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

blob: 5e931d77a8930fb8f8be5147af3e26c763b10637 [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	<title>LLVM Alias Analysis Infrastructure</title>
				6	<link rel="stylesheet" href="llvm.css" type="text/css">
				7	</head>
				8	<body>
				9
				10	<div class="doc_title">
				11	LLVM Alias Analysis Infrastructure
				12	</div>
				13
				14	<ol>
				15	<li><a href="#introduction">Introduction</a></li>
				16
				17	<li><a href="#overview"><tt>AliasAnalysis</tt> Class Overview</a>
				18	<ul>
				19	<li><a href="#pointers">Representation of Pointers</a></li>
				20	<li><a href="#alias">The <tt>alias</tt> method</a></li>
				21	<li><a href="#ModRefInfo">The <tt>getModRefInfo</tt> methods</a></li>
				22	<li><a href="#OtherItfs">Other useful <tt>AliasAnalysis</tt> methods</a></li>
				23	</ul>
				24	</li>
				25
				26	<li><a href="#writingnew">Writing a new <tt>AliasAnalysis</tt> Implementation</a>
				27	<ul>
				28	<li><a href="#passsubclasses">Different Pass styles</a></li>
				29	<li><a href="#requiredcalls">Required initialization calls</a></li>
				30	<li><a href="#interfaces">Interfaces which may be specified</a></li>
				31	<li><a href="#chaining"><tt>AliasAnalysis</tt> chaining behavior</a></li>
				32	<li><a href="#updating">Updating analysis results for transformations</a></li>
				33	<li><a href="#implefficiency">Efficiency Issues</a></li>
Dan Gohman	aa78544	2010-06-24 19:34:03 +0000	[diff] [blame]	34	<li><a href="#passmanager">Pass Manager Issues</a></li>
Dan Gohman	f17a25c	2007-07-18 16:29:46 +0000	[diff] [blame]	35	</ul>
				36	</li>
				37
				38	<li><a href="#using">Using alias analysis results</a>
				39	<ul>
Chris Lattner	72da576	2009-04-25 21:11:37 +0000	[diff] [blame]	40	<li><a href="#memdep">Using the <tt>MemoryDependenceAnalysis</tt> Pass</a></li>
Dan Gohman	f17a25c	2007-07-18 16:29:46 +0000	[diff] [blame]	41	<li><a href="#ast">Using the <tt>AliasSetTracker</tt> class</a></li>
				42	<li><a href="#direct">Using the <tt>AliasAnalysis</tt> interface directly</a></li>
				43	</ul>
				44	</li>
				45
				46	<li><a href="#exist">Existing alias analysis implementations and clients</a>
				47	<ul>
				48	<li><a href="#impls">Available <tt>AliasAnalysis</tt> implementations</a></li>
				49	<li><a href="#aliasanalysis-xforms">Alias analysis driven transformations</a></li>
				50	<li><a href="#aliasanalysis-debug">Clients for debugging and evaluation of
				51	implementations</a></li>
				52	</ul>
				53	</li>
Owen Anderson	4f10b31	2007-10-02 00:44:20 +0000	[diff] [blame]	54	<li><a href="#memdep">Memory Dependence Analysis</a></li>
Dan Gohman	f17a25c	2007-07-18 16:29:46 +0000	[diff] [blame]	55	</ol>
				56
				57	<div class="doc_author">
				58	<p>Written by <a href="mailto:sabre@nondot.org">Chris Lattner</a></p>
				59	</div>
				60
				61	<!-- *********************************************************************** -->
				62	<div class="doc_section">
				63	<a name="introduction">Introduction</a>
				64	</div>
				65	<!-- *********************************************************************** -->
				66
				67	<div class="doc_text">
				68
				69	<p>Alias Analysis (aka Pointer Analysis) is a class of techniques which attempt
				70	to determine whether or not two pointers ever can point to the same object in
				71	memory. There are many different algorithms for alias analysis and many
				72	different ways of classifying them: flow-sensitive vs flow-insensitive,
				73	context-sensitive vs context-insensitive, field-sensitive vs field-insensitive,
				74	unification-based vs subset-based, etc. Traditionally, alias analyses respond
				75	to a query with a <a href="#MustMayNo">Must, May, or No</a> alias response,
				76	indicating that two pointers always point to the same object, might point to the
				77	same object, or are known to never point to the same object.</p>
				78
				79	<p>The LLVM <a
				80	href="http://llvm.org/doxygen/classllvm_1_1AliasAnalysis.html"><tt>AliasAnalysis</tt></a>
				81	class is the primary interface used by clients and implementations of alias
				82	analyses in the LLVM system. This class is the common interface between clients
				83	of alias analysis information and the implementations providing it, and is
				84	designed to support a wide range of implementations and clients (but currently
				85	all clients are assumed to be flow-insensitive). In addition to simple alias
				86	analysis information, this class exposes Mod/Ref information from those
				87	implementations which can provide it, allowing for powerful analyses and
				88	transformations to work well together.</p>
				89
				90	<p>This document contains information necessary to successfully implement this
				91	interface, use it, and to test both sides. It also explains some of the finer
				92	points about what exactly results mean. If you feel that something is unclear
				93	or should be added, please <a href="mailto:sabre@nondot.org">let me
				94	know</a>.</p>
				95
				96	</div>
				97
				98	<!-- *********************************************************************** -->
				99	<div class="doc_section">
				100	<a name="overview"><tt>AliasAnalysis</tt> Class Overview</a>
				101	</div>
				102	<!-- *********************************************************************** -->
				103
				104	<div class="doc_text">
				105
				106	<p>The <a
				107	href="http://llvm.org/doxygen/classllvm_1_1AliasAnalysis.html"><tt>AliasAnalysis</tt></a>
				108	class defines the interface that the various alias analysis implementations
				109	should support. This class exports two important enums: <tt>AliasResult</tt>
				110	and <tt>ModRefResult</tt> which represent the result of an alias query or a
				111	mod/ref query, respectively.</p>
				112
				113	<p>The <tt>AliasAnalysis</tt> interface exposes information about memory,
				114	represented in several different ways. In particular, memory objects are
				115	represented as a starting address and size, and function calls are represented
				116	as the actual <tt>call</tt> or <tt>invoke</tt> instructions that performs the
