Gitiles
Code Review Sign In
gerrit-public.fairphone.software / toolchain / llvm-project / b2d03a6d003a9255655321b649117a010b8efe7f / . / llvm / docs / AliasAnalysis.html
blob: d04e841d7a1a7adf134976f6d020f185a4df8d0d [file] [log] [blame]
Misha Brukman084a3a072003-10-24 18:06:11 +0000[diff] [blame]1<!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.01//EN"
2 "http://www.w3.org/TR/html4/strict.dtd">
Misha Brukman283ff692003-10-22 23:27:16 +0000[diff] [blame]3<html>
4<head>
NAKAMURA Takumib8004d92011-10-31 11:21:59 +0000[diff] [blame]5 <meta http-equiv="Content-Type" content="text/html; charset=utf-8">
Misha Brukman3eef5d42004-07-01 15:33:24 +0000[diff] [blame]6 <title>LLVM Alias Analysis Infrastructure</title>
Misha Brukmanc6ef5592004-01-15 19:04:12 +0000[diff] [blame]7 <link rel="stylesheet" href="llvm.css" type="text/css">
Misha Brukman283ff692003-10-22 23:27:16 +0000[diff] [blame]8</head>
Misha Brukman283ff692003-10-22 23:27:16 +0000[diff] [blame]9<body>
Chris Lattnerb512e9f2003-03-04 19:37:49 +0000[diff] [blame]10
NAKAMURA Takumifc8d9302011-04-18 23:59:50 +0000[diff] [blame]11<h1>
Misha Brukman3eef5d42004-07-01 15:33:24 +0000[diff] [blame]12 LLVM Alias Analysis Infrastructure
NAKAMURA Takumifc8d9302011-04-18 23:59:50 +0000[diff] [blame]13</h1>
Chris Lattnerb512e9f2003-03-04 19:37:49 +0000[diff] [blame]14
15<ol>
Misha Brukman9782b032003-10-23 02:29:42 +0000[diff] [blame]16 <li><a href="#introduction">Introduction</a></li>
Chris Lattnerb512e9f2003-03-04 19:37:49 +0000[diff] [blame]17
Chris Lattner2a19d832004-05-23 21:04:01 +0000[diff] [blame]18 <li><a href="#overview"><tt>AliasAnalysis</tt> Class Overview</a>
Chris Lattnerb512e9f2003-03-04 19:37:49 +0000[diff] [blame]19 <ul>
Misha Brukman9782b032003-10-23 02:29:42 +0000[diff] [blame]20 <li><a href="#pointers">Representation of Pointers</a></li>
Chris Lattner2a19d832004-05-23 21:04:01 +0000[diff] [blame]21 <li><a href="#alias">The <tt>alias</tt> method</a></li>
Misha Brukman9782b032003-10-23 02:29:42 +0000[diff] [blame]22 <li><a href="#ModRefInfo">The <tt>getModRefInfo</tt> methods</a></li>
Chris Lattner2a19d832004-05-23 21:04:01 +0000[diff] [blame]23 <li><a href="#OtherItfs">Other useful <tt>AliasAnalysis</tt> methods</a></li>
24 </ul>
25 </li>
Chris Lattnerb512e9f2003-03-04 19:37:49 +0000[diff] [blame]26
Chris Lattner2a19d832004-05-23 21:04:01 +0000[diff] [blame]27 <li><a href="#writingnew">Writing a new <tt>AliasAnalysis</tt> Implementation</a>
Chris Lattnerb512e9f2003-03-04 19:37:49 +0000[diff] [blame]28 <ul>
Misha Brukman9782b032003-10-23 02:29:42 +0000[diff] [blame]29 <li><a href="#passsubclasses">Different Pass styles</a></li>
30 <li><a href="#requiredcalls">Required initialization calls</a></li>
31 <li><a href="#interfaces">Interfaces which may be specified</a></li>
Chris Lattner2a19d832004-05-23 21:04:01 +0000[diff] [blame]32 <li><a href="#chaining"><tt>AliasAnalysis</tt> chaining behavior</a></li>
33 <li><a href="#updating">Updating analysis results for transformations</a></li>
Misha Brukman9782b032003-10-23 02:29:42 +0000[diff] [blame]34 <li><a href="#implefficiency">Efficiency Issues</a></li>
Dan Gohmane1d8d3a2010-12-15 23:09:41 +0000[diff] [blame]35 <li><a href="#limitations">Limitations</a></li>
Chris Lattner2a19d832004-05-23 21:04:01 +0000[diff] [blame]36 </ul>
37 </li>
Chris Lattnerb512e9f2003-03-04 19:37:49 +0000[diff] [blame]38
Chris Lattner2a19d832004-05-23 21:04:01 +0000[diff] [blame]39 <li><a href="#using">Using alias analysis results</a>
Chris Lattnerb512e9f2003-03-04 19:37:49 +0000[diff] [blame]40 <ul>
Chris Lattner6a6d3cc2009-04-25 21:11:37 +0000[diff] [blame]41 <li><a href="#memdep">Using the <tt>MemoryDependenceAnalysis</tt> Pass</a></li>
Misha Brukman9782b032003-10-23 02:29:42 +0000[diff] [blame]42 <li><a href="#ast">Using the <tt>AliasSetTracker</tt> class</a></li>
Chris Lattner2a19d832004-05-23 21:04:01 +0000[diff] [blame]43 <li><a href="#direct">Using the <tt>AliasAnalysis</tt> interface directly</a></li>
44 </ul>
45 </li>
Misha Brukman084a3a072003-10-24 18:06:11 +0000[diff] [blame]46
Chris Lattner2a19d832004-05-23 21:04:01 +0000[diff] [blame]47 <li><a href="#exist">Existing alias analysis implementations and clients</a>
Chris Lattnerb512e9f2003-03-04 19:37:49 +0000[diff] [blame]48 <ul>
Chris Lattner2a19d832004-05-23 21:04:01 +0000[diff] [blame]49 <li><a href="#impls">Available <tt>AliasAnalysis</tt> implementations</a></li>
50 <li><a href="#aliasanalysis-xforms">Alias analysis driven transformations</a></li>
Misha Brukman3eef5d42004-07-01 15:33:24 +0000[diff] [blame]51 <li><a href="#aliasanalysis-debug">Clients for debugging and evaluation of
52 implementations</a></li>
Chris Lattner2a19d832004-05-23 21:04:01 +0000[diff] [blame]53 </ul>
54 </li>
Owen Anderson88847732007-10-02 00:44:20 +0000[diff] [blame]55 <li><a href="#memdep">Memory Dependence Analysis</a></li>
Misha Brukman9782b032003-10-23 02:29:42 +0000[diff] [blame]56</ol>
Chris Lattnerb512e9f2003-03-04 19:37:49 +0000[diff] [blame]57
Chris Lattner2a19d832004-05-23 21:04:01 +0000[diff] [blame]58<div class="doc_author">
59 <p>Written by <a href="mailto:sabre@nondot.org">Chris Lattner</a></p>
Misha Brukman084a3a072003-10-24 18:06:11 +0000[diff] [blame]60</div>
Chris Lattnerb512e9f2003-03-04 19:37:49 +0000[diff] [blame]61
Chris Lattnerb512e9f2003-03-04 19:37:49 +0000[diff] [blame]62<!-- *********************************************************************** -->
NAKAMURA Takumifc8d9302011-04-18 23:59:50 +0000[diff] [blame]63<h2>
Misha Brukman283ff692003-10-22 23:27:16 +0000[diff] [blame]64 <a name="introduction">Introduction</a>
NAKAMURA Takumifc8d9302011-04-18 23:59:50 +0000[diff] [blame]65</h2>
Chris Lattnerb512e9f2003-03-04 19:37:49 +0000[diff] [blame]66<!-- *********************************************************************** -->
67
NAKAMURA Takumi3ad28282011-04-21 01:52:00 +0000[diff] [blame]68<div>
Misha Brukmanbc2d3302003-11-21 22:30:25 +0000[diff] [blame]69
Chris Lattner2a19d832004-05-23 21:04:01 +0000[diff] [blame]70<p>Alias Analysis (aka Pointer Analysis) is a class of techniques which attempt
71to determine whether or not two pointers ever can point to the same object in
72memory. There are many different algorithms for alias analysis and many
73different ways of classifying them: flow-sensitive vs flow-insensitive,
74context-sensitive vs context-insensitive, field-sensitive vs field-insensitive,
75unification-based vs subset-based, etc. Traditionally, alias analyses respond
Duncan Sands16f122e2007-03-30 12:22:09 +0000[diff] [blame]76to a query with a <a href="#MustMayNo">Must, May, or No</a> alias response,
Chris Lattner2a19d832004-05-23 21:04:01 +0000[diff] [blame]77indicating that two pointers always point to the same object, might point to the
78same object, or are known to never point to the same object.</p>
Misha Brukmanbc2d3302003-11-21 22:30:25 +0000[diff] [blame]79
Chris Lattner2a19d832004-05-23 21:04:01 +0000[diff] [blame]80<p>The LLVM <a
Reid Spencerca058542006-03-14 05:39:39 +0000[diff] [blame]81href="http://llvm.org/doxygen/classllvm_1_1AliasAnalysis.html"><tt>AliasAnalysis</tt></a>
Chris Lattner2a19d832004-05-23 21:04:01 +0000[diff] [blame]82class is the primary interface used by clients and implementations of alias
83analyses in the LLVM system. This class is the common interface between clients
84of alias analysis information and the implementations providing it, and is
85designed to support a wide range of implementations and clients (but currently
86all clients are assumed to be flow-insensitive). In addition to simple alias
87analysis information, this class exposes Mod/Ref information from those
88implementations which can provide it, allowing for powerful analyses and
89transformations to work well together.</p>
Misha Brukmanbc2d3302003-11-21 22:30:25 +0000[diff] [blame]90
91<p>This document contains information necessary to successfully implement this
Chris Lattnerb512e9f2003-03-04 19:37:49 +0000[diff] [blame]92interface, use it, and to test both sides. It also explains some of the finer
93points about what exactly results mean. If you feel that something is unclear
Misha Brukman283ff692003-10-22 23:27:16 +0000[diff] [blame]94or should be added, please <a href="mailto:sabre@nondot.org">let me
Misha Brukmanbc2d3302003-11-21 22:30:25 +0000[diff] [blame]95know</a>.</p>
96
Misha Brukman283ff692003-10-22 23:27:16 +0000[diff] [blame]97</div>
Chris Lattnerb512e9f2003-03-04 19:37:49 +0000[diff] [blame]98
99<!-- *********************************************************************** -->
NAKAMURA Takumifc8d9302011-04-18 23:59:50 +0000[diff] [blame]100<h2>
Chris Lattner2a19d832004-05-23 21:04:01 +0000[diff] [blame]101 <a name="overview"><tt>AliasAnalysis</tt> Class Overview</a>
NAKAMURA Takumifc8d9302011-04-18 23:59:50 +0000[diff] [blame]102</h2>
Chris Lattnerb512e9f2003-03-04 19:37:49 +0000[diff] [blame]103<!-- *********************************************************************** -->
104
NAKAMURA Takumi3ad28282011-04-21 01:52:00 +0000[diff] [blame]105<div>
Misha Brukmanbc2d3302003-11-21 22:30:25 +0000[diff] [blame]106
Chris Lattner2a19d832004-05-23 21:04:01 +0000[diff] [blame]107<p>The <a
Reid Spencerca058542006-03-14 05:39:39 +0000[diff] [blame]108href="http://llvm.org/doxygen/classllvm_1_1AliasAnalysis.html"><tt>AliasAnalysis</tt></a>
Chris Lattner2a19d832004-05-23 21:04:01 +0000[diff] [blame]109class defines the interface that the various alias analysis implementations
110should support. This class exports two important enums: <tt>AliasResult</tt>
111and <tt>ModRefResult</tt> which represent the result of an alias query or a
112mod/ref query, respectively.</p>
Misha Brukmanbc2d3302003-11-21 22:30:25 +0000[diff] [blame]113
Chris Lattner2a19d832004-05-23 21:04:01 +0000[diff] [blame]114<p>The <tt>AliasAnalysis</tt> interface exposes information about memory,
115represented in several different ways. In particular, memory objects are
116represented as a starting address and size, and function calls are represented
117as the actual <tt>call</tt> or <tt>invoke</tt> instructions that performs the
118call. The <tt>AliasAnalysis</tt> interface also exposes some helper methods
