistics are printed at the end of a
				1336	run, when the -stats command line option is enabled on the command
				1337	line. See the <a href="http://llvm.org/docs/ProgrammersManual.html#Statistic">Statistics section</a> in the Programmer's Manual for details.
				1338
				1339	</div>
				1340
				1341
				1342	<!-- *********************************************************************** -->
				1343	<div class="doc_section">
				1344	<a name="passmanager">What PassManager does</a>
				1345	</div>
				1346	<!-- *********************************************************************** -->
				1347
				1348	<div class="doc_text">
				1349
				1350	<p>The <a
				1351	href="http://llvm.org/doxygen/PassManager_8h-source.html"><tt>PassManager</tt></a>
				1352	<a
				1353	href="http://llvm.org/doxygen/classllvm_1_1PassManager.html">class</a>
				1354	takes a list of passes, ensures their <a href="#interaction">prerequisites</a>
				1355	are set up correctly, and then schedules passes to run efficiently. All of the
				1356	LLVM tools that run passes use the <tt>PassManager</tt> for execution of these
				1357	passes.</p>
				1358
				1359	<p>The <tt>PassManager</tt> does two main things to try to reduce the execution
				1360	time of a series of passes:</p>
				1361
				1362	<ol>
				1363	<li><b>Share analysis results</b> - The PassManager attempts to avoid
				1364	recomputing analysis results as much as possible. This means keeping track of
				1365	which analyses are available already, which analyses get invalidated, and which
				1366	analyses are needed to be run for a pass. An important part of work is that the
				1367	<tt>PassManager</tt> tracks the exact lifetime of all analysis results, allowing
				1368	it to <a href="#releaseMemory">free memory</a> allocated to holding analysis
				1369	results as soon as they are no longer needed.</li>
				1370
				1371	<li><b>Pipeline the execution of passes on the program</b> - The
				1372	<tt>PassManager</tt> attempts to get better cache and memory usage behavior out
				1373	of a series of passes by pipelining the passes together. This means that, given
				1374	a series of consequtive <a href="#FunctionPass"><tt>FunctionPass</tt></a>'s, it
				1375	will execute all of the <a href="#FunctionPass"><tt>FunctionPass</tt></a>'s on
				1376	the first function, then all of the <a
				1377	href="#FunctionPass"><tt>FunctionPass</tt></a>es on the second function,
				1378	etc... until the entire program has been run through the passes.
				1379
				1380	<p>This improves the cache behavior of the compiler, because it is only touching
				1381	the LLVM program representation for a single function at a time, instead of
				1382	traversing the entire program. It reduces the memory consumption of compiler,
				1383	because, for example, only one <a
				1384	href="http://llvm.org/doxygen/classllvm_1_1DominatorSet.html"><tt>DominatorSet</tt></a>
John Criswell	8a72615	2007-12-10 20:26:29 +0000	[diff] [blame]	1385	needs to be calculated at a time. This also makes it possible to implement
				1386	some <a
Dan Gohman	f17a25c	2007-07-18 16:29:46 +0000	[diff] [blame]	1387	href="#SMP">interesting enhancements</a> in the future.</p></li>
				1388
				1389	</ol>
				1390
				1391	<p>The effectiveness of the <tt>PassManager</tt> is influenced directly by how
				1392	much information it has about the behaviors of the passes it is scheduling. For
				1393	example, the "preserved" set is intentionally conservative in the face of an
				1394	unimplemented <a href="#getAnalysisUsage"><tt>getAnalysisUsage</tt></a> method.
				1395	Not implementing when it should be implemented will have the effect of not
				1396	allowing any analysis results to live across the execution of your pass.</p>
				1397
				1398	<p>The <tt>PassManager</tt> class exposes a <tt>--debug-pass</tt> command line
				1399	options that is useful for debugging pass execution, seeing how things work, and
				1400	diagnosing when you should be preserving more analyses than you currently are
				1401	(To get information about all of the variants of the <tt>--debug-pass</tt>
				1402	option, just type '<tt>opt --help-hidden</tt>').</p>
				1403
				1404	<p>By using the <tt>--debug-pass=Structure</tt> option, for example, we can see
				1405	how our <a href="#basiccode">Hello World</a> pass interacts with other passes.