				117	call. The <tt>AliasAnalysis</tt> interface also exposes some helper methods
				118	which allow you to get mod/ref information for arbitrary instructions.</p>
				119
				120	</div>
				121
				122	<!-- ======================================================================= -->
				123	<div class="doc_subsection">
				124	<a name="pointers">Representation of Pointers</a>
				125	</div>
				126
				127	<div class="doc_text">
				128
				129	<p>Most importantly, the <tt>AliasAnalysis</tt> class provides several methods
				130	which are used to query whether or not two memory objects alias, whether
				131	function calls can modify or read a memory object, etc. For all of these
				132	queries, memory objects are represented as a pair of their starting address (a
				133	symbolic LLVM <tt>Value*</tt>) and a static size.</p>
				134
				135	<p>Representing memory objects as a starting address and a size is critically
				136	important for correct Alias Analyses. For example, consider this (silly, but
				137	possible) C code:</p>
				138
				139	<div class="doc_code">
				140	<pre>
				141	int i;
				142	char C[2];
				143	char A[10];
				144	/* ... */
				145	for (i = 0; i != 10; ++i) {
				146	C[0] = A[i]; /* One byte store */
				147	C[1] = A[9-i]; /* One byte store */
				148	}
				149	</pre>
				150	</div>
				151
				152	<p>In this case, the <tt>basicaa</tt> pass will disambiguate the stores to
				153	<tt>C[0]</tt> and <tt>C[1]</tt> because they are accesses to two distinct
				154	locations one byte apart, and the accesses are each one byte. In this case, the
				155	LICM pass can use store motion to remove the stores from the loop. In
				156	constrast, the following code:</p>
				157
				158	<div class="doc_code">
				159	<pre>
				160	int i;
				161	char C[2];
				162	char A[10];
				163	/* ... */
				164	for (i = 0; i != 10; ++i) {
				165	((short)C)[0] = A[i]; / Two byte store! */
				166	C[1] = A[9-i]; /* One byte store */
				167	}
				168	</pre>
				169	</div>
				170
				171	<p>In this case, the two stores to C do alias each other, because the access to
				172	the <tt>&C[0]</tt> element is a two byte access. If size information wasn't
				173	available in the query, even the first case would have to conservatively assume
				174	that the accesses alias.</p>
				175
				176	</div>
				177
				178	<!-- ======================================================================= -->
				179	<div class="doc_subsection">
				180	<a name="alias">The <tt>alias</tt> method</a>
				181	</div>
				182
				183	<div class="doc_text">
				184	The <tt>alias</tt> method is the primary interface used to determine whether or
				185	not two memory objects alias each other. It takes two memory objects as input
				186	and returns MustAlias, MayAlias, or NoAlias as appropriate.
				187	</div>
				188
				189	<!-- _______________________________________________________________________ -->
				190	<div class="doc_subsubsection">
				191	<a name="MustMayNo">Must, May, and No Alias Responses</a>
				192	</div>
				193
				194	<div class="doc_text">
Dan Gohman	c820844	2010-07-02 18:41:32 +0000	[diff] [blame^]	195	<p>The NoAlias response may be used when there is never an immediate dependence
				196	between any memory reference <i>based</i> on one pointer and any memory
				197	reference <i>based</i> the other. The most obvious example is when the two
				198	pointers point to non-overlapping memory ranges. Another is when the two
				199	pointers are only ever used for reading memory. Another is when the memory is
				200	freed and reallocated between accesses through one pointer and accesses through
				201	the other -- in this case, there is a dependence, but it's mediated by the free
				202	and reallocation.</p>
Dan Gohman	f17a25c	2007-07-18 16:29:46 +0000	[diff] [blame]	203
Nick Lewycky	74c8756	2008-12-14 21:08:48 +0000	[diff] [blame]	204	<p>The MayAlias response is used whenever the two pointers might refer to the
				205	same object. If the two memory objects overlap, but do not start at the same
				206	location, return MayAlias.</p>
Dan Gohman	f17a25c	2007-07-18 16:29:46 +0000	[diff] [blame]	207
Nick Lewycky	74c8756	2008-12-14 21:08:48 +0000	[diff] [blame]	208	<p>The MustAlias response may only be returned if the two memory objects are
				209	guaranteed to always start at exactly the same location. A MustAlias response
				210	implies that the pointers compare equal.</p>
Dan Gohman	f17a25c	2007-07-18 16:29:46 +0000	[diff] [blame]	211
				212	</div>
				213
				214	<!-- ======================================================================= -->
				215	<div class="doc_subsection">
				216	<a name="ModRefInfo">The <tt>getModRefInfo</tt> methods</a>
				217	</div>
				218
				219	<div class="doc_text">
				220
				221	<p>The <tt>getModRefInfo</tt> methods return information about whether the
				222	execution of an instruction can read or modify a memory location. Mod/Ref
				223	information is always conservative: if an instruction <b>might</b> read or write
				224	a location, ModRef is returned.</p>
				225
				226	<p>The <tt>AliasAnalysis</tt> class also provides a <tt>getModRefInfo</tt>
				227	method for testing dependencies between function calls. This method takes two
				228	call sites (CS1 & CS2), returns NoModRef if the two calls refer to disjoint
				229	memory locations, Ref if CS1 reads memory written by CS2, Mod if CS1 writes to
				230	memory read or written by CS2, or ModRef if CS1 might read or write memory
Chris Lattner	3b80375	2009-11-22 16:01:44 +0000	[diff] [blame]	231	accessed by CS2. Note that this relation is not commutative.</p>
Dan Gohman	f17a25c	2007-07-18 16:29:46 +0000	[diff] [blame]	232
				233	</div>
				234
				235
				236	<!-- ======================================================================= -->
				237	<div class="doc_subsection">
				238	<a name="OtherItfs">Other useful <tt>AliasAnalysis</tt> methods</a>
				239	</div>
				240
				241	<div class="doc_text">
				242
				243	<p>
				244	Several other tidbits of information are often collected by various alias
				245	analysis implementations and can be put to good use by various clients.