119which allow you to get mod/ref information for arbitrary instructions.</p>
Misha Brukmanbc2d3302003-11-21 22:30:25 +0000[diff] [blame]120
Dan Gohman00ef9322010-07-07 14:27:09 +0000[diff] [blame]121<p>All <tt>AliasAnalysis</tt> interfaces require that in queries involving
122multiple values, values which are not
123<a href="LangRef.html#constants">constants</a> are all defined within the
124same function.</p>
125
Chris Lattnerb512e9f2003-03-04 19:37:49 +0000[diff] [blame]126<!-- ======================================================================= -->
NAKAMURA Takumifc8d9302011-04-18 23:59:50 +0000[diff] [blame]127<h3>
Misha Brukman283ff692003-10-22 23:27:16 +0000[diff] [blame]128 <a name="pointers">Representation of Pointers</a>
NAKAMURA Takumifc8d9302011-04-18 23:59:50 +0000[diff] [blame]129</h3>
Chris Lattnerb512e9f2003-03-04 19:37:49 +0000[diff] [blame]130
NAKAMURA Takumi3ad28282011-04-21 01:52:00 +0000[diff] [blame]131<div>
Misha Brukman084a3a072003-10-24 18:06:11 +0000[diff] [blame]132
Chris Lattner2a19d832004-05-23 21:04:01 +0000[diff] [blame]133<p>Most importantly, the <tt>AliasAnalysis</tt> class provides several methods
134which are used to query whether or not two memory objects alias, whether
135function calls can modify or read a memory object, etc. For all of these
136queries, memory objects are represented as a pair of their starting address (a
137symbolic LLVM <tt>Value*</tt>) and a static size.</p>
Misha Brukman084a3a072003-10-24 18:06:11 +0000[diff] [blame]138
139<p>Representing memory objects as a starting address and a size is critically
Chris Lattner2a19d832004-05-23 21:04:01 +0000[diff] [blame]140important for correct Alias Analyses. For example, consider this (silly, but
141possible) C code:</p>
Misha Brukman084a3a072003-10-24 18:06:11 +0000[diff] [blame]142
Misha Brukman2361fcf2004-07-28 22:18:33 +0000[diff] [blame]143<div class="doc_code">
Chris Lattnerb512e9f2003-03-04 19:37:49 +0000[diff] [blame]144<pre>
Misha Brukman2361fcf2004-07-28 22:18:33 +0000[diff] [blame]145int i;
146char C[2];
147char A[10];
148/* ... */
149for (i = 0; i != 10; ++i) {
150 C[0] = A[i]; /* One byte store */
151 C[1] = A[9-i]; /* One byte store */
152}
Chris Lattnerb512e9f2003-03-04 19:37:49 +0000[diff] [blame]153</pre>
Misha Brukman2361fcf2004-07-28 22:18:33 +0000[diff] [blame]154</div>
Misha Brukman084a3a072003-10-24 18:06:11 +0000[diff] [blame]155
156<p>In this case, the <tt>basicaa</tt> pass will disambiguate the stores to
Chris Lattnerb512e9f2003-03-04 19:37:49 +0000[diff] [blame]157<tt>C[0]</tt> and <tt>C[1]</tt> because they are accesses to two distinct
158locations one byte apart, and the accesses are each one byte. In this case, the
159LICM pass can use store motion to remove the stores from the loop. In
Misha Brukman084a3a072003-10-24 18:06:11 +0000[diff] [blame]160constrast, the following code:</p>
161
Misha Brukman2361fcf2004-07-28 22:18:33 +0000[diff] [blame]162<div class="doc_code">
Chris Lattnerb512e9f2003-03-04 19:37:49 +0000[diff] [blame]163<pre>
Misha Brukman2361fcf2004-07-28 22:18:33 +0000[diff] [blame]164int i;
165char C[2];
166char A[10];
167/* ... */
168for (i = 0; i != 10; ++i) {
169 ((short*)C)[0] = A[i]; /* Two byte store! */
170 C[1] = A[9-i]; /* One byte store */
171}
Chris Lattnerb512e9f2003-03-04 19:37:49 +0000[diff] [blame]172</pre>
Misha Brukman2361fcf2004-07-28 22:18:33 +0000[diff] [blame]173</div>
Misha Brukman084a3a072003-10-24 18:06:11 +0000[diff] [blame]174
175<p>In this case, the two stores to C do alias each other, because the access to
176the <tt>&amp;C[0]</tt> element is a two byte access. If size information wasn't
Chris Lattnerb512e9f2003-03-04 19:37:49 +0000[diff] [blame]177available in the query, even the first case would have to conservatively assume
Misha Brukman084a3a072003-10-24 18:06:11 +0000[diff] [blame]178that the accesses alias.</p>
179
Misha Brukman283ff692003-10-22 23:27:16 +0000[diff] [blame]180</div>
Chris Lattnerb512e9f2003-03-04 19:37:49 +0000[diff] [blame]181
182<!-- ======================================================================= -->
NAKAMURA Takumifc8d9302011-04-18 23:59:50 +0000[diff] [blame]183<h3>
Chris Lattner2a19d832004-05-23 21:04:01 +0000[diff] [blame]184 <a name="alias">The <tt>alias</tt> method</a>
NAKAMURA Takumifc8d9302011-04-18 23:59:50 +0000[diff] [blame]185</h3>
Chris Lattner2a19d832004-05-23 21:04:01 +0000[diff] [blame]186
NAKAMURA Takumi3ad28282011-04-21 01:52:00 +0000[diff] [blame]187<div>
Dan Gohman00ef9322010-07-07 14:27:09 +0000[diff] [blame]188<p>The <tt>alias</tt> method is the primary interface used to determine whether
189or not two memory objects alias each other. It takes two memory objects as
Dan Gohman4bd64fa2010-12-10 19:38:58 +0000[diff] [blame]190input and returns MustAlias, PartialAlias, MayAlias, or NoAlias as
191appropriate.</p>
Dan Gohman00ef9322010-07-07 14:27:09 +0000[diff] [blame]192
193<p>Like all <tt>AliasAnalysis</tt> interfaces, the <tt>alias</tt> method requires
194that either the two pointer values be defined within the same function, or at
195least one of the values is a <a href="LangRef.html#constants">constant</a>.</p>
Chris Lattner2a19d832004-05-23 21:04:01 +0000[diff] [blame]196
197<!-- _______________________________________________________________________ -->
NAKAMURA Takumifc8d9302011-04-18 23:59:50 +0000[diff] [blame]198<h4>
Misha Brukman283ff692003-10-22 23:27:16 +0000[diff] [blame]199 <a name="MustMayNo">Must, May, and No Alias Responses</a>
NAKAMURA Takumifc8d9302011-04-18 23:59:50 +0000[diff] [blame]200</h4>
Chris Lattnerb512e9f2003-03-04 19:37:49 +0000[diff] [blame]201
NAKAMURA Takumi3ad28282011-04-21 01:52:00 +0000[diff] [blame]202<div>
Dan Gohmandf12d082010-07-02 18:41:32 +0000[diff] [blame]203<p>The NoAlias response may be used when there is never an immediate dependence
204between any memory reference <i>based</i> on one pointer and any memory
205reference <i>based</i> the other. The most obvious example is when the two
206pointers point to non-overlapping memory ranges. Another is when the two
207pointers are only ever used for reading memory. Another is when the memory is
208freed and reallocated between accesses through one pointer and accesses through
209the other -- in this case, there is a dependence, but it's mediated by the free
210and reallocation.</p>
Misha Brukman084a3a072003-10-24 18:06:11 +0000[diff] [blame]211
Dan Gohmande256292010-07-02 23:46:54 +0000[diff] [blame]212<p>As an exception to this is with the
Dan Gohman00ef9322010-07-07 14:27:09 +0000[diff] [blame]213<a href="LangRef.html#noalias"><tt>noalias</tt></a> keyword; the "irrelevant"
214dependencies are ignored.</p>
Dan Gohmande256292010-07-02 23:46:54 +0000[diff] [blame]215
Nick Lewycky4048ca52008-12-14 21:08:48 +0000[diff] [blame]216<p>The MayAlias response is used whenever the two pointers might refer to the
Dan Gohman4bd64fa2010-12-10 19:38:58 +0000[diff] [blame]217same object.</p>
218
219<p>The PartialAlias response is used when the two memory objects are known
220to be overlapping in some way, but do not start at the same address.</p>
Misha Brukman084a3a072003-10-24 18:06:11 +0000[diff] [blame]221
Nick Lewycky4048ca52008-12-14 21:08:48 +0000[diff] [blame]222<p>The MustAlias response may only be returned if the two memory objects are
223guaranteed to always start at exactly the same location. A MustAlias response
224implies that the pointers compare equal.</p>
Misha Brukman084a3a072003-10-24 18:06:11 +0000[diff] [blame]225
Misha Brukman283ff692003-10-22 23:27:16 +0000[diff] [blame]226</div>
Chris Lattnerb512e9f2003-03-04 19:37:49 +0000[diff] [blame]227
NAKAMURA Takumi3ad28282011-04-21 01:52:00 +0000[diff] [blame]228</div>
229
Chris Lattnerb512e9f2003-03-04 19:37:49 +0000[diff] [blame]230<!-- ======================================================================= -->
NAKAMURA Takumifc8d9302011-04-18 23:59:50 +0000[diff] [blame]231<h3>
Misha Brukman283ff692003-10-22 23:27:16 +0000[diff] [blame]232 <a name="ModRefInfo">The <tt>getModRefInfo</tt> methods</a>
NAKAMURA Takumifc8d9302011-04-18 23:59:50 +0000[diff] [blame]233</h3>
Chris Lattnerb512e9f2003-03-04 19:37:49 +0000[diff] [blame]234
NAKAMURA Takumi3ad28282011-04-21 01:52:00 +0000[diff] [blame]235<div>
Misha Brukman084a3a072003-10-24 18:06:11 +0000[diff] [blame]236
237<p>The <tt>getModRefInfo</tt> methods return information about whether the
Chris Lattnerb512e9f2003-03-04 19:37:49 +0000[diff] [blame]238execution of an instruction can read or modify a memory location. Mod/Ref
Chris Lattner2a19d832004-05-23 21:04:01 +0000[diff] [blame]239information is always conservative: if an instruction <b>might</b> read or write
240a location, ModRef is returned.</p>
241
242<p>The <tt>AliasAnalysis</tt> class also provides a <tt>getModRefInfo</tt>
243method for testing dependencies between function calls. This method takes two
Dan Gohmanf2d81712010-08-30 23:47:24 +0000[diff] [blame]244call sites (CS1 &amp; CS2), returns NoModRef if neither call writes to memory
245read or written by the other, Ref if CS1 reads memory written by CS2, Mod if CS1
246writes to memory read or written by CS2, or ModRef if CS1 might read or write
247memory written to by CS2. Note that this relation is not commutative.</p>
Misha Brukman084a3a072003-10-24 18:06:11 +0000[diff] [blame]248
Misha Brukman283ff692003-10-22 23:27:16 +0000[diff] [blame]249</div>
Chris Lattnerb512e9f2003-03-04 19:37:49 +0000[diff] [blame]250
Chris Lattner2a19d832004-05-23 21:04:01 +0000[diff] [blame]251
252<!-- ======================================================================= -->
NAKAMURA Takumifc8d9302011-04-18 23:59:50 +0000[diff] [blame]253<h3>
Chris Lattner2a19d832004-05-23 21:04:01 +0000[diff] [blame]254 <a name="OtherItfs">Other useful <tt>AliasAnalysis</tt> methods</a>
NAKAMURA Takumifc8d9302011-04-18 23:59:50 +0000[diff] [blame]255</h3>
Chris Lattner2a19d832004-05-23 21:04:01 +0000[diff] [blame]256
NAKAMURA Takumi3ad28282011-04-21 01:52:00 +0000[diff] [blame]257<div>
Chris Lattner2a19d832004-05-23 21:04:01 +0000[diff] [blame]258
259<p>
260Several other tidbits of information are often collected by various alias
261analysis implementations and can be put to good use by various clients.
