Gitiles
Code Review Sign In
gerrit-public.fairphone.software / fp2-dev / platform / external / llvm / 43b8fd728b889f15ea4c65ca957a7420ce2905cd / . / docs / AliasAnalysis.html
blob: 8e6a7fd4d7c606fba2ed5320227b0dc51e942b14 [file] [log] [blame]
Misha Brukman3896be22003-10-24 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 Brukmancd11e452003-10-22 23:27:16 +0000[diff] [blame]3<html>
4<head>
Misha Brukmana97e6cb2004-07-01 15:33:24 +0000[diff] [blame]5 <title>LLVM Alias Analysis Infrastructure</title>
Misha Brukman6e4afd02004-01-15 19:04:12 +0000[diff] [blame]6 <link rel="stylesheet" href="llvm.css" type="text/css">
Misha Brukmancd11e452003-10-22 23:27:16 +0000[diff] [blame]7</head>
Misha Brukmancd11e452003-10-22 23:27:16 +0000[diff] [blame]8<body>
Chris Lattner9f648752003-03-04 19:37:49 +0000[diff] [blame]9
Misha Brukmancd11e452003-10-22 23:27:16 +0000[diff] [blame]10<div class="doc_title">
Misha Brukmana97e6cb2004-07-01 15:33:24 +0000[diff] [blame]11 LLVM Alias Analysis Infrastructure
Misha Brukmancd11e452003-10-22 23:27:16 +0000[diff] [blame]12</div>
Chris Lattner9f648752003-03-04 19:37:49 +0000[diff] [blame]13
14<ol>
Misha Brukmanb2154252003-10-23 02:29:42 +0000[diff] [blame]15 <li><a href="#introduction">Introduction</a></li>
Chris Lattner9f648752003-03-04 19:37:49 +0000[diff] [blame]16
Chris Lattnerccc80652004-05-23 21:04:01 +0000[diff] [blame]17 <li><a href="#overview"><tt>AliasAnalysis</tt> Class Overview</a>
Chris Lattner9f648752003-03-04 19:37:49 +0000[diff] [blame]18 <ul>
Misha Brukmanb2154252003-10-23 02:29:42 +0000[diff] [blame]19 <li><a href="#pointers">Representation of Pointers</a></li>
Chris Lattnerccc80652004-05-23 21:04:01 +0000[diff] [blame]20 <li><a href="#alias">The <tt>alias</tt> method</a></li>
Misha Brukmanb2154252003-10-23 02:29:42 +0000[diff] [blame]21 <li><a href="#ModRefInfo">The <tt>getModRefInfo</tt> methods</a></li>
Chris Lattnerccc80652004-05-23 21:04:01 +0000[diff] [blame]22 <li><a href="#OtherItfs">Other useful <tt>AliasAnalysis</tt> methods</a></li>
23 </ul>
24 </li>
Chris Lattner9f648752003-03-04 19:37:49 +0000[diff] [blame]25
Chris Lattnerccc80652004-05-23 21:04:01 +0000[diff] [blame]26 <li><a href="#writingnew">Writing a new <tt>AliasAnalysis</tt> Implementation</a>
Chris Lattner9f648752003-03-04 19:37:49 +0000[diff] [blame]27 <ul>
Misha Brukmanb2154252003-10-23 02:29:42 +0000[diff] [blame]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>
Chris Lattnerccc80652004-05-23 21:04:01 +0000[diff] [blame]31 <li><a href="#chaining"><tt>AliasAnalysis</tt> chaining behavior</a></li>
32 <li><a href="#updating">Updating analysis results for transformations</a></li>
Misha Brukmanb2154252003-10-23 02:29:42 +0000[diff] [blame]33 <li><a href="#implefficiency">Efficiency Issues</a></li>
Dan Gohman3f43dc32010-06-24 19:34:03 +0000[diff] [blame]34 <li><a href="#passmanager">Pass Manager Issues</a></li>
Chris Lattnerccc80652004-05-23 21:04:01 +0000[diff] [blame]35 </ul>
36 </li>
Chris Lattner9f648752003-03-04 19:37:49 +0000[diff] [blame]37
Chris Lattnerccc80652004-05-23 21:04:01 +0000[diff] [blame]38 <li><a href="#using">Using alias analysis results</a>
Chris Lattner9f648752003-03-04 19:37:49 +0000[diff] [blame]39 <ul>
Chris Lattner4f4365e2009-04-25 21:11:37 +0000[diff] [blame]40 <li><a href="#memdep">Using the <tt>MemoryDependenceAnalysis</tt> Pass</a></li>
Misha Brukmanb2154252003-10-23 02:29:42 +0000[diff] [blame]41 <li><a href="#ast">Using the <tt>AliasSetTracker</tt> class</a></li>
Chris Lattnerccc80652004-05-23 21:04:01 +0000[diff] [blame]42 <li><a href="#direct">Using the <tt>AliasAnalysis</tt> interface directly</a></li>
43 </ul>
44 </li>
Misha Brukman3896be22003-10-24 18:06:11 +0000[diff] [blame]45
Chris Lattnerccc80652004-05-23 21:04:01 +0000[diff] [blame]46 <li><a href="#exist">Existing alias analysis implementations and clients</a>
Chris Lattner9f648752003-03-04 19:37:49 +0000[diff] [blame]47 <ul>
Chris Lattnerccc80652004-05-23 21:04:01 +0000[diff] [blame]48 <li><a href="#impls">Available <tt>AliasAnalysis</tt> implementations</a></li>
49 <li><a href="#aliasanalysis-xforms">Alias analysis driven transformations</a></li>
Misha Brukmana97e6cb2004-07-01 15:33:24 +0000[diff] [blame]50 <li><a href="#aliasanalysis-debug">Clients for debugging and evaluation of
51 implementations</a></li>
Chris Lattnerccc80652004-05-23 21:04:01 +0000[diff] [blame]52 </ul>
53 </li>
Owen Andersond8b47282007-10-02 00:44:20 +0000[diff] [blame]54 <li><a href="#memdep">Memory Dependence Analysis</a></li>
Misha Brukmanb2154252003-10-23 02:29:42 +0000[diff] [blame]55</ol>
Chris Lattner9f648752003-03-04 19:37:49 +0000[diff] [blame]56