				246	</p>
				247
				248	</div>
				249
				250	<!-- _______________________________________________________________________ -->
				251	<div class="doc_subsubsection">
Dan Gohman	f17a25c	2007-07-18 16:29:46 +0000	[diff] [blame]	252	The <tt>pointsToConstantMemory</tt> method
				253	</div>
				254
				255	<div class="doc_text">
				256
				257	<p>The <tt>pointsToConstantMemory</tt> method returns true if and only if the
				258	analysis can prove that the pointer only points to unchanging memory locations
				259	(functions, constant global variables, and the null pointer). This information
				260	can be used to refine mod/ref information: it is impossible for an unchanging
				261	memory location to be modified.</p>
				262
				263	</div>
				264
				265	<!-- _______________________________________________________________________ -->
				266	<div class="doc_subsubsection">
				267	<a name="simplemodref">The <tt>doesNotAccessMemory</tt> and
				268	<tt>onlyReadsMemory</tt> methods</a>
				269	</div>
				270
				271	<div class="doc_text">
				272
				273	<p>These methods are used to provide very simple mod/ref information for
				274	function calls. The <tt>doesNotAccessMemory</tt> method returns true for a
				275	function if the analysis can prove that the function never reads or writes to
				276	memory, or if the function only reads from constant memory. Functions with this
				277	property are side-effect free and only depend on their input arguments, allowing
				278	them to be eliminated if they form common subexpressions or be hoisted out of
				279	loops. Many common functions behave this way (e.g., <tt>sin</tt> and
				280	<tt>cos</tt>) but many others do not (e.g., <tt>acos</tt>, which modifies the
				281	<tt>errno</tt> variable).</p>
				282
				283	<p>The <tt>onlyReadsMemory</tt> method returns true for a function if analysis
				284	can prove that (at most) the function only reads from non-volatile memory.
				285	Functions with this property are side-effect free, only depending on their input
				286	arguments and the state of memory when they are called. This property allows
				287	calls to these functions to be eliminated and moved around, as long as there is
				288	no store instruction that changes the contents of memory. Note that all
				289	functions that satisfy the <tt>doesNotAccessMemory</tt> method also satisfies
				290	<tt>onlyReadsMemory</tt>.</p>
				291
				292	</div>
				293
				294	<!-- *********************************************************************** -->
				295	<div class="doc_section">
				296	<a name="writingnew">Writing a new <tt>AliasAnalysis</tt> Implementation</a>
				297	</div>
				298	<!-- *********************************************************************** -->
				299
				300	<div class="doc_text">
				301
				302	<p>Writing a new alias analysis implementation for LLVM is quite
				303	straight-forward. There are already several implementations that you can use
				304	for examples, and the following information should help fill in any details.
				305	For a examples, take a look at the <a href="#impls">various alias analysis
				306	implementations</a> included with LLVM.</p>
				307
				308	</div>
				309
				310	<!-- ======================================================================= -->
				311	<div class="doc_subsection">
				312	<a name="passsubclasses">Different Pass styles</a>
				313	</div>
				314
				315	<div class="doc_text">
				316
				317	<p>The first step to determining what type of <a
				318	href="WritingAnLLVMPass.html">LLVM pass</a> you need to use for your Alias
				319	Analysis. As is the case with most other analyses and transformations, the
				320	answer should be fairly obvious from what type of problem you are trying to
				321	solve:</p>
				322
				323	<ol>
				324	<li>If you require interprocedural analysis, it should be a
				325	<tt>Pass</tt>.</li>
				326	<li>If you are a function-local analysis, subclass <tt>FunctionPass</tt>.</li>
				327	<li>If you don't need to look at the program at all, subclass
				328	<tt>ImmutablePass</tt>.</li>
				329	</ol>
				330
				331	<p>In addition to the pass that you subclass, you should also inherit from the
				332	<tt>AliasAnalysis</tt> interface, of course, and use the
				333	<tt>RegisterAnalysisGroup</tt> template to register as an implementation of
				334	<tt>AliasAnalysis</tt>.</p>
				335
				336	</div>
				337
				338	<!-- ======================================================================= -->
				339	<div class="doc_subsection">
				340	<a name="requiredcalls">Required initialization calls</a>
				341	</div>
				342
				343	<div class="doc_text">
				344
				345	<p>Your subclass of <tt>AliasAnalysis</tt> is required to invoke two methods on
				346	the <tt>AliasAnalysis</tt> base class: <tt>getAnalysisUsage</tt> and
				347	<tt>InitializeAliasAnalysis</tt>. In particular, your implementation of
				348	<tt>getAnalysisUsage</tt> should explicitly call into the
				349	<tt>AliasAnalysis::getAnalysisUsage</tt> method in addition to doing any
				350	declaring any pass dependencies your pass has. Thus you should have something
				351	like this:</p>
				352
				353	<div class="doc_code">
				354	<pre>
				355	void getAnalysisUsage(AnalysisUsage &AU) const {
				356	AliasAnalysis::getAnalysisUsage(AU);
				357	<i>// declare your dependencies here.</i>
				358	}
				359	</pre>
				360	</div>
				361
				362	<p>Additionally, your must invoke the <tt>InitializeAliasAnalysis</tt> method
				363	from your analysis run method (<tt>run</tt> for a <tt>Pass</tt>,
				364	<tt>runOnFunction</tt> for a <tt>FunctionPass</tt>, or <tt>InitializePass</tt>
				365	for an <tt>ImmutablePass</tt>). For example (as part of a <tt>Pass</tt>):</p>
				366