				1406	Lets try it out with the <tt>gcse</tt> and <tt>licm</tt> passes:</p>
				1407
				1408	<div class="doc_code"><pre>
				1409	$ opt -load ../../../Debug/lib/Hello.so -gcse -licm --debug-pass=Structure < hello.bc > /dev/null
				1410	Module Pass Manager
				1411	Function Pass Manager
				1412	Dominator Set Construction
				1413	Immediate Dominators Construction
				1414	Global Common Subexpression Elimination
				1415	-- Immediate Dominators Construction
				1416	-- Global Common Subexpression Elimination
				1417	Natural Loop Construction
				1418	Loop Invariant Code Motion
				1419	-- Natural Loop Construction
				1420	-- Loop Invariant Code Motion
				1421	Module Verifier
				1422	-- Dominator Set Construction
				1423	-- Module Verifier
				1424	Bitcode Writer
				1425	--Bitcode Writer
				1426	</pre></div>
				1427
				1428	<p>This output shows us when passes are constructed and when the analysis
				1429	results are known to be dead (prefixed with '<tt>--</tt>'). Here we see that
				1430	GCSE uses dominator and immediate dominator information to do its job. The LICM
				1431	pass uses natural loop information, which uses dominator sets, but not immediate
				1432	dominators. Because immediate dominators are no longer useful after the GCSE
				1433	pass, it is immediately destroyed. The dominator sets are then reused to
				1434	compute natural loop information, which is then used by the LICM pass.</p>
				1435
				1436	<p>After the LICM pass, the module verifier runs (which is automatically added
				1437	by the '<tt>opt</tt>' tool), which uses the dominator set to check that the
				1438	resultant LLVM code is well formed. After it finishes, the dominator set
				1439	information is destroyed, after being computed once, and shared by three
				1440	passes.</p>
				1441
				1442	<p>Lets see how this changes when we run the <a href="#basiccode">Hello
				1443	World</a> pass in between the two passes:</p>
				1444
				1445	<div class="doc_code"><pre>
				1446	$ opt -load ../../../Debug/lib/Hello.so -gcse -hello -licm --debug-pass=Structure < hello.bc > /dev/null
				1447	Module Pass Manager
				1448	Function Pass Manager
				1449	Dominator Set Construction
				1450	Immediate Dominators Construction
				1451	Global Common Subexpression Elimination
				1452	<b>-- Dominator Set Construction</b>
				1453	-- Immediate Dominators Construction
				1454	-- Global Common Subexpression Elimination
				1455	<b> Hello World Pass
				1456	-- Hello World Pass
				1457	Dominator Set Construction</b>
				1458	Natural Loop Construction
				1459	Loop Invariant Code Motion
				1460	-- Natural Loop Construction
				1461	-- Loop Invariant Code Motion
				1462	Module Verifier
				1463	-- Dominator Set Construction
				1464	-- Module Verifier
				1465	Bitcode Writer
				1466	--Bitcode Writer
				1467	Hello: __main
				1468	Hello: puts
				1469	Hello: main
				1470	</pre></div>
				1471
				1472	<p>Here we see that the <a href="#basiccode">Hello World</a> pass has killed the
				1473	Dominator Set pass, even though it doesn't modify the code at all! To fix this,
				1474	we need to add the following <a
				1475	href="#getAnalysisUsage"><tt>getAnalysisUsage</tt></a> method to our pass:</p>
				1476
				1477	<div class="doc_code"><pre>
				1478	<i>// We don't modify the program, so we preserve all analyses</i>
				1479	<b>virtual void</b> getAnalysisUsage(AnalysisUsage &AU) <b>const</b> {
				1480	AU.setPreservesAll();
				1481	}
				1482	</pre></div>
				1483
				1484	<p>Now when we run our pass, we get this output:</p>
				1485
				1486	<div class="doc_code"><pre>
				1487	$ opt -load ../../../Debug/lib/Hello.so -gcse -hello -licm --debug-pass=Structure < hello.bc > /dev/null
				1488	Pass Arguments: -gcse -hello -licm
				1489	Module Pass Manager
				1490	Function Pass Manager
				1491	Dominator Set Construction
				1492	Immediate Dominators Construction
				1493	Global Common Subexpression Elimination