Chris Lattnerccc80652004-05-23 21:04:01 +0000[diff] [blame]57<div class="doc_author">
58 <p>Written by <a href="mailto:sabre@nondot.org">Chris Lattner</a></p>
Misha Brukman3896be22003-10-24 18:06:11 +0000[diff] [blame]59</div>
Chris Lattner9f648752003-03-04 19:37:49 +0000[diff] [blame]60
Chris Lattner9f648752003-03-04 19:37:49 +0000[diff] [blame]61<!-- *********************************************************************** -->
Misha Brukmancd11e452003-10-22 23:27:16 +0000[diff] [blame]62<div class="doc_section">
63 <a name="introduction">Introduction</a>
64</div>
Chris Lattner9f648752003-03-04 19:37:49 +0000[diff] [blame]65<!-- *********************************************************************** -->
66
Misha Brukmancd11e452003-10-22 23:27:16 +0000[diff] [blame]67<div class="doc_text">
Misha Brukman1f254d52003-11-21 22:30:25 +0000[diff] [blame]68
Chris Lattnerccc80652004-05-23 21:04:01 +0000[diff] [blame]69<p>Alias Analysis (aka Pointer Analysis) is a class of techniques which attempt
70to determine whether or not two pointers ever can point to the same object in
71memory. There are many different algorithms for alias analysis and many
72different ways of classifying them: flow-sensitive vs flow-insensitive,
73context-sensitive vs context-insensitive, field-sensitive vs field-insensitive,
74unification-based vs subset-based, etc. Traditionally, alias analyses respond
Duncan Sands8036ca42007-03-30 12:22:09 +0000[diff] [blame]75to a query with a <a href="#MustMayNo">Must, May, or No</a> alias response,
Chris Lattnerccc80652004-05-23 21:04:01 +0000[diff] [blame]76indicating that two pointers always point to the same object, might point to the
77same object, or are known to never point to the same object.</p>
Misha Brukman1f254d52003-11-21 22:30:25 +0000[diff] [blame]78
Chris Lattnerccc80652004-05-23 21:04:01 +0000[diff] [blame]79<p>The LLVM <a
Reid Spencer05fe4b02006-03-14 05:39:39 +0000[diff] [blame]80href="http://llvm.org/doxygen/classllvm_1_1AliasAnalysis.html"><tt>AliasAnalysis</tt></a>
Chris Lattnerccc80652004-05-23 21:04:01 +0000[diff] [blame]81class is the primary interface used by clients and implementations of alias
82analyses in the LLVM system. This class is the common interface between clients
83of alias analysis information and the implementations providing it, and is
84designed to support a wide range of implementations and clients (but currently
85all clients are assumed to be flow-insensitive). In addition to simple alias
86analysis information, this class exposes Mod/Ref information from those
87implementations which can provide it, allowing for powerful analyses and
88transformations to work well together.</p>
Misha Brukman1f254d52003-11-21 22:30:25 +0000[diff] [blame]89
90<p>This document contains information necessary to successfully implement this
Chris Lattner9f648752003-03-04 19:37:49 +0000[diff] [blame]91interface, use it, and to test both sides. It also explains some of the finer
92points about what exactly results mean. If you feel that something is unclear
Misha Brukmancd11e452003-10-22 23:27:16 +0000[diff] [blame]93or should be added, please <a href="mailto:sabre@nondot.org">let me
Misha Brukman1f254d52003-11-21 22:30:25 +0000[diff] [blame]94know</a>.</p>
95
Misha Brukmancd11e452003-10-22 23:27:16 +0000[diff] [blame]96</div>
Chris Lattner9f648752003-03-04 19:37:49 +0000[diff] [blame]97
98<!-- *********************************************************************** -->
Misha Brukmancd11e452003-10-22 23:27:16 +0000[diff] [blame]99<div class="doc_section">
Chris Lattnerccc80652004-05-23 21:04:01 +0000[diff] [blame]100 <a name="overview"><tt>AliasAnalysis</tt> Class Overview</a>
Misha Brukmancd11e452003-10-22 23:27:16 +0000[diff] [blame]101</div>
Chris Lattner9f648752003-03-04 19:37:49 +0000[diff] [blame]102<!-- *********************************************************************** -->
103
Misha Brukmancd11e452003-10-22 23:27:16 +0000[diff] [blame]104<div class="doc_text">
Misha Brukman1f254d52003-11-21 22:30:25 +0000[diff] [blame]105
Chris Lattnerccc80652004-05-23 21:04:01 +0000[diff] [blame]106<p>The <a
Reid Spencer05fe4b02006-03-14 05:39:39 +0000[diff] [blame]107href="http://llvm.org/doxygen/classllvm_1_1AliasAnalysis.html"><tt>AliasAnalysis</tt></a>
Chris Lattnerccc80652004-05-23 21:04:01 +0000[diff] [blame]108class defines the interface that the various alias analysis implementations
109should support. This class exports two important enums: <tt>AliasResult</tt>
110and <tt>ModRefResult</tt> which represent the result of an alias query or a
111mod/ref query, respectively.</p>
Misha Brukman1f254d52003-11-21 22:30:25 +0000[diff] [blame]112