				367	<div class="doc_code">
				368	<pre>
				369	bool run(Module &M) {
				370	InitializeAliasAnalysis(this);
				371	<i>// Perform analysis here...</i>
				372	return false;
				373	}
				374	</pre>
				375	</div>
				376
				377	</div>
				378
				379	<!-- ======================================================================= -->
				380	<div class="doc_subsection">
				381	<a name="interfaces">Interfaces which may be specified</a>
				382	</div>
				383
				384	<div class="doc_text">
				385
				386	<p>All of the <a
				387	href="/doxygen/classllvm_1_1AliasAnalysis.html"><tt>AliasAnalysis</tt></a>
				388	virtual methods default to providing <a href="#chaining">chaining</a> to another
				389	alias analysis implementation, which ends up returning conservatively correct
				390	information (returning "May" Alias and "Mod/Ref" for alias and mod/ref queries
				391	respectively). Depending on the capabilities of the analysis you are
				392	implementing, you just override the interfaces you can improve.</p>
				393
				394	</div>
				395
				396
				397
				398	<!-- ======================================================================= -->
				399	<div class="doc_subsection">
				400	<a name="chaining"><tt>AliasAnalysis</tt> chaining behavior</a>
				401	</div>
				402
				403	<div class="doc_text">
				404
				405	<p>With only two special exceptions (the <tt><a
				406	href="#basic-aa">basicaa</a></tt> and <a href="#no-aa"><tt>no-aa</tt></a>
				407	passes) every alias analysis pass chains to another alias analysis
				408	implementation (for example, the user can specify "<tt>-basicaa -ds-aa
Chris Lattner	5dd9205	2010-03-01 19:24:17 +0000	[diff] [blame]	409	-licm</tt>" to get the maximum benefit from both alias
Dan Gohman	f17a25c	2007-07-18 16:29:46 +0000	[diff] [blame]	410	analyses). The alias analysis class automatically takes care of most of this
				411	for methods that you don't override. For methods that you do override, in code
				412	paths that return a conservative MayAlias or Mod/Ref result, simply return
				413	whatever the superclass computes. For example:</p>
				414
				415	<div class="doc_code">
				416	<pre>
				417	AliasAnalysis::AliasResult alias(const Value *V1, unsigned V1Size,
				418	const Value *V2, unsigned V2Size) {
				419	if (...)
				420	return NoAlias;
				421	...
				422
				423	<i>// Couldn't determine a must or no-alias result.</i>
				424	return AliasAnalysis::alias(V1, V1Size, V2, V2Size);
				425	}
				426	</pre>
				427	</div>
				428
				429	<p>In addition to analysis queries, you must make sure to unconditionally pass
				430	LLVM <a href="#updating">update notification</a> methods to the superclass as
				431	well if you override them, which allows all alias analyses in a change to be
				432	updated.</p>
				433
				434	</div>
				435
				436
				437	<!-- ======================================================================= -->
				438	<div class="doc_subsection">
				439	<a name="updating">Updating analysis results for transformations</a>
				440	</div>
				441
				442	<div class="doc_text">
				443	<p>
				444	Alias analysis information is initially computed for a static snapshot of the
				445	program, but clients will use this information to make transformations to the
				446	code. All but the most trivial forms of alias analysis will need to have their
				447	analysis results updated to reflect the changes made by these transformations.
				448	</p>
				449
				450	<p>
				451	The <tt>AliasAnalysis</tt> interface exposes two methods which are used to
				452	communicate program changes from the clients to the analysis implementations.
				453	Various alias analysis implementations should use these methods to ensure that
				454	their internal data structures are kept up-to-date as the program changes (for
				455	example, when an instruction is deleted), and clients of alias analysis must be
				456	sure to call these interfaces appropriately.
				457	</p>
				458	</div>
				459
				460	<!-- _______________________________________________________________________ -->
				461	<div class="doc_subsubsection">The <tt>deleteValue</tt> method</div>
				462
				463	<div class="doc_text">
				464	The <tt>deleteValue</tt> method is called by transformations when they remove an
				465	instruction or any other value from the program (including values that do not
				466	use pointers). Typically alias analyses keep data structures that have entries
				467	for each value in the program. When this method is called, they should remove
				468	any entries for the specified value, if they exist.
				469	</div>
				470
				471	<!-- _______________________________________________________________________ -->
				472	<div class="doc_subsubsection">The <tt>copyValue</tt> method</div>
				473
				474	<div class="doc_text">
				475	The <tt>copyValue</tt> method is used when a new value is introduced into the
				476	program. There is no way to introduce a value into the program that did not
				477	exist before (this doesn't make sense for a safe compiler transformation), so
				478	this is the only way to introduce a new value. This method indicates that the
				479	new value has exactly the same properties as the value being copied.
				480	</div>
				481
				482	<!-- _______________________________________________________________________ -->
				483	<div class="doc_subsubsection">The <tt>replaceWithNewValue</tt> method</div>
				484
				485	<div class="doc_text">
				486	This method is a simple helper method that is provided to make clients easier to
				487	use. It is implemented by copying the old analysis information to the new
				488	value, then deleting the old value. This method cannot be overridden by alias
				489	analysis implementations.