262</p>
263
Chris Lattner2a19d832004-05-23 21:04:01 +0000[diff] [blame]264<!-- _______________________________________________________________________ -->
NAKAMURA Takumifc8d9302011-04-18 23:59:50 +0000[diff] [blame]265<h4>
Chris Lattner2a19d832004-05-23 21:04:01 +0000[diff] [blame]266 The <tt>pointsToConstantMemory</tt> method
NAKAMURA Takumifc8d9302011-04-18 23:59:50 +0000[diff] [blame]267</h4>
Chris Lattner2a19d832004-05-23 21:04:01 +0000[diff] [blame]268
NAKAMURA Takumi3ad28282011-04-21 01:52:00 +0000[diff] [blame]269<div>
Chris Lattner2a19d832004-05-23 21:04:01 +0000[diff] [blame]270
271<p>The <tt>pointsToConstantMemory</tt> method returns true if and only if the
272analysis can prove that the pointer only points to unchanging memory locations
273(functions, constant global variables, and the null pointer). This information
274can be used to refine mod/ref information: it is impossible for an unchanging
275memory location to be modified.</p>
276
277</div>
278
Chris Lattner2a19d832004-05-23 21:04:01 +0000[diff] [blame]279<!-- _______________________________________________________________________ -->
NAKAMURA Takumifc8d9302011-04-18 23:59:50 +0000[diff] [blame]280<h4>
Chris Lattner2a19d832004-05-23 21:04:01 +0000[diff] [blame]281 <a name="simplemodref">The <tt>doesNotAccessMemory</tt> and
282 <tt>onlyReadsMemory</tt> methods</a>
NAKAMURA Takumifc8d9302011-04-18 23:59:50 +0000[diff] [blame]283</h4>
Chris Lattner2a19d832004-05-23 21:04:01 +0000[diff] [blame]284
NAKAMURA Takumi3ad28282011-04-21 01:52:00 +0000[diff] [blame]285<div>
Chris Lattner2a19d832004-05-23 21:04:01 +0000[diff] [blame]286
287<p>These methods are used to provide very simple mod/ref information for
288function calls. The <tt>doesNotAccessMemory</tt> method returns true for a
289function if the analysis can prove that the function never reads or writes to
290memory, or if the function only reads from constant memory. Functions with this
291property are side-effect free and only depend on their input arguments, allowing
292them to be eliminated if they form common subexpressions or be hoisted out of
293loops. Many common functions behave this way (e.g., <tt>sin</tt> and
294<tt>cos</tt>) but many others do not (e.g., <tt>acos</tt>, which modifies the
295<tt>errno</tt> variable).</p>
296
297<p>The <tt>onlyReadsMemory</tt> method returns true for a function if analysis
298can prove that (at most) the function only reads from non-volatile memory.
299Functions with this property are side-effect free, only depending on their input
300arguments and the state of memory when they are called. This property allows
301calls to these functions to be eliminated and moved around, as long as there is
302no store instruction that changes the contents of memory. Note that all
303functions that satisfy the <tt>doesNotAccessMemory</tt> method also satisfies
304<tt>onlyReadsMemory</tt>.</p>
305
306</div>
307
NAKAMURA Takumi3ad28282011-04-21 01:52:00 +0000[diff] [blame]308</div>
309
310</div>
311
Chris Lattnerb512e9f2003-03-04 19:37:49 +0000[diff] [blame]312<!-- *********************************************************************** -->
NAKAMURA Takumifc8d9302011-04-18 23:59:50 +0000[diff] [blame]313<h2>
Chris Lattner2a19d832004-05-23 21:04:01 +0000[diff] [blame]314 <a name="writingnew">Writing a new <tt>AliasAnalysis</tt> Implementation</a>
NAKAMURA Takumifc8d9302011-04-18 23:59:50 +0000[diff] [blame]315</h2>
Chris Lattnerb512e9f2003-03-04 19:37:49 +0000[diff] [blame]316<!-- *********************************************************************** -->
317
NAKAMURA Takumi3ad28282011-04-21 01:52:00 +0000[diff] [blame]318<div>
Misha Brukman084a3a072003-10-24 18:06:11 +0000[diff] [blame]319
320<p>Writing a new alias analysis implementation for LLVM is quite
321straight-forward. There are already several implementations that you can use
322for examples, and the following information should help fill in any details.
Chris Lattner2a19d832004-05-23 21:04:01 +0000[diff] [blame]323For a examples, take a look at the <a href="#impls">various alias analysis
324implementations</a> included with LLVM.</p>
Misha Brukman084a3a072003-10-24 18:06:11 +0000[diff] [blame]325
Chris Lattnerb512e9f2003-03-04 19:37:49 +0000[diff] [blame]326<!-- ======================================================================= -->
NAKAMURA Takumifc8d9302011-04-18 23:59:50 +0000[diff] [blame]327<h3>
Misha Brukman283ff692003-10-22 23:27:16 +0000[diff] [blame]328 <a name="passsubclasses">Different Pass styles</a>
NAKAMURA Takumifc8d9302011-04-18 23:59:50 +0000[diff] [blame]329</h3>
Chris Lattnerb512e9f2003-03-04 19:37:49 +0000[diff] [blame]330
NAKAMURA Takumi3ad28282011-04-21 01:52:00 +0000[diff] [blame]331<div>
Misha Brukman084a3a072003-10-24 18:06:11 +0000[diff] [blame]332
333<p>The first step to determining what type of <a
334href="WritingAnLLVMPass.html">LLVM pass</a> you need to use for your Alias
335Analysis. As is the case with most other analyses and transformations, the
336answer should be fairly obvious from what type of problem you are trying to
337solve:</p>
338
Chris Lattnerb512e9f2003-03-04 19:37:49 +0000[diff] [blame]339<ol>
Misha Brukman283ff692003-10-22 23:27:16 +0000[diff] [blame]340 <li>If you require interprocedural analysis, it should be a
341 <tt>Pass</tt>.</li>
Chris Lattner2a19d832004-05-23 21:04:01 +0000[diff] [blame]342 <li>If you are a function-local analysis, subclass <tt>FunctionPass</tt>.</li>
Misha Brukman283ff692003-10-22 23:27:16 +0000[diff] [blame]343 <li>If you don't need to look at the program at all, subclass
344 <tt>ImmutablePass</tt>.</li>
345</ol>
Misha Brukman084a3a072003-10-24 18:06:11 +0000[diff] [blame]346
347<p>In addition to the pass that you subclass, you should also inherit from the
Misha Brukmanb4dd5392003-05-07 21:47:16 +0000[diff] [blame]348<tt>AliasAnalysis</tt> interface, of course, and use the
Chris Lattnerb512e9f2003-03-04 19:37:49 +0000[diff] [blame]349<tt>RegisterAnalysisGroup</tt> template to register as an implementation of
Misha Brukman084a3a072003-10-24 18:06:11 +0000[diff] [blame]350<tt>AliasAnalysis</tt>.</p>
351
Misha Brukman283ff692003-10-22 23:27:16 +0000[diff] [blame]352</div>
Chris Lattnerb512e9f2003-03-04 19:37:49 +0000[diff] [blame]353
354<!-- ======================================================================= -->
NAKAMURA Takumifc8d9302011-04-18 23:59:50 +0000[diff] [blame]355<h3>
Misha Brukman283ff692003-10-22 23:27:16 +0000[diff] [blame]356 <a name="requiredcalls">Required initialization calls</a>
NAKAMURA Takumifc8d9302011-04-18 23:59:50 +0000[diff] [blame]357</h3>
Chris Lattnerb512e9f2003-03-04 19:37:49 +0000[diff] [blame]358
NAKAMURA Takumi3ad28282011-04-21 01:52:00 +0000[diff] [blame]359<div>
Misha Brukman084a3a072003-10-24 18:06:11 +0000[diff] [blame]360
Chris Lattner2a19d832004-05-23 21:04:01 +0000[diff] [blame]361<p>Your subclass of <tt>AliasAnalysis</tt> is required to invoke two methods on
362the <tt>AliasAnalysis</tt> base class: <tt>getAnalysisUsage</tt> and
Chris Lattnerb512e9f2003-03-04 19:37:49 +0000[diff] [blame]363<tt>InitializeAliasAnalysis</tt>. In particular, your implementation of
364<tt>getAnalysisUsage</tt> should explicitly call into the
365<tt>AliasAnalysis::getAnalysisUsage</tt> method in addition to doing any
366declaring any pass dependencies your pass has. Thus you should have something
Misha Brukman084a3a072003-10-24 18:06:11 +0000[diff] [blame]367like this:</p>
368
Misha Brukman2361fcf2004-07-28 22:18:33 +0000[diff] [blame]369<div class="doc_code">
Chris Lattnerb512e9f2003-03-04 19:37:49 +0000[diff] [blame]370<pre>
Misha Brukman2361fcf2004-07-28 22:18:33 +0000[diff] [blame]371void getAnalysisUsage(AnalysisUsage &amp;AU) const {
372 AliasAnalysis::getAnalysisUsage(AU);
373 <i>// declare your dependencies here.</i>
374}
Chris Lattnerb512e9f2003-03-04 19:37:49 +0000[diff] [blame]375</pre>
Misha Brukman2361fcf2004-07-28 22:18:33 +0000[diff] [blame]376</div>
Misha Brukman084a3a072003-10-24 18:06:11 +0000[diff] [blame]377
378<p>Additionally, your must invoke the <tt>InitializeAliasAnalysis</tt> method
379from your analysis run method (<tt>run</tt> for a <tt>Pass</tt>,
Chris Lattner2a19d832004-05-23 21:04:01 +0000[diff] [blame]380<tt>runOnFunction</tt> for a <tt>FunctionPass</tt>, or <tt>InitializePass</tt>
381for an <tt>ImmutablePass</tt>). For example (as part of a <tt>Pass</tt>):</p>
Misha Brukman084a3a072003-10-24 18:06:11 +0000[diff] [blame]382
Misha Brukman2361fcf2004-07-28 22:18:33 +0000[diff] [blame]383<div class="doc_code">
Chris Lattnerb512e9f2003-03-04 19:37:49 +0000[diff] [blame]384<pre>
Misha Brukman2361fcf2004-07-28 22:18:33 +0000[diff] [blame]385bool run(Module &amp;M) {
386 InitializeAliasAnalysis(this);
387 <i>// Perform analysis here...</i>
388 return false;
389}
Chris Lattnerb512e9f2003-03-04 19:37:49 +0000[diff] [blame]390</pre>