Chris Lattnerccc80652004-05-23 21:04:01 +0000[diff] [blame]113<p>The <tt>AliasAnalysis</tt> interface exposes information about memory,
114represented in several different ways. In particular, memory objects are
115represented as a starting address and size, and function calls are represented
116as the actual <tt>call</tt> or <tt>invoke</tt> instructions that performs the
117call. The <tt>AliasAnalysis</tt> interface also exposes some helper methods
118which allow you to get mod/ref information for arbitrary instructions.</p>
Misha Brukman1f254d52003-11-21 22:30:25 +0000[diff] [blame]119
Misha Brukmancd11e452003-10-22 23:27:16 +0000[diff] [blame]120</div>
Chris Lattner9f648752003-03-04 19:37:49 +0000[diff] [blame]121
122<!-- ======================================================================= -->
Misha Brukmancd11e452003-10-22 23:27:16 +0000[diff] [blame]123<div class="doc_subsection">
124 <a name="pointers">Representation of Pointers</a>
125</div>
Chris Lattner9f648752003-03-04 19:37:49 +0000[diff] [blame]126
Misha Brukmancd11e452003-10-22 23:27:16 +0000[diff] [blame]127<div class="doc_text">
Misha Brukman3896be22003-10-24 18:06:11 +0000[diff] [blame]128
Chris Lattnerccc80652004-05-23 21:04:01 +0000[diff] [blame]129<p>Most importantly, the <tt>AliasAnalysis</tt> class provides several methods
130which are used to query whether or not two memory objects alias, whether
131function calls can modify or read a memory object, etc. For all of these
132queries, memory objects are represented as a pair of their starting address (a
133symbolic LLVM <tt>Value*</tt>) and a static size.</p>
Misha Brukman3896be22003-10-24 18:06:11 +0000[diff] [blame]134
135<p>Representing memory objects as a starting address and a size is critically
Chris Lattnerccc80652004-05-23 21:04:01 +0000[diff] [blame]136important for correct Alias Analyses. For example, consider this (silly, but
137possible) C code:</p>
Misha Brukman3896be22003-10-24 18:06:11 +0000[diff] [blame]138
Misha Brukman294611a2004-07-28 22:18:33 +0000[diff] [blame]139<div class="doc_code">
Chris Lattner9f648752003-03-04 19:37:49 +0000[diff] [blame]140<pre>
Misha Brukman294611a2004-07-28 22:18:33 +0000[diff] [blame]141int i;
142char C[2];
143char A[10];
144/* ... */
145for (i = 0; i != 10; ++i) {
146 C[0] = A[i]; /* One byte store */
147 C[1] = A[9-i]; /* One byte store */
148}
Chris Lattner9f648752003-03-04 19:37:49 +0000[diff] [blame]149</pre>
Misha Brukman294611a2004-07-28 22:18:33 +0000[diff] [blame]150</div>
Misha Brukman3896be22003-10-24 18:06:11 +0000[diff] [blame]151
152<p>In this case, the <tt>basicaa</tt> pass will disambiguate the stores to
Chris Lattner9f648752003-03-04 19:37:49 +0000[diff] [blame]153<tt>C[0]</tt> and <tt>C[1]</tt> because they are accesses to two distinct
154locations one byte apart, and the accesses are each one byte. In this case, the
155LICM pass can use store motion to remove the stores from the loop. In
Misha Brukman3896be22003-10-24 18:06:11 +0000[diff] [blame]156constrast, the following code:</p>
157
Misha Brukman294611a2004-07-28 22:18:33 +0000[diff] [blame]158<div class="doc_code">
Chris Lattner9f648752003-03-04 19:37:49 +0000[diff] [blame]159<pre>
Misha Brukman294611a2004-07-28 22:18:33 +0000[diff] [blame]160int i;
161char C[2];
162char A[10];
163/* ... */
164for (i = 0; i != 10; ++i) {
165 ((short*)C)[0] = A[i]; /* Two byte store! */
166 C[1] = A[9-i]; /* One byte store */
167}
Chris Lattner9f648752003-03-04 19:37:49 +0000[diff] [blame]168</pre>
Misha Brukman294611a2004-07-28 22:18:33 +0000[diff] [blame]169</div>
Misha Brukman3896be22003-10-24 18:06:11 +0000[diff] [blame]170
171<p>In this case, the two stores to C do alias each other, because the access to
172the <tt>&amp;C[0]</tt> element is a two byte access. If size information wasn't
Chris Lattner9f648752003-03-04 19:37:49 +0000[diff] [blame]173available in the query, even the first case would have to conservatively assume
Misha Brukman3896be22003-10-24 18:06:11 +0000[diff] [blame]174that the accesses alias.</p>
175
Misha Brukmancd11e452003-10-22 23:27:16 +0000[diff] [blame]176</div>
Chris Lattner9f648752003-03-04 19:37:49 +0000[diff] [blame]177
178<!-- ======================================================================= -->
Misha Brukmancd11e452003-10-22 23:27:16 +0000[diff] [blame]179<div class="doc_subsection">
Chris Lattnerccc80652004-05-23 21:04:01 +0000[diff] [blame]180 <a name="alias">The <tt>alias</tt> method</a>
181</div>
182
183<div class="doc_text">
184The <tt>alias</tt> method is the primary interface used to determine whether or
185not two memory objects alias each other. It takes two memory objects as input
186and returns MustAlias, MayAlias, or NoAlias as appropriate.