				490	</div>
				491
				492	<!-- ======================================================================= -->
				493	<div class="doc_subsection">
				494	<a name="implefficiency">Efficiency Issues</a>
				495	</div>
				496
				497	<div class="doc_text">
				498
				499	<p>From the LLVM perspective, the only thing you need to do to provide an
				500	efficient alias analysis is to make sure that alias analysis <b>queries</b> are
				501	serviced quickly. The actual calculation of the alias analysis results (the
				502	"run" method) is only performed once, but many (perhaps duplicate) queries may
				503	be performed. Because of this, try to move as much computation to the run
				504	method as possible (within reason).</p>
				505
				506	</div>
				507
Dan Gohman	aa78544	2010-06-24 19:34:03 +0000	[diff] [blame]	508	<!-- ======================================================================= -->
				509	<div class="doc_subsection">
				510	<a name="passmanager">Pass Manager Issues</a>
				511	</div>
				512
				513	<div class="doc_text">
				514
				515	<p>PassManager support for alternative AliasAnalysis implementation
				516	has some issues.</p>
				517
				518	<p>There is no way to override the default alias analysis. It would
				519	be very useful to be able to do something like "opt -my-aa -O2" and
				520	have it use -my-aa for all passes which need AliasAnalysis, but there
				521	is currently no support for that, short of changing the source code
				522	and recompiling. Similarly, there is also no way of setting a chain
				523	of analyses as the default.</p>
				524
				525	<p>There is no way for transform passes to declare that they preserve
				526	<tt>AliasAnalysis</tt> implementations. The <tt>AliasAnalysis</tt>
				527	interface includes <tt>deleteValue</tt> and <tt>copyValue</tt> methods
				528	which are intended to allow a pass to keep an AliasAnalysis consistent,
				529	however there's no way for a pass to declare in its
				530	<tt>getAnalysisUsage</tt> that it does so. Some passes attempt to use
				531	<tt>AU.addPreserved<AliasAnalysis></tt>, however this doesn't
				532	actually have any effect.</tt>
				533
				534	<p><tt>AliasAnalysisCounter</tt> (<tt>-count-aa</tt>) and <tt>AliasDebugger</tt>
				535	(<tt>-debug-aa</tt>) are implemented as <tt>ModulePass</tt> classes, so if your
				536	alias analysis uses <tt>FunctionPass</tt>, it won't be able to use
				537	these utilities. If you try to use them, the pass manager will
				538	silently route alias analysis queries directly to
				539	<tt>BasicAliasAnalysis</tt> instead.</p>
				540
				541	<p>Similarly, the <tt>opt -p</tt> option introduces <tt>ModulePass</tt>
				542	passes between each pass, which prevents the use of <tt>FunctionPass</tt>
				543	alias analysis passes.</p>
				544
				545	</div>
				546
Dan Gohman	f17a25c	2007-07-18 16:29:46 +0000	[diff] [blame]	547	<!-- *********************************************************************** -->
				548	<div class="doc_section">
				549	<a name="using">Using alias analysis results</a>
				550	</div>
				551	<!-- *********************************************************************** -->
				552
				553	<div class="doc_text">
				554
				555	<p>There are several different ways to use alias analysis results. In order of
				556	preference, these are...</p>
				557
				558	</div>
				559
				560	<!-- ======================================================================= -->
				561	<div class="doc_subsection">
Chris Lattner	72da576	2009-04-25 21:11:37 +0000	[diff] [blame]	562	<a name="memdep">Using the <tt>MemoryDependenceAnalysis</tt> Pass</a>
Dan Gohman	f17a25c	2007-07-18 16:29:46 +0000	[diff] [blame]	563	</div>
				564
				565	<div class="doc_text">
				566
Chris Lattner	72da576	2009-04-25 21:11:37 +0000	[diff] [blame]	567	<p>The <tt>memdep</tt> pass uses alias analysis to provide high-level dependence
				568	information about memory-using instructions. This will tell you which store
				569	feeds into a load, for example. It uses caching and other techniques to be
				570	efficient, and is used by Dead Store Elimination, GVN, and memcpy optimizations.
				571	</p>
Dan Gohman	f17a25c	2007-07-18 16:29:46 +0000	[diff] [blame]	572
				573	</div>
				574
				575	<!-- ======================================================================= -->
				576	<div class="doc_subsection">
				577	<a name="ast">Using the <tt>AliasSetTracker</tt> class</a>
				578	</div>
				579
				580	<div class="doc_text">
				581
				582	<p>Many transformations need information about alias <b>sets</b> that are active
				583	in some scope, rather than information about pairwise aliasing. The <tt><a
				584	href="/doxygen/classllvm_1_1AliasSetTracker.html">AliasSetTracker</a></tt> class
				585	is used to efficiently build these Alias Sets from the pairwise alias analysis
				586	information provided by the <tt>AliasAnalysis</tt> interface.</p>
				587
				588	<p>First you initialize the AliasSetTracker by using the "<tt>add</tt>" methods
				589	to add information about various potentially aliasing instructions in the scope
				590	you are interested in. Once all of the alias sets are completed, your pass
				591	should simply iterate through the constructed alias sets, using the
				592	<tt>AliasSetTracker</tt> <tt>begin()</tt>/<tt>end()</tt> methods.</p>
				593
				594	<p>The <tt>AliasSet</tt>s formed by the <tt>AliasSetTracker</tt> are guaranteed
				595	to be disjoint, calculate mod/ref information and volatility for the set, and
				596	keep track of whether or not all of the pointers in the set are Must aliases.