Misha Brukman2361fcf2004-07-28 22:18:33 +0000[diff] [blame]391</div>
Misha Brukman084a3a072003-10-24 18:06:11 +0000[diff] [blame]392
Misha Brukman283ff692003-10-22 23:27:16 +0000[diff] [blame]393</div>
Chris Lattnerb512e9f2003-03-04 19:37:49 +0000[diff] [blame]394
395<!-- ======================================================================= -->
NAKAMURA Takumifc8d9302011-04-18 23:59:50 +0000[diff] [blame]396<h3>
Misha Brukman283ff692003-10-22 23:27:16 +0000[diff] [blame]397 <a name="interfaces">Interfaces which may be specified</a>
NAKAMURA Takumifc8d9302011-04-18 23:59:50 +0000[diff] [blame]398</h3>
Chris Lattnerb512e9f2003-03-04 19:37:49 +0000[diff] [blame]399
NAKAMURA Takumi3ad28282011-04-21 01:52:00 +0000[diff] [blame]400<div>
Misha Brukman084a3a072003-10-24 18:06:11 +0000[diff] [blame]401
Chris Lattner2a19d832004-05-23 21:04:01 +0000[diff] [blame]402<p>All of the <a
403href="/doxygen/classllvm_1_1AliasAnalysis.html"><tt>AliasAnalysis</tt></a>
404virtual methods default to providing <a href="#chaining">chaining</a> to another
405alias analysis implementation, which ends up returning conservatively correct
406information (returning "May" Alias and "Mod/Ref" for alias and mod/ref queries
Misha Brukman084a3a072003-10-24 18:06:11 +0000[diff] [blame]407respectively). Depending on the capabilities of the analysis you are
408implementing, you just override the interfaces you can improve.</p>
409
Misha Brukman283ff692003-10-22 23:27:16 +0000[diff] [blame]410</div>
Chris Lattnerb512e9f2003-03-04 19:37:49 +0000[diff] [blame]411
Chris Lattner2a19d832004-05-23 21:04:01 +0000[diff] [blame]412
413
Chris Lattnerb512e9f2003-03-04 19:37:49 +0000[diff] [blame]414<!-- ======================================================================= -->
NAKAMURA Takumifc8d9302011-04-18 23:59:50 +0000[diff] [blame]415<h3>
Chris Lattner2a19d832004-05-23 21:04:01 +0000[diff] [blame]416 <a name="chaining"><tt>AliasAnalysis</tt> chaining behavior</a>
NAKAMURA Takumifc8d9302011-04-18 23:59:50 +0000[diff] [blame]417</h3>
Chris Lattnerb512e9f2003-03-04 19:37:49 +0000[diff] [blame]418
NAKAMURA Takumi3ad28282011-04-21 01:52:00 +0000[diff] [blame]419<div>
Misha Brukman084a3a072003-10-24 18:06:11 +0000[diff] [blame]420
Chris Lattner2a19d832004-05-23 21:04:01 +0000[diff] [blame]421<p>With only two special exceptions (the <tt><a
422href="#basic-aa">basicaa</a></tt> and <a href="#no-aa"><tt>no-aa</tt></a>
423passes) every alias analysis pass chains to another alias analysis
424implementation (for example, the user can specify "<tt>-basicaa -ds-aa
Chris Lattner5ea3e652010-03-01 19:24:17 +0000[diff] [blame]425-licm</tt>" to get the maximum benefit from both alias
Chris Lattner2a19d832004-05-23 21:04:01 +0000[diff] [blame]426analyses). The alias analysis class automatically takes care of most of this
427for methods that you don't override. For methods that you do override, in code
428paths that return a conservative MayAlias or Mod/Ref result, simply return
429whatever the superclass computes. For example:</p>
Misha Brukman084a3a072003-10-24 18:06:11 +0000[diff] [blame]430
Misha Brukman2361fcf2004-07-28 22:18:33 +0000[diff] [blame]431<div class="doc_code">
Chris Lattner2a19d832004-05-23 21:04:01 +0000[diff] [blame]432<pre>
Misha Brukman2361fcf2004-07-28 22:18:33 +0000[diff] [blame]433AliasAnalysis::AliasResult alias(const Value *V1, unsigned V1Size,
434 const Value *V2, unsigned V2Size) {
435 if (...)
436 return NoAlias;
437 ...
Chris Lattner2a19d832004-05-23 21:04:01 +0000[diff] [blame]438
Misha Brukman2361fcf2004-07-28 22:18:33 +0000[diff] [blame]439 <i>// Couldn't determine a must or no-alias result.</i>
440 return AliasAnalysis::alias(V1, V1Size, V2, V2Size);
441}
Chris Lattner2a19d832004-05-23 21:04:01 +0000[diff] [blame]442</pre>
Misha Brukman2361fcf2004-07-28 22:18:33 +0000[diff] [blame]443</div>
Chris Lattner2a19d832004-05-23 21:04:01 +0000[diff] [blame]444
445<p>In addition to analysis queries, you must make sure to unconditionally pass
446LLVM <a href="#updating">update notification</a> methods to the superclass as
447well if you override them, which allows all alias analyses in a change to be
448updated.</p>
449
450</div>
451
452
453<!-- ======================================================================= -->
NAKAMURA Takumifc8d9302011-04-18 23:59:50 +0000[diff] [blame]454<h3>
Chris Lattner2a19d832004-05-23 21:04:01 +0000[diff] [blame]455 <a name="updating">Updating analysis results for transformations</a>
NAKAMURA Takumifc8d9302011-04-18 23:59:50 +0000[diff] [blame]456</h3>
Chris Lattner2a19d832004-05-23 21:04:01 +0000[diff] [blame]457
NAKAMURA Takumi3ad28282011-04-21 01:52:00 +0000[diff] [blame]458<div>
Chris Lattner2a19d832004-05-23 21:04:01 +0000[diff] [blame]459<p>
460Alias analysis information is initially computed for a static snapshot of the
461program, but clients will use this information to make transformations to the
462code. All but the most trivial forms of alias analysis will need to have their
463analysis results updated to reflect the changes made by these transformations.
464</p>
465
466<p>
Owen Andersonb6e4ff02011-01-03 21:38:41 +0000[diff] [blame]467The <tt>AliasAnalysis</tt> interface exposes four methods which are used to
Chris Lattner2a19d832004-05-23 21:04:01 +0000[diff] [blame]468communicate program changes from the clients to the analysis implementations.
469Various alias analysis implementations should use these methods to ensure that
470their internal data structures are kept up-to-date as the program changes (for
471example, when an instruction is deleted), and clients of alias analysis must be
472sure to call these interfaces appropriately.
473</p>
Chris Lattner2a19d832004-05-23 21:04:01 +0000[diff] [blame]474
475<!-- _______________________________________________________________________ -->
NAKAMURA Takumifc8d9302011-04-18 23:59:50 +0000[diff] [blame]476<h4>The <tt>deleteValue</tt> method</h4>
Chris Lattner2a19d832004-05-23 21:04:01 +0000[diff] [blame]477
NAKAMURA Takumi3ad28282011-04-21 01:52:00 +0000[diff] [blame]478<div>
Chris Lattner2a19d832004-05-23 21:04:01 +0000[diff] [blame]479The <tt>deleteValue</tt> method is called by transformations when they remove an
480instruction or any other value from the program (including values that do not
481use pointers). Typically alias analyses keep data structures that have entries
482for each value in the program. When this method is called, they should remove
483any entries for the specified value, if they exist.
484</div>
485
Chris Lattner2a19d832004-05-23 21:04:01 +0000[diff] [blame]486<!-- _______________________________________________________________________ -->
NAKAMURA Takumifc8d9302011-04-18 23:59:50 +0000[diff] [blame]487<h4>The <tt>copyValue</tt> method</h4>
Chris Lattner2a19d832004-05-23 21:04:01 +0000[diff] [blame]488
NAKAMURA Takumi3ad28282011-04-21 01:52:00 +0000[diff] [blame]489<div>
Chris Lattner2a19d832004-05-23 21:04:01 +0000[diff] [blame]490The <tt>copyValue</tt> method is used when a new value is introduced into the
491program. There is no way to introduce a value into the program that did not
492exist before (this doesn't make sense for a safe compiler transformation), so
493this is the only way to introduce a new value. This method indicates that the
494new value has exactly the same properties as the value being copied.
495</div>
496
Chris Lattner2a19d832004-05-23 21:04:01 +0000[diff] [blame]497<!-- _______________________________________________________________________ -->
NAKAMURA Takumifc8d9302011-04-18 23:59:50 +0000[diff] [blame]498<h4>The <tt>replaceWithNewValue</tt> method</h4>
Chris Lattner2a19d832004-05-23 21:04:01 +0000[diff] [blame]499
NAKAMURA Takumi3ad28282011-04-21 01:52:00 +0000[diff] [blame]500<div>
Chris Lattner2a19d832004-05-23 21:04:01 +0000[diff] [blame]501This method is a simple helper method that is provided to make clients easier to
502use. It is implemented by copying the old analysis information to the new
503value, then deleting the old value. This method cannot be overridden by alias
504analysis implementations.
Misha Brukman283ff692003-10-22 23:27:16 +0000[diff] [blame]505</div>
Chris Lattnerb512e9f2003-03-04 19:37:49 +0000[diff] [blame]506
Owen Andersonb6e4ff02011-01-03 21:38:41 +0000[diff] [blame]507<!-- _______________________________________________________________________ -->
NAKAMURA Takumifc8d9302011-04-18 23:59:50 +0000[diff] [blame]508<h4>The <tt>addEscapingUse</tt> method</h4>
Owen Andersonb6e4ff02011-01-03 21:38:41 +0000[diff] [blame]509
NAKAMURA Takumi3ad28282011-04-21 01:52:00 +0000[diff] [blame]510<div>
Owen Andersonb6e4ff02011-01-03 21:38:41 +0000[diff] [blame]511<p>The <tt>addEscapingUse</tt> method is used when the uses of a pointer
512value have changed in ways that may invalidate precomputed analysis information.