				1494	-- Immediate Dominators Construction
				1495	-- Global Common Subexpression Elimination
				1496	Hello World Pass
				1497	-- Hello World Pass
				1498	Natural Loop Construction
				1499	Loop Invariant Code Motion
				1500	-- Loop Invariant Code Motion
				1501	-- Natural Loop Construction
				1502	Module Verifier
				1503	-- Dominator Set Construction
				1504	-- Module Verifier
				1505	Bitcode Writer
				1506	--Bitcode Writer
				1507	Hello: __main
				1508	Hello: puts
				1509	Hello: main
				1510	</pre></div>
				1511
				1512	<p>Which shows that we don't accidentally invalidate dominator information
				1513	anymore, and therefore do not have to compute it twice.</p>
				1514
				1515	</div>
				1516
				1517	<!-- _______________________________________________________________________ -->
				1518	<div class="doc_subsubsection">
				1519	<a name="releaseMemory">The <tt>releaseMemory</tt> method</a>
				1520	</div>
				1521
				1522	<div class="doc_text">
				1523
				1524	<div class="doc_code"><pre>
				1525	<b>virtual void</b> releaseMemory();
				1526	</pre></div>
				1527
				1528	<p>The <tt>PassManager</tt> automatically determines when to compute analysis
				1529	results, and how long to keep them around for. Because the lifetime of the pass
				1530	object itself is effectively the entire duration of the compilation process, we
				1531	need some way to free analysis results when they are no longer useful. The
				1532	<tt>releaseMemory</tt> virtual method is the way to do this.</p>
				1533
				1534	<p>If you are writing an analysis or any other pass that retains a significant
				1535	amount of state (for use by another pass which "requires" your pass and uses the
				1536	<a href="#getAnalysis">getAnalysis</a> method) you should implement
				1537	<tt>releaseMEmory</tt> to, well, release the memory allocated to maintain this
				1538	internal state. This method is called after the <tt>run*</tt> method for the
				1539	class, before the next call of <tt>run*</tt> in your pass.</p>
				1540
				1541	</div>
				1542
				1543	<!-- *********************************************************************** -->
				1544	<div class="doc_section">
				1545	<a name="registering">Registering dynamically loaded passes</a>
				1546	</div>
				1547	<!-- *********************************************************************** -->
				1548
				1549	<div class="doc_text">
				1550
				1551	<p><i>Size matters</i> when constructing production quality tools using llvm,
				1552	both for the purposes of distribution, and for regulating the resident code size
				1553	when running on the target system. Therefore, it becomes desirable to
				1554	selectively use some passes, while omitting others and maintain the flexibility
				1555	to change configurations later on. You want to be able to do all this, and,
				1556	provide feedback to the user. This is where pass registration comes into
				1557	play.</p>
				1558
				1559	<p>The fundamental mechanisms for pass registration are the
				1560	<tt>MachinePassRegistry</tt> class and subclasses of
				1561	<tt>MachinePassRegistryNode</tt>.</p>
				1562
				1563	<p>An instance of <tt>MachinePassRegistry</tt> is used to maintain a list of
				1564	<tt>MachinePassRegistryNode</tt> objects. This instance maintains the list and
				1565	communicates additions and deletions to the command line interface.</p>
				1566
				1567	<p>An instance of <tt>MachinePassRegistryNode</tt> subclass is used to maintain
				1568	information provided about a particular pass. This information includes the
				1569	command line name, the command help string and the address of the function used
				1570	to create an instance of the pass. A global static constructor of one of these
				1571	instances <i>registers</i> with a corresponding <tt>MachinePassRegistry</tt>,
				1572	the static destructor <i>unregisters</i>. Thus a pass that is statically linked
				1573	in the tool will be registered at start up. A dynamically loaded pass will