				597	The AliasSetTracker also makes sure that sets are properly folded due to call
				598	instructions, and can provide a list of pointers in each set.</p>
				599
				600	<p>As an example user of this, the <a href="/doxygen/structLICM.html">Loop
				601	Invariant Code Motion</a> pass uses <tt>AliasSetTracker</tt>s to calculate alias
				602	sets for each loop nest. If an <tt>AliasSet</tt> in a loop is not modified,
				603	then all load instructions from that set may be hoisted out of the loop. If any
				604	alias sets are stored to <b>and</b> are must alias sets, then the stores may be
				605	sunk to outside of the loop, promoting the memory location to a register for the
				606	duration of the loop nest. Both of these transformations only apply if the
				607	pointer argument is loop-invariant.</p>
				608
				609	</div>
				610
				611	<!-- _______________________________________________________________________ -->
				612	<div class="doc_subsubsection">
				613	The AliasSetTracker implementation
				614	</div>
				615
				616	<div class="doc_text">
				617
				618	<p>The AliasSetTracker class is implemented to be as efficient as possible. It
				619	uses the union-find algorithm to efficiently merge AliasSets when a pointer is
				620	inserted into the AliasSetTracker that aliases multiple sets. The primary data
				621	structure is a hash table mapping pointers to the AliasSet they are in.</p>
				622
				623	<p>The AliasSetTracker class must maintain a list of all of the LLVM Value*'s
				624	that are in each AliasSet. Since the hash table already has entries for each
				625	LLVM Value* of interest, the AliasesSets thread the linked list through these
				626	hash-table nodes to avoid having to allocate memory unnecessarily, and to make
				627	merging alias sets extremely efficient (the linked list merge is constant time).
				628	</p>
				629
				630	<p>You shouldn't need to understand these details if you are just a client of
				631	the AliasSetTracker, but if you look at the code, hopefully this brief
				632	description will help make sense of why things are designed the way they
				633	are.</p>
				634
				635	</div>
				636
				637	<!-- ======================================================================= -->
				638	<div class="doc_subsection">
				639	<a name="direct">Using the <tt>AliasAnalysis</tt> interface directly</a>
				640	</div>
				641
				642	<div class="doc_text">
				643
				644	<p>If neither of these utility class are what your pass needs, you should use
				645	the interfaces exposed by the <tt>AliasAnalysis</tt> class directly. Try to use
				646	the higher-level methods when possible (e.g., use mod/ref information instead of
				647	the <a href="#alias"><tt>alias</tt></a> method directly if possible) to get the
				648	best precision and efficiency.</p>
				649
				650	</div>
				651
				652	<!-- *********************************************************************** -->
				653	<div class="doc_section">
				654	<a name="exist">Existing alias analysis implementations and clients</a>
				655	</div>
				656	<!-- *********************************************************************** -->
				657
				658	<div class="doc_text">
				659
				660	<p>If you're going to be working with the LLVM alias analysis infrastructure,
				661	you should know what clients and implementations of alias analysis are
				662	available. In particular, if you are implementing an alias analysis, you should
				663	be aware of the <a href="#aliasanalysis-debug">the clients</a> that are useful
				664	for monitoring and evaluating different implementations.</p>
				665
				666	</div>
				667
				668	<!-- ======================================================================= -->
				669	<div class="doc_subsection">
				670	<a name="impls">Available <tt>AliasAnalysis</tt> implementations</a>
				671	</div>
				672
				673	<div class="doc_text">
				674
				675	<p>This section lists the various implementations of the <tt>AliasAnalysis</tt>
				676	interface. With the exception of the <a href="#no-aa"><tt>-no-aa</tt></a> and
				677	<a href="#basic-aa"><tt>-basicaa</tt></a> implementations, all of these <a
				678	href="#chaining">chain</a> to other alias analysis implementations.</p>
				679
				680	</div>
				681
				682	<!-- _______________________________________________________________________ -->
				683	<div class="doc_subsubsection">
				684	<a name="no-aa">The <tt>-no-aa</tt> pass</a>
				685	</div>
				686
				687	<div class="doc_text">
				688
				689	<p>The <tt>-no-aa</tt> pass is just like what it sounds: an alias analysis that
				690	never returns any useful information. This pass can be useful if you think that
				691	alias analysis is doing something wrong and are trying to narrow down a
				692	problem.</p>
				693
				694	</div>
				695
				696	<!-- _______________________________________________________________________ -->
				697	<div class="doc_subsubsection">
				698	<a name="basic-aa">The <tt>-basicaa</tt> pass</a>
				699	</div>
				700
				701	<div class="doc_text">
				702
				703	<p>The <tt>-basicaa</tt> pass is the default LLVM alias analysis. It is an
				704	aggressive local analysis that "knows" many important facts:</p>
				705
				706	<ul>
				707	<li>Distinct globals, stack allocations, and heap allocations can never
				708	alias.</li>
				709	<li>Globals, stack allocations, and heap allocations never alias the null
				710	pointer.</li>
				711	<li>Different fields of a structure do not alias.</li>
				712	<li>Indexes into arrays with statically differing subscripts cannot alias.</li>
				713	<li>Many common standard C library functions <a
				714	href="#simplemodref">never access memory or only read memory</a>.</li>
				715	<li>Pointers that obviously point to constant globals
				716	"<tt>pointToConstantMemory</tt>".</li>
				717	<li>Function calls can not modify or references stack allocations if they never
				718	escape from the function that allocates them (a common case for automatic
				719	arrays).</li>
				720	</ul>
				721
				722	</div>
				723
				724	<!-- _______________________________________________________________________ -->
				725	<div class="doc_subsubsection">
				726	<a name="globalsmodref">The <tt>-globalsmodref-aa</tt> pass</a>
				727	</div>
				728
				729	<div class="doc_text">
				730
				731	<p>This pass implements a simple context-sensitive mod/ref and alias analysis
				732	for internal global variables that don't "have their address taken". If a
				733	global does not have its address taken, the pass knows that no pointers alias
				734	the global. This pass also keeps track of functions that it knows never access
Chris Lattner	72da576	2009-04-25 21:11:37 +0000	[diff] [blame]	735	memory or never read memory. This allows certain optimizations (e.g. GVN) to
Dan Gohman	f17a25c	2007-07-18 16:29:46 +0000	[diff] [blame]	736	eliminate call instructions entirely.