513Implementations may either use this callback to provide conservative responses
514for points whose uses have change since analysis time, or may recompute some
515or all of their internal state to continue providing accurate responses.</p>
516
517<p>In general, any new use of a pointer value is considered an escaping use,
518and must be reported through this callback, <em>except</em> for the
519uses below:</p>
520
521<ul>
522 <li>A <tt>bitcast</tt> or <tt>getelementptr</tt> of the pointer</li>
523 <li>A <tt>store</tt> through the pointer (but not a <tt>store</tt>
524 <em>of</em> the pointer)</li>
525 <li>A <tt>load</tt> through the pointer</li>
526</ul>
527</div>
528
NAKAMURA Takumi3ad28282011-04-21 01:52:00 +0000[diff] [blame]529</div>
530
Chris Lattnerb512e9f2003-03-04 19:37:49 +0000[diff] [blame]531<!-- ======================================================================= -->
NAKAMURA Takumifc8d9302011-04-18 23:59:50 +0000[diff] [blame]532<h3>
Misha Brukman283ff692003-10-22 23:27:16 +0000[diff] [blame]533 <a name="implefficiency">Efficiency Issues</a>
NAKAMURA Takumifc8d9302011-04-18 23:59:50 +0000[diff] [blame]534</h3>
Chris Lattnerb512e9f2003-03-04 19:37:49 +0000[diff] [blame]535
NAKAMURA Takumi3ad28282011-04-21 01:52:00 +0000[diff] [blame]536<div>
Misha Brukman084a3a072003-10-24 18:06:11 +0000[diff] [blame]537
538<p>From the LLVM perspective, the only thing you need to do to provide an
539efficient alias analysis is to make sure that alias analysis <b>queries</b> are
540serviced quickly. The actual calculation of the alias analysis results (the
541"run" method) is only performed once, but many (perhaps duplicate) queries may
542be performed. Because of this, try to move as much computation to the run
543method as possible (within reason).</p>
544
Misha Brukman283ff692003-10-22 23:27:16 +0000[diff] [blame]545</div>
Chris Lattnerb512e9f2003-03-04 19:37:49 +0000[diff] [blame]546
Dan Gohmana3de0182010-06-24 19:34:03 +0000[diff] [blame]547<!-- ======================================================================= -->
NAKAMURA Takumifc8d9302011-04-18 23:59:50 +0000[diff] [blame]548<h3>
Dan Gohmanb3f04f42010-12-15 18:45:20 +0000[diff] [blame]549 <a name="limitations">Limitations</a>
NAKAMURA Takumifc8d9302011-04-18 23:59:50 +0000[diff] [blame]550</h3>
Dan Gohmana3de0182010-06-24 19:34:03 +0000[diff] [blame]551
NAKAMURA Takumi3ad28282011-04-21 01:52:00 +0000[diff] [blame]552<div>
Dan Gohmana3de0182010-06-24 19:34:03 +0000[diff] [blame]553
Dan Gohmanb3f04f42010-12-15 18:45:20 +0000[diff] [blame]554<p>The AliasAnalysis infrastructure has several limitations which make
555writing a new <tt>AliasAnalysis</tt> implementation difficult.</p>
Dan Gohmana3de0182010-06-24 19:34:03 +0000[diff] [blame]556
557<p>There is no way to override the default alias analysis. It would
558be very useful to be able to do something like "opt -my-aa -O2" and
559have it use -my-aa for all passes which need AliasAnalysis, but there
560is currently no support for that, short of changing the source code
561and recompiling. Similarly, there is also no way of setting a chain
562of analyses as the default.</p>
563
564<p>There is no way for transform passes to declare that they preserve
565<tt>AliasAnalysis</tt> implementations. The <tt>AliasAnalysis</tt>
566interface includes <tt>deleteValue</tt> and <tt>copyValue</tt> methods
567which are intended to allow a pass to keep an AliasAnalysis consistent,
568however there's no way for a pass to declare in its
569<tt>getAnalysisUsage</tt> that it does so. Some passes attempt to use
570<tt>AU.addPreserved&lt;AliasAnalysis&gt;</tt>, however this doesn't
NAKAMURA Takumi6bd36d52011-04-09 09:51:57 +0000[diff] [blame]571actually have any effect.</p>
Dan Gohmana3de0182010-06-24 19:34:03 +0000[diff] [blame]572
573<p><tt>AliasAnalysisCounter</tt> (<tt>-count-aa</tt>) and <tt>AliasDebugger</tt>
574(<tt>-debug-aa</tt>) are implemented as <tt>ModulePass</tt> classes, so if your
575alias analysis uses <tt>FunctionPass</tt>, it won't be able to use
576these utilities. If you try to use them, the pass manager will
577silently route alias analysis queries directly to
578<tt>BasicAliasAnalysis</tt> instead.</p>
579
580<p>Similarly, the <tt>opt -p</tt> option introduces <tt>ModulePass</tt>
581passes between each pass, which prevents the use of <tt>FunctionPass</tt>
582alias analysis passes.</p>
583
Dan Gohmanb3f04f42010-12-15 18:45:20 +0000[diff] [blame]584<p>The <tt>AliasAnalysis</tt> API does have functions for notifying
585implementations when values are deleted or copied, however these
586aren't sufficient. There are many other ways that LLVM IR can be
587modified which could be relevant to <tt>AliasAnalysis</tt>
588implementations which can not be expressed.</p>
589
590<p>The <tt>AliasAnalysisDebugger</tt> utility seems to suggest that
591<tt>AliasAnalysis</tt> implementations can expect that they will be
592informed of any relevant <tt>Value</tt> before it appears in an
593alias query. However, popular clients such as <tt>GVN</tt> don't
594support this, and are known to trigger errors when run with the
595<tt>AliasAnalysisDebugger</tt>.</p>
596
597<p>Due to several of the above limitations, the most obvious use for
598the <tt>AliasAnalysisCounter</tt> utility, collecting stats on all
599alias queries in a compilation, doesn't work, even if the
600<tt>AliasAnalysis</tt> implementations don't use <tt>FunctionPass</tt>.
601There's no way to set a default, much less a default sequence,
602and there's no way to preserve it.</p>
603
604<p>The <tt>AliasSetTracker</tt> class (which is used by <tt>LICM</tt>
605makes a non-deterministic number of alias queries. This can cause stats
606collected by <tt>AliasAnalysisCounter</tt> to have fluctuations among
607identical runs, for example. Another consequence is that debugging
608techniques involving pausing execution after a predetermined number
609of queries can be unreliable.</p>
610
611<p>Many alias queries can be reformulated in terms of other alias
612queries. When multiple <tt>AliasAnalysis</tt> queries are chained together,
613it would make sense to start those queries from the beginning of the chain,
614with care taken to avoid infinite looping, however currently an
615implementation which wants to do this can only start such queries
616from itself.</p>
617
Dan Gohmana3de0182010-06-24 19:34:03 +0000[diff] [blame]618</div>
619
NAKAMURA Takumi3ad28282011-04-21 01:52:00 +0000[diff] [blame]620</div>
621
Chris Lattnerb512e9f2003-03-04 19:37:49 +0000[diff] [blame]622<!-- *********************************************************************** -->
NAKAMURA Takumifc8d9302011-04-18 23:59:50 +0000[diff] [blame]623<h2>
Chris Lattner2a19d832004-05-23 21:04:01 +0000[diff] [blame]624 <a name="using">Using alias analysis results</a>
NAKAMURA Takumifc8d9302011-04-18 23:59:50 +0000[diff] [blame]625</h2>
Chris Lattnerb512e9f2003-03-04 19:37:49 +0000[diff] [blame]626<!-- *********************************************************************** -->
627
NAKAMURA Takumi3ad28282011-04-21 01:52:00 +0000[diff] [blame]628<div>
Misha Brukman084a3a072003-10-24 18:06:11 +0000[diff] [blame]629
630<p>There are several different ways to use alias analysis results. In order of
631preference, these are...</p>
632
Chris Lattnerb512e9f2003-03-04 19:37:49 +0000[diff] [blame]633<!-- ======================================================================= -->
NAKAMURA Takumifc8d9302011-04-18 23:59:50 +0000[diff] [blame]634<h3>
Chris Lattner6a6d3cc2009-04-25 21:11:37 +0000[diff] [blame]635 <a name="memdep">Using the <tt>MemoryDependenceAnalysis</tt> Pass</a>
NAKAMURA Takumifc8d9302011-04-18 23:59:50 +0000[diff] [blame]636</h3>
Chris Lattnerb512e9f2003-03-04 19:37:49 +0000[diff] [blame]637
NAKAMURA Takumi3ad28282011-04-21 01:52:00 +0000[diff] [blame]638<div>
Misha Brukman084a3a072003-10-24 18:06:11 +0000[diff] [blame]639
Chris Lattner6a6d3cc2009-04-25 21:11:37 +0000[diff] [blame]640<p>The <tt>memdep</tt> pass uses alias analysis to provide high-level dependence
641information about memory-using instructions. This will tell you which store
642feeds into a load, for example. It uses caching and other techniques to be
643efficient, and is used by Dead Store Elimination, GVN, and memcpy optimizations.
644</p>
Misha Brukman084a3a072003-10-24 18:06:11 +0000[diff] [blame]645
Misha Brukman283ff692003-10-22 23:27:16 +0000[diff] [blame]646</div>
Chris Lattnerb512e9f2003-03-04 19:37:49 +0000[diff] [blame]647
648<!-- ======================================================================= -->
NAKAMURA Takumifc8d9302011-04-18 23:59:50 +0000[diff] [blame]649<h3>
Misha Brukman283ff692003-10-22 23:27:16 +0000[diff] [blame]650 <a name="ast">Using the <tt>AliasSetTracker</tt> class</a>
NAKAMURA Takumifc8d9302011-04-18 23:59:50 +0000[diff] [blame]651</h3>
Chris Lattnerb512e9f2003-03-04 19:37:49 +0000[diff] [blame]652
NAKAMURA Takumi3ad28282011-04-21 01:52:00 +0000[diff] [blame]653<div>
Misha Brukman084a3a072003-10-24 18:06:11 +0000[diff] [blame]654
655<p>Many transformations need information about alias <b>sets</b> that are active
656in some scope, rather than information about pairwise aliasing. The <tt><a
Misha Brukman2361fcf2004-07-28 22:18:33 +0000[diff] [blame]657href="/doxygen/classllvm_1_1AliasSetTracker.html">AliasSetTracker</a></tt> class
658is used to efficiently build these Alias Sets from the pairwise alias analysis
Chris Lattner2a19d832004-05-23 21:04:01 +0000[diff] [blame]659information provided by the <tt>AliasAnalysis</tt> interface.</p>
Misha Brukman084a3a072003-10-24 18:06:11 +0000[diff] [blame]660
Chris Lattner2a19d832004-05-23 21:04:01 +0000[diff] [blame]661<p>First you initialize the AliasSetTracker by using the "<tt>add</tt>" methods
662to add information about various potentially aliasing instructions in the scope
663you are interested in. Once all of the alias sets are completed, your pass
664should simply iterate through the constructed alias sets, using the
665<tt>AliasSetTracker</tt> <tt>begin()</tt>/<tt>end()</tt> methods.</p>
Misha Brukman084a3a072003-10-24 18:06:11 +0000[diff] [blame]666
667<p>The <tt>AliasSet</tt>s formed by the <tt>AliasSetTracker</tt> are guaranteed
Chris Lattnerb8b9163a2003-12-19 08:43:07 +0000[diff] [blame]668to be disjoint, calculate mod/ref information and volatility for the set, and
669keep track of whether or not all of the pointers in the set are Must aliases.