187</div>
188
189<!-- _______________________________________________________________________ -->
190<div class="doc_subsubsection">
Misha Brukmancd11e452003-10-22 23:27:16 +0000[diff] [blame]191 <a name="MustMayNo">Must, May, and No Alias Responses</a>
192</div>
Chris Lattner9f648752003-03-04 19:37:49 +0000[diff] [blame]193
Misha Brukmancd11e452003-10-22 23:27:16 +0000[diff] [blame]194<div class="doc_text">
Dan Gohman1e109622010-07-02 18:41:32 +0000[diff] [blame]195<p>The NoAlias response may be used when there is never an immediate dependence
196between any memory reference <i>based</i> on one pointer and any memory
197reference <i>based</i> the other. The most obvious example is when the two
198pointers point to non-overlapping memory ranges. Another is when the two
199pointers are only ever used for reading memory. Another is when the memory is
200freed and reallocated between accesses through one pointer and accesses through
201the other -- in this case, there is a dependence, but it's mediated by the free
202and reallocation.</p>
Misha Brukman3896be22003-10-24 18:06:11 +0000[diff] [blame]203
Dan Gohmanefca7f92010-07-02 23:46:54 +0000[diff] [blame]204<p>As an exception to this is with the
205<a href="LangRef.html#noalias"><tt>noalias</tt></a> keyword. AliasAnalysis
Dan Gohman43b8fd72010-07-02 23:52:36 +0000[diff] [blame^]206implementations may choose to respect the <tt>noalias</tt> keyword and ignore
207the "irrelevant" dependencies, provided their clients do not need to be aware
208of these dependencies for correctness.</p>
Dan Gohmanefca7f92010-07-02 23:46:54 +0000[diff] [blame]209
Nick Lewycky01557ce2008-12-14 21:08:48 +0000[diff] [blame]210<p>The MayAlias response is used whenever the two pointers might refer to the
211same object. If the two memory objects overlap, but do not start at the same
212location, return MayAlias.</p>
Misha Brukman3896be22003-10-24 18:06:11 +0000[diff] [blame]213
Nick Lewycky01557ce2008-12-14 21:08:48 +0000[diff] [blame]214<p>The MustAlias response may only be returned if the two memory objects are
215guaranteed to always start at exactly the same location. A MustAlias response
216implies that the pointers compare equal.</p>
Misha Brukman3896be22003-10-24 18:06:11 +0000[diff] [blame]217
Misha Brukmancd11e452003-10-22 23:27:16 +0000[diff] [blame]218</div>
Chris Lattner9f648752003-03-04 19:37:49 +0000[diff] [blame]219
220<!-- ======================================================================= -->
Misha Brukmancd11e452003-10-22 23:27:16 +0000[diff] [blame]221<div class="doc_subsection">
222 <a name="ModRefInfo">The <tt>getModRefInfo</tt> methods</a>
223</div>
Chris Lattner9f648752003-03-04 19:37:49 +0000[diff] [blame]224
Misha Brukmancd11e452003-10-22 23:27:16 +0000[diff] [blame]225<div class="doc_text">
Misha Brukman3896be22003-10-24 18:06:11 +0000[diff] [blame]226
227<p>The <tt>getModRefInfo</tt> methods return information about whether the
Chris Lattner9f648752003-03-04 19:37:49 +0000[diff] [blame]228execution of an instruction can read or modify a memory location. Mod/Ref
Chris Lattnerccc80652004-05-23 21:04:01 +0000[diff] [blame]229information is always conservative: if an instruction <b>might</b> read or write
230a location, ModRef is returned.</p>
231
232<p>The <tt>AliasAnalysis</tt> class also provides a <tt>getModRefInfo</tt>
233method for testing dependencies between function calls. This method takes two
234call sites (CS1 &amp; CS2), returns NoModRef if the two calls refer to disjoint
235memory locations, Ref if CS1 reads memory written by CS2, Mod if CS1 writes to
236memory read or written by CS2, or ModRef if CS1 might read or write memory
Chris Lattner66e08cf2009-11-22 16:01:44 +0000[diff] [blame]237accessed by CS2. Note that this relation is not commutative.</p>
Misha Brukman3896be22003-10-24 18:06:11 +0000[diff] [blame]238
Misha Brukmancd11e452003-10-22 23:27:16 +0000[diff] [blame]239</div>
Chris Lattner9f648752003-03-04 19:37:49 +0000[diff] [blame]240
Chris Lattnerccc80652004-05-23 21:04:01 +0000[diff] [blame]241
242<!-- ======================================================================= -->
243<div class="doc_subsection">
244 <a name="OtherItfs">Other useful <tt>AliasAnalysis</tt> methods</a>
245</div>
246
247<div class="doc_text">
248
249<p>
250Several other tidbits of information are often collected by various alias
251analysis implementations and can be put to good use by various clients.
252</p>
253
254</div>
255
256<!-- _______________________________________________________________________ -->
257<div class="doc_subsubsection">
Chris Lattnerccc80652004-05-23 21:04:01 +0000[diff] [blame]258 The <tt>pointsToConstantMemory</tt> method
259</div>
260
261<div class="doc_text">
262
263<p>The <tt>pointsToConstantMemory</tt> method returns true if and only if the
264analysis can prove that the pointer only points to unchanging memory locations
265(functions, constant global variables, and the null pointer). This information
266can be used to refine mod/ref information: it is impossible for an unchanging
267memory location to be modified.</p>
268
269</div>
270
Chris Lattnerccc80652004-05-23 21:04:01 +0000[diff] [blame]271<!-- _______________________________________________________________________ -->
272<div class="doc_subsubsection">
273 <a name="simplemodref">The <tt>doesNotAccessMemory</tt> and
274 <tt>onlyReadsMemory</tt> methods</a>
275</div>
276
277<div class="doc_text">
278
279<p>These methods are used to provide very simple mod/ref information for
280function calls. The <tt>doesNotAccessMemory</tt> method returns true for a
281function if the analysis can prove that the function never reads or writes to
282memory, or if the function only reads from constant memory. Functions with this
283property are side-effect free and only depend on their input arguments, allowing
284them to be eliminated if they form common subexpressions or be hoisted out of
285loops. Many common functions behave this way (e.g., <tt>sin</tt> and
286<tt>cos</tt>) but many others do not (e.g., <tt>acos</tt>, which modifies the
287<tt>errno</tt> variable).</p>
288
289<p>The <tt>onlyReadsMemory</tt> method returns true for a function if analysis
290can prove that (at most) the function only reads from non-volatile memory.
291Functions with this property are side-effect free, only depending on their input
292arguments and the state of memory when they are called. This property allows
293calls to these functions to be eliminated and moved around, as long as there is
294no store instruction that changes the contents of memory. Note that all
295functions that satisfy the <tt>doesNotAccessMemory</tt> method also satisfies
296<tt>onlyReadsMemory</tt>.</p>
297
298</div>
299
Chris Lattner9f648752003-03-04 19:37:49 +0000[diff] [blame]300<!-- *********************************************************************** -->
Misha Brukmancd11e452003-10-22 23:27:16 +0000[diff] [blame]301<div class="doc_section">
Chris Lattnerccc80652004-05-23 21:04:01 +0000[diff] [blame]302 <a name="writingnew">Writing a new <tt>AliasAnalysis</tt> Implementation</a>
Misha Brukmancd11e452003-10-22 23:27:16 +0000[diff] [blame]303</div>
Chris Lattner9f648752003-03-04 19:37:49 +0000[diff] [blame]304<!-- *********************************************************************** -->
305
Misha Brukmancd11e452003-10-22 23:27:16 +0000[diff] [blame]306<div class="doc_text">
Misha Brukman3896be22003-10-24 18:06:11 +0000[diff] [blame]307
308<p>Writing a new alias analysis implementation for LLVM is quite
309straight-forward. There are already several implementations that you can use
310for examples, and the following information should help fill in any details.