				737	</p>
				738
				739	<p>The real power of this pass is that it provides context-sensitive mod/ref
				740	information for call instructions. This allows the optimizer to know that
				741	calls to a function do not clobber or read the value of the global, allowing
				742	loads and stores to be eliminated.</p>
				743
				744	<p>Note that this pass is somewhat limited in its scope (only support
				745	non-address taken globals), but is very quick analysis.</p>
				746	</div>
				747
				748	<!-- _______________________________________________________________________ -->
				749	<div class="doc_subsubsection">
Dan Gohman	f17a25c	2007-07-18 16:29:46 +0000	[diff] [blame]	750	<a name="steens-aa">The <tt>-steens-aa</tt> pass</a>
				751	</div>
				752
				753	<div class="doc_text">
				754
				755	<p>The <tt>-steens-aa</tt> pass implements a variation on the well-known
				756	"Steensgaard's algorithm" for interprocedural alias analysis. Steensgaard's
				757	algorithm is a unification-based, flow-insensitive, context-insensitive, and
				758	field-insensitive alias analysis that is also very scalable (effectively linear
				759	time).</p>
				760
				761	<p>The LLVM <tt>-steens-aa</tt> pass implements a "speculatively
				762	field-<b>sensitive</b>" version of Steensgaard's algorithm using the Data
				763	Structure Analysis framework. This gives it substantially more precision than
				764	the standard algorithm while maintaining excellent analysis scalability.</p>
				765
				766	<p>Note that <tt>-steens-aa</tt> is available in the optional "poolalloc"
				767	module, it is not part of the LLVM core.</p>
				768
				769	</div>
				770
				771	<!-- _______________________________________________________________________ -->
				772	<div class="doc_subsubsection">
				773	<a name="ds-aa">The <tt>-ds-aa</tt> pass</a>
				774	</div>
				775
				776	<div class="doc_text">
				777
				778	<p>The <tt>-ds-aa</tt> pass implements the full Data Structure Analysis
				779	algorithm. Data Structure Analysis is a modular unification-based,
				780	flow-insensitive, context-<b>sensitive</b>, and speculatively
				781	field-<b>sensitive</b> alias analysis that is also quite scalable, usually at
				782	O(n*log(n)).</p>
				783
				784	<p>This algorithm is capable of responding to a full variety of alias analysis
				785	queries, and can provide context-sensitive mod/ref information as well. The
				786	only major facility not implemented so far is support for must-alias
				787	information.</p>
				788
				789	<p>Note that <tt>-ds-aa</tt> is available in the optional "poolalloc"
				790	module, it is not part of the LLVM core.</p>
				791
				792	</div>
				793
Dan Gohman	3e6ba36	2010-06-28 22:09:52 +0000	[diff] [blame]	794	<!-- _______________________________________________________________________ -->
				795	<div class="doc_subsubsection">
				796	<a name="scev-aa">The <tt>-scev-aa</tt> pass</a>
				797	</div>
				798
				799	<div class="doc_text">
				800
				801	<p>The <tt>-scev-aa</tt> pass implements AliasAnalysis queries by
				802	translating them into ScalarEvolution queries. This gives it a
				803	more complete understanding of <tt>getelementptr</tt> instructions
				804	and loop induction variables than other alias analyses have.</p>
				805
				806	</div>
Dan Gohman	f17a25c	2007-07-18 16:29:46 +0000	[diff] [blame]	807
				808	<!-- ======================================================================= -->
				809	<div class="doc_subsection">
				810	<a name="aliasanalysis-xforms">Alias analysis driven transformations</a>
				811	</div>
				812
				813	<div class="doc_text">
				814	LLVM includes several alias-analysis driven transformations which can be used
				815	with any of the implementations above.
				816	</div>
				817
				818	<!-- _______________________________________________________________________ -->
				819	<div class="doc_subsubsection">
				820	<a name="adce">The <tt>-adce</tt> pass</a>
				821	</div>
				822
				823	<div class="doc_text">
				824
				825	<p>The <tt>-adce</tt> pass, which implements Aggressive Dead Code Elimination
				826	uses the <tt>AliasAnalysis</tt> interface to delete calls to functions that do
				827	not have side-effects and are not used.</p>
				828
				829	</div>
				830
				831
				832	<!-- _______________________________________________________________________ -->
				833	<div class="doc_subsubsection">
				834	<a name="licm">The <tt>-licm</tt> pass</a>
				835	</div>
				836
				837	<div class="doc_text">
				838
				839	<p>The <tt>-licm</tt> pass implements various Loop Invariant Code Motion related
				840	transformations. It uses the <tt>AliasAnalysis</tt> interface for several
				841	different transformations:</p>
				842
				843	<ul>
				844	<li>It uses mod/ref information to hoist or sink load instructions out of loops
				845	if there are no instructions in the loop that modifies the memory loaded.</li>
				846
				847	<li>It uses mod/ref information to hoist function calls out of loops that do not
				848	write to memory and are loop-invariant.</li>
				849
				850	<li>If uses alias information to promote memory objects that are loaded and
				851	stored to in loops to live in a register instead. It can do this if there are
				852	no may aliases to the loaded/stored memory location.</li>
				853	</ul>
				854
				855	</div>
				856
				857	<!-- _______________________________________________________________________ -->
				858	<div class="doc_subsubsection">
				859	<a name="argpromotion">The <tt>-argpromotion</tt> pass</a>
				860	</div>
				861
				862	<div class="doc_text">
				863	<p>
				864	The <tt>-argpromotion</tt> pass promotes by-reference arguments to be passed in
				865	by-value instead. In particular, if pointer arguments are only loaded from it
				866	passes in the value loaded instead of the address to the function. This pass
				867	uses alias information to make sure that the value loaded from the argument
				868	pointer is not modified between the entry of the function and any load of the
				869	pointer.</p>
				870	</div>
				871
				872	<!-- _______________________________________________________________________ -->
				873	<div class="doc_subsubsection">
Chris Lattner	72da576	2009-04-25 21:11:37 +0000	[diff] [blame]	874	<a name="gvn">The <tt>-gvn</tt>, <tt>-memcpyopt</tt>, and <tt>-dse</tt>
				875	passes</a>
Dan Gohman	f17a25c	2007-07-18 16:29:46 +0000	[diff] [blame]	876	</div>
				877
				878	<div class="doc_text">
				879
Chris Lattner	72da576	2009-04-25 21:11:37 +0000	[diff] [blame]	880	<p>These passes use AliasAnalysis information to reason about loads and stores.