				1574	register on load and unregister at unload.</p>
				1575
				1576	</div>
				1577
				1578	<!-- _______________________________________________________________________ -->
				1579	<div class="doc_subsection">
				1580	<a name="registering_existing">Using existing registries</a>
				1581	</div>
				1582
				1583	<div class="doc_text">
				1584
				1585	<p>There are predefined registries to track instruction scheduling
				1586	(<tt>RegisterScheduler</tt>) and register allocation (<tt>RegisterRegAlloc</tt>)
				1587	machine passes. Here we will describe how to <i>register</i> a register
				1588	allocator machine pass.</p>
				1589
				1590	<p>Implement your register allocator machine pass. In your register allocator
				1591	.cpp file add the following include;</p>
				1592
				1593	<div class="doc_code"><pre>
				1594	#include "llvm/CodeGen/RegAllocRegistry.h"
				1595	</pre></div>
				1596
				1597	<p>Also in your register allocator .cpp file, define a creator function in the
				1598	form; </p>
				1599
				1600	<div class="doc_code"><pre>
				1601	FunctionPass *createMyRegisterAllocator() {
				1602	return new MyRegisterAllocator();
				1603	}
				1604	</pre></div>
				1605
				1606	<p>Note that the signature of this function should match the type of
				1607	<tt>RegisterRegAlloc::FunctionPassCtor</tt>. In the same file add the
				1608	"installing" declaration, in the form;</p>
				1609
				1610	<div class="doc_code"><pre>
				1611	static RegisterRegAlloc myRegAlloc("myregalloc",
				1612	" my register allocator help string",
				1613	createMyRegisterAllocator);
				1614	</pre></div>
				1615
				1616	<p>Note the two spaces prior to the help string produces a tidy result on the
				1617	--help query.</p>
				1618
				1619	<div class="doc_code"><pre>
				1620	$ llc --help
				1621	...
				1622	-regalloc - Register allocator to use: (default = linearscan)
				1623	=linearscan - linear scan register allocator
				1624	=local - local register allocator
				1625	=simple - simple register allocator
				1626	=myregalloc - my register allocator help string
				1627	...
				1628	</pre></div>
				1629
				1630	<p>And that's it. The user is now free to use <tt>-regalloc=myregalloc</tt> as
				1631	an option. Registering instruction schedulers is similar except use the
				1632	<tt>RegisterScheduler</tt> class. Note that the
				1633	<tt>RegisterScheduler::FunctionPassCtor</tt> is significantly different from
				1634	<tt>RegisterRegAlloc::FunctionPassCtor</tt>.</p>
				1635
				1636	<p>To force the load/linking of your register allocator into the llc/lli tools,
				1637	add your creator function's global declaration to "Passes.h" and add a "pseudo"
				1638	call line to <tt>llvm/Codegen/LinkAllCodegenComponents.h</tt>.</p>
				1639
				1640	</div>
				1641
				1642
				1643	<!-- _______________________________________________________________________ -->
				1644	<div class="doc_subsection">
				1645	<a name="registering_new">Creating new registries</a>
				1646	</div>
				1647
				1648	<div class="doc_text">
				1649
				1650	<p>The easiest way to get started is to clone one of the existing registries; we
				1651	recommend <tt>llvm/CodeGen/RegAllocRegistry.h</tt>. The key things to modify
				1652	are the class name and the <tt>FunctionPassCtor</tt> type.</p>
				1653
				1654	<p>Then you need to declare the registry. Example: if your pass registry is
				1655	<tt>RegisterMyPasses</tt> then define;</p>
				1656
				1657	<div class="doc_code"><pre>
				1658	MachinePassRegistry RegisterMyPasses::Registry;
				1659	</pre></div>
				1660
				1661	<p>And finally, declare the command line option for your passes. Example:</p>
				1662
				1663	<div class="doc_code"><pre>
				1664	cl::opt<RegisterMyPasses::FunctionPassCtor, false,
				1665	RegisterPassParser<RegisterMyPasses&gt &gt
				1666	MyPassOpt("mypass",
				1667	cl::init(&createDefaultMyPass),
				1668	cl::desc("my pass option help"));