Chris Lattnerccc80652004-05-23 21:04:01 +0000[diff] [blame]311For a examples, take a look at the <a href="#impls">various alias analysis
312implementations</a> included with LLVM.</p>
Misha Brukman3896be22003-10-24 18:06:11 +0000[diff] [blame]313
Misha Brukmancd11e452003-10-22 23:27:16 +0000[diff] [blame]314</div>
Chris Lattner9f648752003-03-04 19:37:49 +0000[diff] [blame]315
316<!-- ======================================================================= -->
Misha Brukmancd11e452003-10-22 23:27:16 +0000[diff] [blame]317<div class="doc_subsection">
318 <a name="passsubclasses">Different Pass styles</a>
319</div>
Chris Lattner9f648752003-03-04 19:37:49 +0000[diff] [blame]320
Misha Brukmancd11e452003-10-22 23:27:16 +0000[diff] [blame]321<div class="doc_text">
Misha Brukman3896be22003-10-24 18:06:11 +0000[diff] [blame]322
323<p>The first step to determining what type of <a
324href="WritingAnLLVMPass.html">LLVM pass</a> you need to use for your Alias
325Analysis. As is the case with most other analyses and transformations, the
326answer should be fairly obvious from what type of problem you are trying to
327solve:</p>
328
Chris Lattner9f648752003-03-04 19:37:49 +0000[diff] [blame]329<ol>
Misha Brukmancd11e452003-10-22 23:27:16 +0000[diff] [blame]330 <li>If you require interprocedural analysis, it should be a
331 <tt>Pass</tt>.</li>
Chris Lattnerccc80652004-05-23 21:04:01 +0000[diff] [blame]332 <li>If you are a function-local analysis, subclass <tt>FunctionPass</tt>.</li>
Misha Brukmancd11e452003-10-22 23:27:16 +0000[diff] [blame]333 <li>If you don't need to look at the program at all, subclass
334 <tt>ImmutablePass</tt>.</li>
335</ol>
Misha Brukman3896be22003-10-24 18:06:11 +0000[diff] [blame]336
337<p>In addition to the pass that you subclass, you should also inherit from the
Misha Brukman700fd492003-05-07 21:47:16 +0000[diff] [blame]338<tt>AliasAnalysis</tt> interface, of course, and use the
Chris Lattner9f648752003-03-04 19:37:49 +0000[diff] [blame]339<tt>RegisterAnalysisGroup</tt> template to register as an implementation of
Misha Brukman3896be22003-10-24 18:06:11 +0000[diff] [blame]340<tt>AliasAnalysis</tt>.</p>
341
Misha Brukmancd11e452003-10-22 23:27:16 +0000[diff] [blame]342</div>
Chris Lattner9f648752003-03-04 19:37:49 +0000[diff] [blame]343
344<!-- ======================================================================= -->
Misha Brukmancd11e452003-10-22 23:27:16 +0000[diff] [blame]345<div class="doc_subsection">
346 <a name="requiredcalls">Required initialization calls</a>
347</div>
Chris Lattner9f648752003-03-04 19:37:49 +0000[diff] [blame]348
Misha Brukmancd11e452003-10-22 23:27:16 +0000[diff] [blame]349<div class="doc_text">
Misha Brukman3896be22003-10-24 18:06:11 +0000[diff] [blame]350
Chris Lattnerccc80652004-05-23 21:04:01 +0000[diff] [blame]351<p>Your subclass of <tt>AliasAnalysis</tt> is required to invoke two methods on
352the <tt>AliasAnalysis</tt> base class: <tt>getAnalysisUsage</tt> and
Chris Lattner9f648752003-03-04 19:37:49 +0000[diff] [blame]353<tt>InitializeAliasAnalysis</tt>. In particular, your implementation of
354<tt>getAnalysisUsage</tt> should explicitly call into the
355<tt>AliasAnalysis::getAnalysisUsage</tt> method in addition to doing any
356declaring any pass dependencies your pass has. Thus you should have something
Misha Brukman3896be22003-10-24 18:06:11 +0000[diff] [blame]357like this:</p>
358
Misha Brukman294611a2004-07-28 22:18:33 +0000[diff] [blame]359<div class="doc_code">
Chris Lattner9f648752003-03-04 19:37:49 +0000[diff] [blame]360<pre>
Misha Brukman294611a2004-07-28 22:18:33 +0000[diff] [blame]361void getAnalysisUsage(AnalysisUsage &amp;AU) const {
362 AliasAnalysis::getAnalysisUsage(AU);
363 <i>// declare your dependencies here.</i>
364}
Chris Lattner9f648752003-03-04 19:37:49 +0000[diff] [blame]365</pre>
Misha Brukman294611a2004-07-28 22:18:33 +0000[diff] [blame]366</div>
Misha Brukman3896be22003-10-24 18:06:11 +0000[diff] [blame]367
368<p>Additionally, your must invoke the <tt>InitializeAliasAnalysis</tt> method
369from your analysis run method (<tt>run</tt> for a <tt>Pass</tt>,
Chris Lattnerccc80652004-05-23 21:04:01 +0000[diff] [blame]370<tt>runOnFunction</tt> for a <tt>FunctionPass</tt>, or <tt>InitializePass</tt>
371for an <tt>ImmutablePass</tt>). For example (as part of a <tt>Pass</tt>):</p>
Misha Brukman3896be22003-10-24 18:06:11 +0000[diff] [blame]372
Misha Brukman294611a2004-07-28 22:18:33 +0000[diff] [blame]373<div class="doc_code">
Chris Lattner9f648752003-03-04 19:37:49 +0000[diff] [blame]374<pre>
Misha Brukman294611a2004-07-28 22:18:33 +0000[diff] [blame]375bool run(Module &amp;M) {
376 InitializeAliasAnalysis(this);
377 <i>// Perform analysis here...</i>
378 return false;
379}
Chris Lattner9f648752003-03-04 19:37:49 +0000[diff] [blame]380</pre>
Misha Brukman294611a2004-07-28 22:18:33 +0000[diff] [blame]381</div>
Misha Brukman3896be22003-10-24 18:06:11 +0000[diff] [blame]382
Misha Brukmancd11e452003-10-22 23:27:16 +0000[diff] [blame]383</div>
Chris Lattner9f648752003-03-04 19:37:49 +0000[diff] [blame]384
385<!-- ======================================================================= -->
Misha Brukmancd11e452003-10-22 23:27:16 +0000[diff] [blame]386<div class="doc_subsection">
387 <a name="interfaces">Interfaces which may be specified</a>
388</div>
Chris Lattner9f648752003-03-04 19:37:49 +0000[diff] [blame]389
Misha Brukmancd11e452003-10-22 23:27:16 +0000[diff] [blame]390<div class="doc_text">
Misha Brukman3896be22003-10-24 18:06:11 +0000[diff] [blame]391
Chris Lattnerccc80652004-05-23 21:04:01 +0000[diff] [blame]392<p>All of the <a