670The AliasSetTracker also makes sure that sets are properly folded due to call
Misha Brukman084a3a072003-10-24 18:06:11 +0000[diff] [blame]671instructions, and can provide a list of pointers in each set.</p>
672
673<p>As an example user of this, the <a href="/doxygen/structLICM.html">Loop
Chris Lattner2a19d832004-05-23 21:04:01 +0000[diff] [blame]674Invariant Code Motion</a> pass uses <tt>AliasSetTracker</tt>s to calculate alias
675sets for each loop nest. If an <tt>AliasSet</tt> in a loop is not modified,
676then all load instructions from that set may be hoisted out of the loop. If any
677alias sets are stored to <b>and</b> are must alias sets, then the stores may be
678sunk to outside of the loop, promoting the memory location to a register for the
679duration of the loop nest. Both of these transformations only apply if the
680pointer argument is loop-invariant.</p>
Misha Brukman084a3a072003-10-24 18:06:11 +0000[diff] [blame]681
Chris Lattner2a19d832004-05-23 21:04:01 +0000[diff] [blame]682<!-- _______________________________________________________________________ -->
NAKAMURA Takumifc8d9302011-04-18 23:59:50 +0000[diff] [blame]683<h4>
Chris Lattnerb8b9163a2003-12-19 08:43:07 +0000[diff] [blame]684 The AliasSetTracker implementation
NAKAMURA Takumifc8d9302011-04-18 23:59:50 +0000[diff] [blame]685</h4>
Chris Lattnerb8b9163a2003-12-19 08:43:07 +0000[diff] [blame]686
NAKAMURA Takumi3ad28282011-04-21 01:52:00 +0000[diff] [blame]687<div>
Chris Lattnerb8b9163a2003-12-19 08:43:07 +0000[diff] [blame]688
689<p>The AliasSetTracker class is implemented to be as efficient as possible. It
690uses the union-find algorithm to efficiently merge AliasSets when a pointer is
691inserted into the AliasSetTracker that aliases multiple sets. The primary data
692structure is a hash table mapping pointers to the AliasSet they are in.</p>
693
694<p>The AliasSetTracker class must maintain a list of all of the LLVM Value*'s
695that are in each AliasSet. Since the hash table already has entries for each
696LLVM Value* of interest, the AliasesSets thread the linked list through these
697hash-table nodes to avoid having to allocate memory unnecessarily, and to make
698merging alias sets extremely efficient (the linked list merge is constant time).
699</p>
700
701<p>You shouldn't need to understand these details if you are just a client of
702the AliasSetTracker, but if you look at the code, hopefully this brief
703description will help make sense of why things are designed the way they
704are.</p>
705
706</div>
707
NAKAMURA Takumi3ad28282011-04-21 01:52:00 +0000[diff] [blame]708</div>
709
Chris Lattnerb512e9f2003-03-04 19:37:49 +0000[diff] [blame]710<!-- ======================================================================= -->
NAKAMURA Takumifc8d9302011-04-18 23:59:50 +0000[diff] [blame]711<h3>
Chris Lattner2a19d832004-05-23 21:04:01 +0000[diff] [blame]712 <a name="direct">Using the <tt>AliasAnalysis</tt> interface directly</a>
NAKAMURA Takumifc8d9302011-04-18 23:59:50 +0000[diff] [blame]713</h3>
Chris Lattnerb512e9f2003-03-04 19:37:49 +0000[diff] [blame]714
NAKAMURA Takumi3ad28282011-04-21 01:52:00 +0000[diff] [blame]715<div>
Misha Brukman084a3a072003-10-24 18:06:11 +0000[diff] [blame]716
Chris Lattner2a19d832004-05-23 21:04:01 +0000[diff] [blame]717<p>If neither of these utility class are what your pass needs, you should use
718the interfaces exposed by the <tt>AliasAnalysis</tt> class directly. Try to use
719the higher-level methods when possible (e.g., use mod/ref information instead of
720the <a href="#alias"><tt>alias</tt></a> method directly if possible) to get the
721best precision and efficiency.</p>
Misha Brukman084a3a072003-10-24 18:06:11 +0000[diff] [blame]722
Misha Brukman283ff692003-10-22 23:27:16 +0000[diff] [blame]723</div>
Chris Lattnerb512e9f2003-03-04 19:37:49 +0000[diff] [blame]724
NAKAMURA Takumi3ad28282011-04-21 01:52:00 +0000[diff] [blame]725</div>
726
Chris Lattnerb512e9f2003-03-04 19:37:49 +0000[diff] [blame]727<!-- *********************************************************************** -->
NAKAMURA Takumifc8d9302011-04-18 23:59:50 +0000[diff] [blame]728<h2>
Chris Lattner2a19d832004-05-23 21:04:01 +0000[diff] [blame]729 <a name="exist">Existing alias analysis implementations and clients</a>
NAKAMURA Takumifc8d9302011-04-18 23:59:50 +0000[diff] [blame]730</h2>
Chris Lattnerb512e9f2003-03-04 19:37:49 +0000[diff] [blame]731<!-- *********************************************************************** -->
732
NAKAMURA Takumi3ad28282011-04-21 01:52:00 +0000[diff] [blame]733<div>
Misha Brukman084a3a072003-10-24 18:06:11 +0000[diff] [blame]734
Chris Lattner2a19d832004-05-23 21:04:01 +0000[diff] [blame]735<p>If you're going to be working with the LLVM alias analysis infrastructure,
736you should know what clients and implementations of alias analysis are
737available. In particular, if you are implementing an alias analysis, you should
738be aware of the <a href="#aliasanalysis-debug">the clients</a> that are useful
739for monitoring and evaluating different implementations.</p>
Misha Brukman084a3a072003-10-24 18:06:11 +0000[diff] [blame]740
Chris Lattnerb512e9f2003-03-04 19:37:49 +0000[diff] [blame]741<!-- ======================================================================= -->
NAKAMURA Takumifc8d9302011-04-18 23:59:50 +0000[diff] [blame]742<h3>
Chris Lattner2a19d832004-05-23 21:04:01 +0000[diff] [blame]743 <a name="impls">Available <tt>AliasAnalysis</tt> implementations</a>
NAKAMURA Takumifc8d9302011-04-18 23:59:50 +0000[diff] [blame]744</h3>
Chris Lattner2a19d832004-05-23 21:04:01 +0000[diff] [blame]745
NAKAMURA Takumi3ad28282011-04-21 01:52:00 +0000[diff] [blame]746<div>
Chris Lattner2a19d832004-05-23 21:04:01 +0000[diff] [blame]747
748<p>This section lists the various implementations of the <tt>AliasAnalysis</tt>
Dan Gohman4a70bfe2011-05-24 22:01:32 +0000[diff] [blame]749interface. With the exception of the <a href="#no-aa"><tt>-no-aa</tt></a>
750implementation, all of these <a href="#chaining">chain</a> to other alias
751analysis implementations.</p>
Chris Lattner2a19d832004-05-23 21:04:01 +0000[diff] [blame]752
Chris Lattner2a19d832004-05-23 21:04:01 +0000[diff] [blame]753<!-- _______________________________________________________________________ -->
NAKAMURA Takumifc8d9302011-04-18 23:59:50 +0000[diff] [blame]754<h4>
Misha Brukman283ff692003-10-22 23:27:16 +0000[diff] [blame]755 <a name="no-aa">The <tt>-no-aa</tt> pass</a>
NAKAMURA Takumifc8d9302011-04-18 23:59:50 +0000[diff] [blame]756</h4>
Chris Lattnerb512e9f2003-03-04 19:37:49 +0000[diff] [blame]757
NAKAMURA Takumi3ad28282011-04-21 01:52:00 +0000[diff] [blame]758<div>
Misha Brukman084a3a072003-10-24 18:06:11 +0000[diff] [blame]759
Chris Lattner2a19d832004-05-23 21:04:01 +0000[diff] [blame]760<p>The <tt>-no-aa</tt> pass is just like what it sounds: an alias analysis that
761never returns any useful information. This pass can be useful if you think that
762alias analysis is doing something wrong and are trying to narrow down a
763problem.</p>
764
765</div>
766
Chris Lattner2a19d832004-05-23 21:04:01 +0000[diff] [blame]767<!-- _______________________________________________________________________ -->
NAKAMURA Takumifc8d9302011-04-18 23:59:50 +0000[diff] [blame]768<h4>
Chris Lattner2a19d832004-05-23 21:04:01 +0000[diff] [blame]769 <a name="basic-aa">The <tt>-basicaa</tt> pass</a>
NAKAMURA Takumifc8d9302011-04-18 23:59:50 +0000[diff] [blame]770</h4>
Chris Lattner2a19d832004-05-23 21:04:01 +0000[diff] [blame]771
NAKAMURA Takumi3ad28282011-04-21 01:52:00 +0000[diff] [blame]772<div>
Chris Lattner2a19d832004-05-23 21:04:01 +0000[diff] [blame]773
Dan Gohmanb83aed92010-11-15 18:07:16 +0000[diff] [blame]774<p>The <tt>-basicaa</tt> pass is an aggressive local analysis that "knows"
775many important facts:</p>
Chris Lattner2a19d832004-05-23 21:04:01 +0000[diff] [blame]776
777<ul>
778<li>Distinct globals, stack allocations, and heap allocations can never
779 alias.</li>
780<li>Globals, stack allocations, and heap allocations never alias the null
781 pointer.</li>
782<li>Different fields of a structure do not alias.</li>
783<li>Indexes into arrays with statically differing subscripts cannot alias.</li>
784<li>Many common standard C library functions <a
785 href="#simplemodref">never access memory or only read memory</a>.</li>
786<li>Pointers that obviously point to constant globals
787 "<tt>pointToConstantMemory</tt>".</li>
788<li>Function calls can not modify or references stack allocations if they never
789 escape from the function that allocates them (a common case for automatic
790 arrays).</li>
791</ul>
792
793</div>
794
Chris Lattner9475b962004-06-28 19:19:47 +0000[diff] [blame]795<!-- _______________________________________________________________________ -->
NAKAMURA Takumifc8d9302011-04-18 23:59:50 +0000[diff] [blame]796<h4>
Chris Lattner9475b962004-06-28 19:19:47 +0000[diff] [blame]797 <a name="globalsmodref">The <tt>-globalsmodref-aa</tt> pass</a>
NAKAMURA Takumifc8d9302011-04-18 23:59:50 +0000[diff] [blame]798</h4>
Chris Lattner9475b962004-06-28 19:19:47 +0000[diff] [blame]799
NAKAMURA Takumi3ad28282011-04-21 01:52:00 +0000[diff] [blame]800<div>
Chris Lattner9475b962004-06-28 19:19:47 +0000[diff] [blame]801
802<p>This pass implements a simple context-sensitive mod/ref and alias analysis
Chris Lattnerd1384372004-07-27 07:50:07 +0000[diff] [blame]803for internal global variables that don't "have their address taken". If a
804global does not have its address taken, the pass knows that no pointers alias
805the global. This pass also keeps track of functions that it knows never access
Chris Lattner6a6d3cc2009-04-25 21:11:37 +0000[diff] [blame]806memory or never read memory. This allows certain optimizations (e.g. GVN) to
Chris Lattnerd1384372004-07-27 07:50:07 +0000[diff] [blame]807eliminate call instructions entirely.