				1669	</pre></div>
				1670
				1671	<p>Here the command option is "mypass", with createDefaultMyPass as the default
				1672	creator.</p>
				1673
				1674	</div>
				1675
				1676	<!-- *********************************************************************** -->
				1677	<div class="doc_section">
				1678	<a name="debughints">Using GDB with dynamically loaded passes</a>
				1679	</div>
				1680	<!-- *********************************************************************** -->
				1681
				1682	<div class="doc_text">
				1683
				1684	<p>Unfortunately, using GDB with dynamically loaded passes is not as easy as it
				1685	should be. First of all, you can't set a breakpoint in a shared object that has
				1686	not been loaded yet, and second of all there are problems with inlined functions
				1687	in shared objects. Here are some suggestions to debugging your pass with
				1688	GDB.</p>
				1689
				1690	<p>For sake of discussion, I'm going to assume that you are debugging a
				1691	transformation invoked by <tt>opt</tt>, although nothing described here depends
				1692	on that.</p>
				1693
				1694	</div>
				1695
				1696	<!-- _______________________________________________________________________ -->
				1697	<div class="doc_subsubsection">
				1698	<a name="breakpoint">Setting a breakpoint in your pass</a>
				1699	</div>
				1700
				1701	<div class="doc_text">
				1702
				1703	<p>First thing you do is start <tt>gdb</tt> on the <tt>opt</tt> process:</p>
				1704
				1705	<div class="doc_code"><pre>
				1706	$ <b>gdb opt</b>
				1707	GNU gdb 5.0
				1708	Copyright 2000 Free Software Foundation, Inc.
				1709	GDB is free software, covered by the GNU General Public License, and you are
				1710	welcome to change it and/or distribute copies of it under certain conditions.
				1711	Type "show copying" to see the conditions.
				1712	There is absolutely no warranty for GDB. Type "show warranty" for details.
				1713	This GDB was configured as "sparc-sun-solaris2.6"...
				1714	(gdb)
				1715	</pre></div>
				1716
				1717	<p>Note that <tt>opt</tt> has a lot of debugging information in it, so it takes
				1718	time to load. Be patient. Since we cannot set a breakpoint in our pass yet
				1719	(the shared object isn't loaded until runtime), we must execute the process, and
				1720	have it stop before it invokes our pass, but after it has loaded the shared
				1721	object. The most foolproof way of doing this is to set a breakpoint in
				1722	<tt>PassManager::run</tt> and then run the process with the arguments you
				1723	want:</p>
				1724
				1725	<div class="doc_code"><pre>
				1726	(gdb) <b>break llvm::PassManager::run</b>
				1727	Breakpoint 1 at 0x2413bc: file Pass.cpp, line 70.
				1728	(gdb) <b>run test.bc -load $(LLVMTOP)/llvm/Debug/lib/[libname].so -[passoption]</b>
				1729	Starting program: opt test.bc -load $(LLVMTOP)/llvm/Debug/lib/[libname].so -[passoption]
				1730	Breakpoint 1, PassManager::run (this=0xffbef174, M=@0x70b298) at Pass.cpp:70
				1731	70 bool PassManager::run(Module &M) { return PM->run(M); }
				1732	(gdb)
				1733	</pre></div>
				1734
				1735	<p>Once the <tt>opt</tt> stops in the <tt>PassManager::run</tt> method you are
				1736	now free to set breakpoints in your pass so that you can trace through execution
				1737	or do other standard debugging stuff.</p>
				1738
				1739	</div>
				1740
				1741	<!-- _______________________________________________________________________ -->
				1742	<div class="doc_subsubsection">
				1743	<a name="debugmisc">Miscellaneous Problems</a>
				1744	</div>
				1745
				1746	<div class="doc_text">
				1747
				1748	<p>Once you have the basics down, there are a couple of problems that GDB has,
				1749	some with solutions, some without.</p>
				1750
				1751	<ul>
				1752	<li>Inline functions have bogus stack information. In general, GDB does a
				1753	pretty good job getting stack traces and stepping through inline functions.