393href="/doxygen/classllvm_1_1AliasAnalysis.html"><tt>AliasAnalysis</tt></a>
394virtual methods default to providing <a href="#chaining">chaining</a> to another
395alias analysis implementation, which ends up returning conservatively correct
396information (returning "May" Alias and "Mod/Ref" for alias and mod/ref queries
Misha Brukman3896be22003-10-24 18:06:11 +0000[diff] [blame]397respectively). Depending on the capabilities of the analysis you are
398implementing, you just override the interfaces you can improve.</p>
399
Misha Brukmancd11e452003-10-22 23:27:16 +0000[diff] [blame]400</div>
Chris Lattner9f648752003-03-04 19:37:49 +0000[diff] [blame]401
Chris Lattnerccc80652004-05-23 21:04:01 +0000[diff] [blame]402
403
Chris Lattner9f648752003-03-04 19:37:49 +0000[diff] [blame]404<!-- ======================================================================= -->
Misha Brukmancd11e452003-10-22 23:27:16 +0000[diff] [blame]405<div class="doc_subsection">
Chris Lattnerccc80652004-05-23 21:04:01 +0000[diff] [blame]406 <a name="chaining"><tt>AliasAnalysis</tt> chaining behavior</a>
Misha Brukmancd11e452003-10-22 23:27:16 +0000[diff] [blame]407</div>
Chris Lattner9f648752003-03-04 19:37:49 +0000[diff] [blame]408
Misha Brukmancd11e452003-10-22 23:27:16 +0000[diff] [blame]409<div class="doc_text">
Misha Brukman3896be22003-10-24 18:06:11 +0000[diff] [blame]410
Chris Lattnerccc80652004-05-23 21:04:01 +0000[diff] [blame]411<p>With only two special exceptions (the <tt><a
412href="#basic-aa">basicaa</a></tt> and <a href="#no-aa"><tt>no-aa</tt></a>
413passes) every alias analysis pass chains to another alias analysis
414implementation (for example, the user can specify "<tt>-basicaa -ds-aa
Chris Lattnera9cf1962010-03-01 19:24:17 +0000[diff] [blame]415-licm</tt>" to get the maximum benefit from both alias
Chris Lattnerccc80652004-05-23 21:04:01 +0000[diff] [blame]416analyses). The alias analysis class automatically takes care of most of this
417for methods that you don't override. For methods that you do override, in code
418paths that return a conservative MayAlias or Mod/Ref result, simply return
419whatever the superclass computes. For example:</p>
Misha Brukman3896be22003-10-24 18:06:11 +0000[diff] [blame]420
Misha Brukman294611a2004-07-28 22:18:33 +0000[diff] [blame]421<div class="doc_code">
Chris Lattnerccc80652004-05-23 21:04:01 +0000[diff] [blame]422<pre>
Misha Brukman294611a2004-07-28 22:18:33 +0000[diff] [blame]423AliasAnalysis::AliasResult alias(const Value *V1, unsigned V1Size,
424 const Value *V2, unsigned V2Size) {
425 if (...)
426 return NoAlias;
427 ...
Chris Lattnerccc80652004-05-23 21:04:01 +0000[diff] [blame]428
Misha Brukman294611a2004-07-28 22:18:33 +0000[diff] [blame]429 <i>// Couldn't determine a must or no-alias result.</i>
430 return AliasAnalysis::alias(V1, V1Size, V2, V2Size);
431}
Chris Lattnerccc80652004-05-23 21:04:01 +0000[diff] [blame]432</pre>
Misha Brukman294611a2004-07-28 22:18:33 +0000[diff] [blame]433</div>
Chris Lattnerccc80652004-05-23 21:04:01 +0000[diff] [blame]434
435<p>In addition to analysis queries, you must make sure to unconditionally pass
436LLVM <a href="#updating">update notification</a> methods to the superclass as
437well if you override them, which allows all alias analyses in a change to be
438updated.</p>
439
440</div>
441
442
443<!-- ======================================================================= -->
444<div class="doc_subsection">
445 <a name="updating">Updating analysis results for transformations</a>
446</div>
447
448<div class="doc_text">
449<p>
450Alias analysis information is initially computed for a static snapshot of the
451program, but clients will use this information to make transformations to the
452code. All but the most trivial forms of alias analysis will need to have their
453analysis results updated to reflect the changes made by these transformations.
454</p>
455
456<p>
457The <tt>AliasAnalysis</tt> interface exposes two methods which are used to
458communicate program changes from the clients to the analysis implementations.
459Various alias analysis implementations should use these methods to ensure that
460their internal data structures are kept up-to-date as the program changes (for
461example, when an instruction is deleted), and clients of alias analysis must be
462sure to call these interfaces appropriately.
463</p>
464</div>
465
466<!-- _______________________________________________________________________ -->
467<div class="doc_subsubsection">The <tt>deleteValue</tt> method</div>
468
469<div class="doc_text">
470The <tt>deleteValue</tt> method is called by transformations when they remove an
471instruction or any other value from the program (including values that do not
472use pointers). Typically alias analyses keep data structures that have entries
473for each value in the program. When this method is called, they should remove
474any entries for the specified value, if they exist.
475</div>
476
Chris Lattnerccc80652004-05-23 21:04:01 +0000[diff] [blame]477<!-- _______________________________________________________________________ -->
478<div class="doc_subsubsection">The <tt>copyValue</tt> method</div>
479
480<div class="doc_text">
481The <tt>copyValue</tt> method is used when a new value is introduced into the
482program. There is no way to introduce a value into the program that did not
483exist before (this doesn't make sense for a safe compiler transformation), so
484this is the only way to introduce a new value. This method indicates that the
485new value has exactly the same properties as the value being copied.