Chris Lattner9475b962004-06-28 19:19:47 +0000[diff] [blame]808</p>
809
810<p>The real power of this pass is that it provides context-sensitive mod/ref
811information for call instructions. This allows the optimizer to know that
812calls to a function do not clobber or read the value of the global, allowing
813loads and stores to be eliminated.</p>
814
815<p>Note that this pass is somewhat limited in its scope (only support
816non-address taken globals), but is very quick analysis.</p>
817</div>
Chris Lattner2a19d832004-05-23 21:04:01 +0000[diff] [blame]818
819<!-- _______________________________________________________________________ -->
NAKAMURA Takumifc8d9302011-04-18 23:59:50 +0000[diff] [blame]820<h4>
Chris Lattner2a19d832004-05-23 21:04:01 +0000[diff] [blame]821 <a name="steens-aa">The <tt>-steens-aa</tt> pass</a>
NAKAMURA Takumifc8d9302011-04-18 23:59:50 +0000[diff] [blame]822</h4>
Chris Lattner2a19d832004-05-23 21:04:01 +0000[diff] [blame]823
NAKAMURA Takumi3ad28282011-04-21 01:52:00 +0000[diff] [blame]824<div>
Chris Lattner2a19d832004-05-23 21:04:01 +0000[diff] [blame]825
826<p>The <tt>-steens-aa</tt> pass implements a variation on the well-known
827"Steensgaard's algorithm" for interprocedural alias analysis. Steensgaard's
828algorithm is a unification-based, flow-insensitive, context-insensitive, and
829field-insensitive alias analysis that is also very scalable (effectively linear
830time).</p>
831
832<p>The LLVM <tt>-steens-aa</tt> pass implements a "speculatively
833field-<b>sensitive</b>" version of Steensgaard's algorithm using the Data
834Structure Analysis framework. This gives it substantially more precision than
835the standard algorithm while maintaining excellent analysis scalability.</p>
836
Chris Lattner4b325282007-07-03 04:41:50 +0000[diff] [blame]837<p>Note that <tt>-steens-aa</tt> is available in the optional "poolalloc"
838module, it is not part of the LLVM core.</p>
839
Chris Lattner2a19d832004-05-23 21:04:01 +0000[diff] [blame]840</div>
841
842<!-- _______________________________________________________________________ -->
NAKAMURA Takumifc8d9302011-04-18 23:59:50 +0000[diff] [blame]843<h4>
Chris Lattner2a19d832004-05-23 21:04:01 +0000[diff] [blame]844 <a name="ds-aa">The <tt>-ds-aa</tt> pass</a>
NAKAMURA Takumifc8d9302011-04-18 23:59:50 +0000[diff] [blame]845</h4>
Chris Lattner2a19d832004-05-23 21:04:01 +0000[diff] [blame]846
NAKAMURA Takumi3ad28282011-04-21 01:52:00 +0000[diff] [blame]847<div>
Chris Lattner2a19d832004-05-23 21:04:01 +0000[diff] [blame]848
849<p>The <tt>-ds-aa</tt> pass implements the full Data Structure Analysis
850algorithm. Data Structure Analysis is a modular unification-based,
851flow-insensitive, context-<b>sensitive</b>, and speculatively
852field-<b>sensitive</b> alias analysis that is also quite scalable, usually at
853O(n*log(n)).</p>
854
855<p>This algorithm is capable of responding to a full variety of alias analysis
856queries, and can provide context-sensitive mod/ref information as well. The
857only major facility not implemented so far is support for must-alias
858information.</p>
Misha Brukman084a3a072003-10-24 18:06:11 +0000[diff] [blame]859
Chris Lattner4b325282007-07-03 04:41:50 +0000[diff] [blame]860<p>Note that <tt>-ds-aa</tt> is available in the optional "poolalloc"
861module, it is not part of the LLVM core.</p>
862
Misha Brukman283ff692003-10-22 23:27:16 +0000[diff] [blame]863</div>
Chris Lattnerb512e9f2003-03-04 19:37:49 +0000[diff] [blame]864
Dan Gohmanb160ba42010-06-28 22:09:52 +0000[diff] [blame]865<!-- _______________________________________________________________________ -->
NAKAMURA Takumifc8d9302011-04-18 23:59:50 +0000[diff] [blame]866<h4>
Dan Gohmanb160ba42010-06-28 22:09:52 +0000[diff] [blame]867 <a name="scev-aa">The <tt>-scev-aa</tt> pass</a>
NAKAMURA Takumifc8d9302011-04-18 23:59:50 +0000[diff] [blame]868</h4>
Dan Gohmanb160ba42010-06-28 22:09:52 +0000[diff] [blame]869
NAKAMURA Takumi3ad28282011-04-21 01:52:00 +0000[diff] [blame]870<div>
Dan Gohmanb160ba42010-06-28 22:09:52 +0000[diff] [blame]871
872<p>The <tt>-scev-aa</tt> pass implements AliasAnalysis queries by
873translating them into ScalarEvolution queries. This gives it a
874more complete understanding of <tt>getelementptr</tt> instructions
875and loop induction variables than other alias analyses have.</p>
876
877</div>
Chris Lattnerb512e9f2003-03-04 19:37:49 +0000[diff] [blame]878
NAKAMURA Takumi3ad28282011-04-21 01:52:00 +0000[diff] [blame]879</div>
880
Chris Lattnerb512e9f2003-03-04 19:37:49 +0000[diff] [blame]881<!-- ======================================================================= -->
NAKAMURA Takumifc8d9302011-04-18 23:59:50 +0000[diff] [blame]882<h3>
Chris Lattner2a19d832004-05-23 21:04:01 +0000[diff] [blame]883 <a name="aliasanalysis-xforms">Alias analysis driven transformations</a>
NAKAMURA Takumifc8d9302011-04-18 23:59:50 +0000[diff] [blame]884</h3>
Chris Lattner2a19d832004-05-23 21:04:01 +0000[diff] [blame]885
NAKAMURA Takumi3ad28282011-04-21 01:52:00 +0000[diff] [blame]886<div>
Chris Lattner2a19d832004-05-23 21:04:01 +0000[diff] [blame]887LLVM includes several alias-analysis driven transformations which can be used
888with any of the implementations above.
Chris Lattner2a19d832004-05-23 21:04:01 +0000[diff] [blame]889
890<!-- _______________________________________________________________________ -->
NAKAMURA Takumifc8d9302011-04-18 23:59:50 +0000[diff] [blame]891<h4>
Chris Lattner2a19d832004-05-23 21:04:01 +0000[diff] [blame]892 <a name="adce">The <tt>-adce</tt> pass</a>
NAKAMURA Takumifc8d9302011-04-18 23:59:50 +0000[diff] [blame]893</h4>
Chris Lattner2a19d832004-05-23 21:04:01 +0000[diff] [blame]894
NAKAMURA Takumi3ad28282011-04-21 01:52:00 +0000[diff] [blame]895<div>
Chris Lattner2a19d832004-05-23 21:04:01 +0000[diff] [blame]896
897<p>The <tt>-adce</tt> pass, which implements Aggressive Dead Code Elimination
898uses the <tt>AliasAnalysis</tt> interface to delete calls to functions that do
899not have side-effects and are not used.</p>
900
901</div>
902
903
904<!-- _______________________________________________________________________ -->
NAKAMURA Takumifc8d9302011-04-18 23:59:50 +0000[diff] [blame]905<h4>
Chris Lattner2a19d832004-05-23 21:04:01 +0000[diff] [blame]906 <a name="licm">The <tt>-licm</tt> pass</a>
NAKAMURA Takumifc8d9302011-04-18 23:59:50 +0000[diff] [blame]907</h4>
Chris Lattner2a19d832004-05-23 21:04:01 +0000[diff] [blame]908
NAKAMURA Takumi3ad28282011-04-21 01:52:00 +0000[diff] [blame]909<div>
Chris Lattner2a19d832004-05-23 21:04:01 +0000[diff] [blame]910
911<p>The <tt>-licm</tt> pass implements various Loop Invariant Code Motion related
912transformations. It uses the <tt>AliasAnalysis</tt> interface for several
913different transformations:</p>
914
915<ul>
916<li>It uses mod/ref information to hoist or sink load instructions out of loops
917if there are no instructions in the loop that modifies the memory loaded.</li>
918
919<li>It uses mod/ref information to hoist function calls out of loops that do not
920write to memory and are loop-invariant.</li>
921
922<li>If uses alias information to promote memory objects that are loaded and
923stored to in loops to live in a register instead. It can do this if there are
924no may aliases to the loaded/stored memory location.</li>
925</ul>
926
927</div>
928
929<!-- _______________________________________________________________________ -->
NAKAMURA Takumifc8d9302011-04-18 23:59:50 +0000[diff] [blame]930<h4>
Chris Lattner2a19d832004-05-23 21:04:01 +0000[diff] [blame]931 <a name="argpromotion">The <tt>-argpromotion</tt> pass</a>
NAKAMURA Takumifc8d9302011-04-18 23:59:50 +0000[diff] [blame]932</h4>
Chris Lattner2a19d832004-05-23 21:04:01 +0000[diff] [blame]933
NAKAMURA Takumi3ad28282011-04-21 01:52:00 +0000[diff] [blame]934<div>
Chris Lattner2a19d832004-05-23 21:04:01 +0000[diff] [blame]935<p>
936The <tt>-argpromotion</tt> pass promotes by-reference arguments to be passed in
937by-value instead. In particular, if pointer arguments are only loaded from it
938passes in the value loaded instead of the address to the function. This pass
939uses alias information to make sure that the value loaded from the argument
940pointer is not modified between the entry of the function and any load of the
941pointer.</p>
942</div>
943
944<!-- _______________________________________________________________________ -->
NAKAMURA Takumifc8d9302011-04-18 23:59:50 +0000[diff] [blame]945<h4>
Chris Lattner6a6d3cc2009-04-25 21:11:37 +0000[diff] [blame]946 <a name="gvn">The <tt>-gvn</tt>, <tt>-memcpyopt</tt>, and <tt>-dse</tt>
947 passes</a>
NAKAMURA Takumifc8d9302011-04-18 23:59:50 +0000[diff] [blame]948</h4>
Chris Lattner2a19d832004-05-23 21:04:01 +0000[diff] [blame]949
NAKAMURA Takumi3ad28282011-04-21 01:52:00 +0000[diff] [blame]950<div>
Misha Brukman2361fcf2004-07-28 22:18:33 +0000[diff] [blame]951
Chris Lattner6a6d3cc2009-04-25 21:11:37 +0000[diff] [blame]952<p>These passes use AliasAnalysis information to reason about loads and stores.