				881	</p>
Dan Gohman	f17a25c	2007-07-18 16:29:46 +0000	[diff] [blame]	882
				883	</div>
				884
				885	<!-- ======================================================================= -->
				886	<div class="doc_subsection">
				887	<a name="aliasanalysis-debug">Clients for debugging and evaluation of
				888	implementations</a>
				889	</div>
				890
				891	<div class="doc_text">
				892
				893	<p>These passes are useful for evaluating the various alias analysis
Chris Lattner	5dd9205	2010-03-01 19:24:17 +0000	[diff] [blame]	894	implementations. You can use them with commands like '<tt>opt -ds-aa
Dan Gohman	f17a25c	2007-07-18 16:29:46 +0000	[diff] [blame]	895	-aa-eval foo.bc -disable-output -stats</tt>'.</p>
				896
				897	</div>
				898
				899	<!-- _______________________________________________________________________ -->
				900	<div class="doc_subsubsection">
				901	<a name="print-alias-sets">The <tt>-print-alias-sets</tt> pass</a>
				902	</div>
				903
				904	<div class="doc_text">
				905
				906	<p>The <tt>-print-alias-sets</tt> pass is exposed as part of the
				907	<tt>opt</tt> tool to print out the Alias Sets formed by the <a
				908	href="#ast"><tt>AliasSetTracker</tt></a> class. This is useful if you're using
				909	the <tt>AliasSetTracker</tt> class. To use it, use something like:</p>
				910
				911	<div class="doc_code">
				912	<pre>
				913	% opt -ds-aa -print-alias-sets -disable-output
				914	</pre>
				915	</div>
				916
				917	</div>
				918
				919
				920	<!-- _______________________________________________________________________ -->
				921	<div class="doc_subsubsection">
				922	<a name="count-aa">The <tt>-count-aa</tt> pass</a>
				923	</div>
				924
				925	<div class="doc_text">
				926
				927	<p>The <tt>-count-aa</tt> pass is useful to see how many queries a particular
				928	pass is making and what responses are returned by the alias analysis. As an
				929	example,</p>
				930
				931	<div class="doc_code">
				932	<pre>
				933	% opt -basicaa -count-aa -ds-aa -count-aa -licm
				934	</pre>
				935	</div>
				936
				937	<p>will print out how many queries (and what responses are returned) by the
				938	<tt>-licm</tt> pass (of the <tt>-ds-aa</tt> pass) and how many queries are made
				939	of the <tt>-basicaa</tt> pass by the <tt>-ds-aa</tt> pass. This can be useful
				940	when debugging a transformation or an alias analysis implementation.</p>
				941
				942	</div>
				943
				944	<!-- _______________________________________________________________________ -->
				945	<div class="doc_subsubsection">
				946	<a name="aa-eval">The <tt>-aa-eval</tt> pass</a>
				947	</div>
				948
				949	<div class="doc_text">
				950
				951	<p>The <tt>-aa-eval</tt> pass simply iterates through all pairs of pointers in a
				952	function and asks an alias analysis whether or not the pointers alias. This
				953	gives an indication of the precision of the alias analysis. Statistics are
				954	printed indicating the percent of no/may/must aliases found (a more precise
				955	algorithm will have a lower number of may aliases).</p>
				956
				957	</div>
				958
				959	<!-- *********************************************************************** -->
Owen Anderson	05e080f	2007-10-02 00:43:25 +0000	[diff] [blame]	960	<div class="doc_section">
				961	<a name="memdep">Memory Dependence Analysis</a>
				962	</div>
				963	<!-- *********************************************************************** -->
				964
				965	<div class="doc_text">
				966
				967	<p>If you're just looking to be a client of alias analysis information, consider
				968	using the Memory Dependence Analysis interface instead. MemDep is a lazy,
				969	caching layer on top of alias analysis that is able to answer the question of
				970	what preceding memory operations a given instruction depends on, either at an
				971	intra- or inter-block level. Because of its laziness and caching
				972	policy, using MemDep can be a significant performance win over accessing alias
				973	analysis directly.</p>
				974
				975	</div>
				976
				977	<!-- *********************************************************************** -->
Dan Gohman	f17a25c	2007-07-18 16:29:46 +0000	[diff] [blame]	978
				979	<hr>
				980	<address>
				981	<a href="http://jigsaw.w3.org/css-validator/check/referer"><img
Misha Brukman	947321d	2008-12-11 17:34:48 +0000	[diff] [blame]	982	src="http://jigsaw.w3.org/css-validator/images/vcss-blue" alt="Valid CSS"></a>
Dan Gohman	f17a25c	2007-07-18 16:29:46 +0000	[diff] [blame]	983	<a href="http://validator.w3.org/check/referer"><img
Misha Brukman	947321d	2008-12-11 17:34:48 +0000	[diff] [blame]	984	src="http://www.w3.org/Icons/valid-html401-blue" alt="Valid HTML 4.01"></a>
Dan Gohman	f17a25c	2007-07-18 16:29:46 +0000	[diff] [blame]	985
				986	<a href="mailto:sabre@nondot.org">Chris Lattner</a><br>
				987	<a href="http://llvm.org">LLVM Compiler Infrastructure</a><br>
				988	Last modified: $Date$
				989	</address>
				990
				991	</body>
				992	</html>