486</div>
487
Chris Lattnerccc80652004-05-23 21:04:01 +0000[diff] [blame]488<!-- _______________________________________________________________________ -->
489<div class="doc_subsubsection">The <tt>replaceWithNewValue</tt> method</div>
490
491<div class="doc_text">
492This method is a simple helper method that is provided to make clients easier to
493use. It is implemented by copying the old analysis information to the new
494value, then deleting the old value. This method cannot be overridden by alias
495analysis implementations.
Misha Brukmancd11e452003-10-22 23:27:16 +0000[diff] [blame]496</div>
Chris Lattner9f648752003-03-04 19:37:49 +0000[diff] [blame]497
498<!-- ======================================================================= -->
Misha Brukmancd11e452003-10-22 23:27:16 +0000[diff] [blame]499<div class="doc_subsection">
500 <a name="implefficiency">Efficiency Issues</a>
501</div>
Chris Lattner9f648752003-03-04 19:37:49 +0000[diff] [blame]502
Misha Brukmancd11e452003-10-22 23:27:16 +0000[diff] [blame]503<div class="doc_text">
Misha Brukman3896be22003-10-24 18:06:11 +0000[diff] [blame]504
505<p>From the LLVM perspective, the only thing you need to do to provide an
506efficient alias analysis is to make sure that alias analysis <b>queries</b> are
507serviced quickly. The actual calculation of the alias analysis results (the
508"run" method) is only performed once, but many (perhaps duplicate) queries may
509be performed. Because of this, try to move as much computation to the run
510method as possible (within reason).</p>
511
Misha Brukmancd11e452003-10-22 23:27:16 +0000[diff] [blame]512</div>
Chris Lattner9f648752003-03-04 19:37:49 +0000[diff] [blame]513
Dan Gohman3f43dc32010-06-24 19:34:03 +0000[diff] [blame]514<!-- ======================================================================= -->
515<div class="doc_subsection">
516 <a name="passmanager">Pass Manager Issues</a>
517</div>
518
519<div class="doc_text">
520
521<p>PassManager support for alternative AliasAnalysis implementation
522has some issues.</p>
523
524<p>There is no way to override the default alias analysis. It would
525be very useful to be able to do something like "opt -my-aa -O2" and
526have it use -my-aa for all passes which need AliasAnalysis, but there
527is currently no support for that, short of changing the source code
528and recompiling. Similarly, there is also no way of setting a chain
529of analyses as the default.</p>
530
531<p>There is no way for transform passes to declare that they preserve
532<tt>AliasAnalysis</tt> implementations. The <tt>AliasAnalysis</tt>
533interface includes <tt>deleteValue</tt> and <tt>copyValue</tt> methods
534which are intended to allow a pass to keep an AliasAnalysis consistent,
535however there's no way for a pass to declare in its
536<tt>getAnalysisUsage</tt> that it does so. Some passes attempt to use
537<tt>AU.addPreserved&lt;AliasAnalysis&gt;</tt>, however this doesn't
538actually have any effect.</tt>
539
540<p><tt>AliasAnalysisCounter</tt> (<tt>-count-aa</tt>) and <tt>AliasDebugger</tt>
541(<tt>-debug-aa</tt>) are implemented as <tt>ModulePass</tt> classes, so if your
542alias analysis uses <tt>FunctionPass</tt>, it won't be able to use
543these utilities. If you try to use them, the pass manager will
544silently route alias analysis queries directly to
545<tt>BasicAliasAnalysis</tt> instead.</p>
546
547<p>Similarly, the <tt>opt -p</tt> option introduces <tt>ModulePass</tt>
548passes between each pass, which prevents the use of <tt>FunctionPass</tt>
549alias analysis passes.</p>
550
551</div>
552
Chris Lattner9f648752003-03-04 19:37:49 +0000[diff] [blame]553<!-- *********************************************************************** -->
Misha Brukmancd11e452003-10-22 23:27:16 +0000[diff] [blame]554<div class="doc_section">
Chris Lattnerccc80652004-05-23 21:04:01 +0000[diff] [blame]555 <a name="using">Using alias analysis results</a>
Misha Brukmancd11e452003-10-22 23:27:16 +0000[diff] [blame]556</div>
Chris Lattner9f648752003-03-04 19:37:49 +0000[diff] [blame]557<!-- *********************************************************************** -->
558
Misha Brukmancd11e452003-10-22 23:27:16 +0000[diff] [blame]559<div class="doc_text">
Misha Brukman3896be22003-10-24 18:06:11 +0000[diff] [blame]560
561<p>There are several different ways to use alias analysis results. In order of
562preference, these are...</p>
563
Misha Brukmancd11e452003-10-22 23:27:16 +0000[diff] [blame]564</div>
Chris Lattner9f648752003-03-04 19:37:49 +0000[diff] [blame]565
566<!-- ======================================================================= -->
Misha Brukmancd11e452003-10-22 23:27:16 +0000[diff] [blame]567<div class="doc_subsection">
Chris Lattner4f4365e2009-04-25 21:11:37 +0000[diff] [blame]568 <a name="memdep">Using the <tt>MemoryDependenceAnalysis</tt> Pass</a>
Misha Brukmancd11e452003-10-22 23:27:16 +0000[diff] [blame]569</div>
Chris Lattner9f648752003-03-04 19:37:49 +0000[diff] [blame]570
Misha Brukmancd11e452003-10-22 23:27:16 +0000[diff] [blame]571<div class="doc_text">
Misha Brukman3896be22003-10-24 18:06:11 +0000[diff] [blame]572
Chris Lattner4f4365e2009-04-25 21:11:37 +0000[diff] [blame]573<p>The <tt>memdep</tt> pass uses alias analysis to provide high-level dependence
574information about memory-using instructions. This will tell you which store
575feeds into a load, for example. It uses caching and other techniques to be
576efficient, and is used by Dead Store Elimination, GVN, and memcpy optimizations.