				1754	When a pass is dynamically loaded however, it somehow completely loses this
				1755	capability. The only solution I know of is to de-inline a function (move it
				1756	from the body of a class to a .cpp file).</li>
				1757
				1758	<li>Restarting the program breaks breakpoints. After following the information
				1759	above, you have succeeded in getting some breakpoints planted in your pass. Nex
				1760	thing you know, you restart the program (i.e., you type '<tt>run</tt>' again),
				1761	and you start getting errors about breakpoints being unsettable. The only way I
				1762	have found to "fix" this problem is to <tt>delete</tt> the breakpoints that are
				1763	already set in your pass, run the program, and re-set the breakpoints once
				1764	execution stops in <tt>PassManager::run</tt>.</li>
				1765
				1766	</ul>
				1767
				1768	<p>Hopefully these tips will help with common case debugging situations. If
				1769	you'd like to contribute some tips of your own, just contact <a
				1770	href="mailto:sabre@nondot.org">Chris</a>.</p>
				1771
				1772	</div>
				1773
				1774	<!-- *********************************************************************** -->
				1775	<div class="doc_section">
				1776	<a name="future">Future extensions planned</a>
				1777	</div>
				1778	<!-- *********************************************************************** -->
				1779
				1780	<div class="doc_text">
				1781
				1782	<p>Although the LLVM Pass Infrastructure is very capable as it stands, and does
				1783	some nifty stuff, there are things we'd like to add in the future. Here is
				1784	where we are going:</p>
				1785
				1786	</div>
				1787
				1788	<!-- _______________________________________________________________________ -->
				1789	<div class="doc_subsubsection">
				1790	<a name="SMP">Multithreaded LLVM</a>
				1791	</div>
				1792
				1793	<div class="doc_text">
				1794
				1795	<p>Multiple CPU machines are becoming more common and compilation can never be
				1796	fast enough: obviously we should allow for a multithreaded compiler. Because of
				1797	the semantics defined for passes above (specifically they cannot maintain state
				1798	across invocations of their <tt>run*</tt> methods), a nice clean way to
				1799	implement a multithreaded compiler would be for the <tt>PassManager</tt> class
				1800	to create multiple instances of each pass object, and allow the separate
				1801	instances to be hacking on different parts of the program at the same time.</p>
				1802
				1803	<p>This implementation would prevent each of the passes from having to implement
				1804	multithreaded constructs, requiring only the LLVM core to have locking in a few
				1805	places (for global resources). Although this is a simple extension, we simply
				1806	haven't had time (or multiprocessor machines, thus a reason) to implement this.
				1807	Despite that, we have kept the LLVM passes SMP ready, and you should too.</p>
				1808
				1809	</div>
				1810
				1811	<!-- *********************************************************************** -->
				1812	<hr>
				1813	<address>
				1814	<a href="http://jigsaw.w3.org/css-validator/check/referer"><img
				1815	src="http://jigsaw.w3.org/css-validator/images/vcss" alt="Valid CSS!"></a>
				1816	<a href="http://validator.w3.org/check/referer"><img
				1817	src="http://www.w3.org/Icons/valid-html401" alt="Valid HTML 4.01!" /></a>
				1818
				1819	<a href="mailto:sabre@nondot.org">Chris Lattner</a><br>
				1820	<a href="http://llvm.org">The LLVM Compiler Infrastructure</a><br>
				1821	Last modified: $Date$
				1822	</address>
				1823
				1824	</body>
				1825	</html>