953</p>
Chris Lattner2a19d832004-05-23 21:04:01 +0000[diff] [blame]954
955</div>
956
NAKAMURA Takumi3ad28282011-04-21 01:52:00 +0000[diff] [blame]957</div>
958
Chris Lattner2a19d832004-05-23 21:04:01 +0000[diff] [blame]959<!-- ======================================================================= -->
NAKAMURA Takumifc8d9302011-04-18 23:59:50 +0000[diff] [blame]960<h3>
Misha Brukman2361fcf2004-07-28 22:18:33 +0000[diff] [blame]961 <a name="aliasanalysis-debug">Clients for debugging and evaluation of
962 implementations</a>
NAKAMURA Takumifc8d9302011-04-18 23:59:50 +0000[diff] [blame]963</h3>
Chris Lattner2a19d832004-05-23 21:04:01 +0000[diff] [blame]964
NAKAMURA Takumi3ad28282011-04-21 01:52:00 +0000[diff] [blame]965<div>
Chris Lattner2a19d832004-05-23 21:04:01 +0000[diff] [blame]966
Misha Brukman2361fcf2004-07-28 22:18:33 +0000[diff] [blame]967<p>These passes are useful for evaluating the various alias analysis
Chris Lattner5ea3e652010-03-01 19:24:17 +0000[diff] [blame]968implementations. You can use them with commands like '<tt>opt -ds-aa
Misha Brukman2361fcf2004-07-28 22:18:33 +0000[diff] [blame]969-aa-eval foo.bc -disable-output -stats</tt>'.</p>
970
Chris Lattner2a19d832004-05-23 21:04:01 +0000[diff] [blame]971<!-- _______________________________________________________________________ -->
NAKAMURA Takumifc8d9302011-04-18 23:59:50 +0000[diff] [blame]972<h4>
Misha Brukman283ff692003-10-22 23:27:16 +0000[diff] [blame]973 <a name="print-alias-sets">The <tt>-print-alias-sets</tt> pass</a>
NAKAMURA Takumifc8d9302011-04-18 23:59:50 +0000[diff] [blame]974</h4>
Chris Lattnerb512e9f2003-03-04 19:37:49 +0000[diff] [blame]975
NAKAMURA Takumi3ad28282011-04-21 01:52:00 +0000[diff] [blame]976<div>
Misha Brukman084a3a072003-10-24 18:06:11 +0000[diff] [blame]977
978<p>The <tt>-print-alias-sets</tt> pass is exposed as part of the
Chris Lattner2e6f4482006-01-03 06:04:48 +0000[diff] [blame]979<tt>opt</tt> tool to print out the Alias Sets formed by the <a
Chris Lattnerb512e9f2003-03-04 19:37:49 +0000[diff] [blame]980href="#ast"><tt>AliasSetTracker</tt></a> class. This is useful if you're using
Chris Lattner2e6f4482006-01-03 06:04:48 +0000[diff] [blame]981the <tt>AliasSetTracker</tt> class. To use it, use something like:</p>
982
983<div class="doc_code">
984<pre>
985% opt -ds-aa -print-alias-sets -disable-output
986</pre>
987</div>
Misha Brukman084a3a072003-10-24 18:06:11 +0000[diff] [blame]988
Misha Brukman283ff692003-10-22 23:27:16 +0000[diff] [blame]989</div>
Chris Lattnerb512e9f2003-03-04 19:37:49 +0000[diff] [blame]990
Chris Lattner2a19d832004-05-23 21:04:01 +0000[diff] [blame]991
992<!-- _______________________________________________________________________ -->
NAKAMURA Takumifc8d9302011-04-18 23:59:50 +0000[diff] [blame]993<h4>
Misha Brukman283ff692003-10-22 23:27:16 +0000[diff] [blame]994 <a name="count-aa">The <tt>-count-aa</tt> pass</a>
NAKAMURA Takumifc8d9302011-04-18 23:59:50 +0000[diff] [blame]995</h4>
Chris Lattnerb512e9f2003-03-04 19:37:49 +0000[diff] [blame]996
NAKAMURA Takumi3ad28282011-04-21 01:52:00 +0000[diff] [blame]997<div>
Misha Brukman084a3a072003-10-24 18:06:11 +0000[diff] [blame]998
999<p>The <tt>-count-aa</tt> pass is useful to see how many queries a particular
Misha Brukman2361fcf2004-07-28 22:18:33 +0000[diff] [blame]1000pass is making and what responses are returned by the alias analysis. As an
1001example,</p>
Misha Brukman084a3a072003-10-24 18:06:11 +0000[diff] [blame]1002
Misha Brukman2361fcf2004-07-28 22:18:33 +0000[diff] [blame]1003<div class="doc_code">
Chris Lattnerb512e9f2003-03-04 19:37:49 +0000[diff] [blame]1004<pre>
Misha Brukman2361fcf2004-07-28 22:18:33 +0000[diff] [blame]1005% opt -basicaa -count-aa -ds-aa -count-aa -licm
Chris Lattnerb512e9f2003-03-04 19:37:49 +0000[diff] [blame]1006</pre>
Misha Brukman2361fcf2004-07-28 22:18:33 +0000[diff] [blame]1007</div>
Misha Brukman084a3a072003-10-24 18:06:11 +0000[diff] [blame]1008
Misha Brukman2361fcf2004-07-28 22:18:33 +0000[diff] [blame]1009<p>will print out how many queries (and what responses are returned) by the
1010<tt>-licm</tt> pass (of the <tt>-ds-aa</tt> pass) and how many queries are made
1011of the <tt>-basicaa</tt> pass by the <tt>-ds-aa</tt> pass. This can be useful
1012when debugging a transformation or an alias analysis implementation.</p>
Misha Brukman084a3a072003-10-24 18:06:11 +0000[diff] [blame]1013
Misha Brukman283ff692003-10-22 23:27:16 +0000[diff] [blame]1014</div>
Chris Lattnerb512e9f2003-03-04 19:37:49 +0000[diff] [blame]1015
Chris Lattner2a19d832004-05-23 21:04:01 +0000[diff] [blame]1016<!-- _______________________________________________________________________ -->
NAKAMURA Takumifc8d9302011-04-18 23:59:50 +0000[diff] [blame]1017<h4>
Misha Brukman283ff692003-10-22 23:27:16 +0000[diff] [blame]1018 <a name="aa-eval">The <tt>-aa-eval</tt> pass</a>
NAKAMURA Takumifc8d9302011-04-18 23:59:50 +0000[diff] [blame]1019</h4>
Chris Lattnerb512e9f2003-03-04 19:37:49 +0000[diff] [blame]1020
NAKAMURA Takumi3ad28282011-04-21 01:52:00 +0000[diff] [blame]1021<div>
Misha Brukman084a3a072003-10-24 18:06:11 +0000[diff] [blame]1022
1023<p>The <tt>-aa-eval</tt> pass simply iterates through all pairs of pointers in a
Chris Lattnerb512e9f2003-03-04 19:37:49 +0000[diff] [blame]1024function and asks an alias analysis whether or not the pointers alias. This
1025gives an indication of the precision of the alias analysis. Statistics are
Chris Lattner2a19d832004-05-23 21:04:01 +0000[diff] [blame]1026printed indicating the percent of no/may/must aliases found (a more precise
1027algorithm will have a lower number of may aliases).</p>
Misha Brukman084a3a072003-10-24 18:06:11 +0000[diff] [blame]1028
Misha Brukman283ff692003-10-22 23:27:16 +0000[diff] [blame]1029</div>
Chris Lattnerb512e9f2003-03-04 19:37:49 +0000[diff] [blame]1030
NAKAMURA Takumi3ad28282011-04-21 01:52:00 +0000[diff] [blame]1031</div>
1032
1033</div>
1034
Chris Lattnerb512e9f2003-03-04 19:37:49 +0000[diff] [blame]1035<!-- *********************************************************************** -->
NAKAMURA Takumifc8d9302011-04-18 23:59:50 +0000[diff] [blame]1036<h2>
Owen Andersond263cfa2007-10-02 00:43:25 +0000[diff] [blame]1037 <a name="memdep">Memory Dependence Analysis</a>
NAKAMURA Takumifc8d9302011-04-18 23:59:50 +0000[diff] [blame]1038</h2>
Owen Andersond263cfa2007-10-02 00:43:25 +0000[diff] [blame]1039<!-- *********************************************************************** -->
1040
NAKAMURA Takumi3ad28282011-04-21 01:52:00 +0000[diff] [blame]1041<div>
Owen Andersond263cfa2007-10-02 00:43:25 +0000[diff] [blame]1042
1043<p>If you're just looking to be a client of alias analysis information, consider
1044using the Memory Dependence Analysis interface instead. MemDep is a lazy,
1045caching layer on top of alias analysis that is able to answer the question of
1046what preceding memory operations a given instruction depends on, either at an
1047intra- or inter-block level. Because of its laziness and caching
1048policy, using MemDep can be a significant performance win over accessing alias
1049analysis directly.</p>
1050
1051</div>
1052
1053<!-- *********************************************************************** -->
Chris Lattnerb512e9f2003-03-04 19:37:49 +0000[diff] [blame]1054
Misha Brukman084a3a072003-10-24 18:06:11 +0000[diff] [blame]1055<hr>
Misha Brukmane8e932a2003-11-22 01:26:21 +0000[diff] [blame]1056<address>
Misha Brukman8834af92003-12-21 22:53:21 +0000[diff] [blame]1057 <a href="http://jigsaw.w3.org/css-validator/check/referer"><img
Misha Brukman86242e12008-12-11 17:34:48 +0000[diff] [blame]1058 src="http://jigsaw.w3.org/css-validator/images/vcss-blue" alt="Valid CSS"></a>
Misha Brukmane8e932a2003-11-22 01:26:21 +0000[diff] [blame]1059 <a href="http://validator.w3.org/check/referer"><img
Misha Brukman86242e12008-12-11 17:34:48 +0000[diff] [blame]1060 src="http://www.w3.org/Icons/valid-html401-blue" alt="Valid HTML 4.01"></a>
Misha Brukmane8e932a2003-11-22 01:26:21 +0000[diff] [blame]1061
1062 <a href="mailto:sabre@nondot.org">Chris Lattner</a><br>
NAKAMURA Takumica46f5a2011-04-09 02:13:37 +0000[diff] [blame]1063 <a href="http://llvm.org/">LLVM Compiler Infrastructure</a><br>
Misha Brukmane8e932a2003-11-22 01:26:21 +0000[diff] [blame]1064 Last modified: $Date$
1065</address>
Misha Brukman283ff692003-10-22 23:27:16 +0000[diff] [blame]1066
1067</body>
1068</html>