577</p>
Misha Brukman3896be22003-10-24 18:06:11 +0000[diff] [blame]578
Misha Brukmancd11e452003-10-22 23:27:16 +0000[diff] [blame]579</div>
Chris Lattner9f648752003-03-04 19:37:49 +0000[diff] [blame]580
581<!-- ======================================================================= -->
Misha Brukmancd11e452003-10-22 23:27:16 +0000[diff] [blame]582<div class="doc_subsection">
583 <a name="ast">Using the <tt>AliasSetTracker</tt> class</a>
584</div>
Chris Lattner9f648752003-03-04 19:37:49 +0000[diff] [blame]585
Misha Brukmancd11e452003-10-22 23:27:16 +0000[diff] [blame]586<div class="doc_text">
Misha Brukman3896be22003-10-24 18:06:11 +0000[diff] [blame]587
588<p>Many transformations need information about alias <b>sets</b> that are active
589in some scope, rather than information about pairwise aliasing. The <tt><a
Misha Brukman294611a2004-07-28 22:18:33 +0000[diff] [blame]590href="/doxygen/classllvm_1_1AliasSetTracker.html">AliasSetTracker</a></tt> class
591is used to efficiently build these Alias Sets from the pairwise alias analysis
Chris Lattnerccc80652004-05-23 21:04:01 +0000[diff] [blame]592information provided by the <tt>AliasAnalysis</tt> interface.</p>
Misha Brukman3896be22003-10-24 18:06:11 +0000[diff] [blame]593
Chris Lattnerccc80652004-05-23 21:04:01 +0000[diff] [blame]594<p>First you initialize the AliasSetTracker by using the "<tt>add</tt>" methods
595to add information about various potentially aliasing instructions in the scope
596you are interested in. Once all of the alias sets are completed, your pass
597should simply iterate through the constructed alias sets, using the
598<tt>AliasSetTracker</tt> <tt>begin()</tt>/<tt>end()</tt> methods.</p>
Misha Brukman3896be22003-10-24 18:06:11 +0000[diff] [blame]599
600<p>The <tt>AliasSet</tt>s formed by the <tt>AliasSetTracker</tt> are guaranteed
Chris Lattner539ca702003-12-19 08:43:07 +0000[diff] [blame]601to be disjoint, calculate mod/ref information and volatility for the set, and
602keep track of whether or not all of the pointers in the set are Must aliases.
603The AliasSetTracker also makes sure that sets are properly folded due to call
Misha Brukman3896be22003-10-24 18:06:11 +0000[diff] [blame]604instructions, and can provide a list of pointers in each set.</p>
605
606<p>As an example user of this, the <a href="/doxygen/structLICM.html">Loop
Chris Lattnerccc80652004-05-23 21:04:01 +0000[diff] [blame]607Invariant Code Motion</a> pass uses <tt>AliasSetTracker</tt>s to calculate alias
608sets for each loop nest. If an <tt>AliasSet</tt> in a loop is not modified,
609then all load instructions from that set may be hoisted out of the loop. If any
610alias sets are stored to <b>and</b> are must alias sets, then the stores may be
611sunk to outside of the loop, promoting the memory location to a register for the
612duration of the loop nest. Both of these transformations only apply if the
613pointer argument is loop-invariant.</p>
Misha Brukman3896be22003-10-24 18:06:11 +0000[diff] [blame]614
Misha Brukmancd11e452003-10-22 23:27:16 +0000[diff] [blame]615</div>
Chris Lattner9f648752003-03-04 19:37:49 +0000[diff] [blame]616
Chris Lattnerccc80652004-05-23 21:04:01 +0000[diff] [blame]617<!-- _______________________________________________________________________ -->
Chris Lattner539ca702003-12-19 08:43:07 +0000[diff] [blame]618<div class="doc_subsubsection">
619 The AliasSetTracker implementation
620</div>
621
622<div class="doc_text">
623
624<p>The AliasSetTracker class is implemented to be as efficient as possible. It
625uses the union-find algorithm to efficiently merge AliasSets when a pointer is
626inserted into the AliasSetTracker that aliases multiple sets. The primary data
627structure is a hash table mapping pointers to the AliasSet they are in.</p>
628
629<p>The AliasSetTracker class must maintain a list of all of the LLVM Value*'s
630that are in each AliasSet. Since the hash table already has entries for each
631LLVM Value* of interest, the AliasesSets thread the linked list through these
632hash-table nodes to avoid having to allocate memory unnecessarily, and to make
633merging alias sets extremely efficient (the linked list merge is constant time).
634</p>
635
636<p>You shouldn't need to understand these details if you are just a client of
637the AliasSetTracker, but if you look at the code, hopefully this brief
638description will help make sense of why things are designed the way they
639are.</p>
640
641</div>
642
Chris Lattner9f648752003-03-04 19:37:49 +0000[diff] [blame]643<!-- ======================================================================= -->
Misha Brukmancd11e452003-10-22 23:27:16 +0000[diff] [blame]644<div class="doc_subsection">
Chris Lattnerccc80652004-05-23 21:04:01 +0000[diff] [blame]645 <a name="direct">Using the <tt>AliasAnalysis</tt> interface directly</a>
Misha Brukmancd11e452003-10-22 23:27:16 +0000[diff] [blame]646</div>
Chris Lattner9f648752003-03-04 19:37:49 +0000[diff] [blame]647
Misha Brukmancd11e452003-10-22 23:27:16 +0000[diff] [blame]648<div class="doc_text">
Misha Brukman3896be22003-10-24 18:06:11 +0000[diff] [blame]649
Chris Lattnerccc80652004-05-23 21:04:01 +0000[diff] [blame]650<p>If neither of these utility class are what your pass needs, you should use
651the interfaces exposed by the <tt>AliasAnalysis</tt> class directly. Try to use
652the higher-level methods when possible (e.g., use mod/ref information instead of
653the <a href="#alias"><tt>alias</tt></a> method directly if possible) to get the
654best precision and efficiency.</p>
Misha Brukman3896be22003-10-24 18:06:11 +0000[diff] [blame]655
Misha Brukmancd11e452003-10-22 23:27:16 +0000[diff] [blame]656</div>
Chris Lattner9f648752003-03-04 19:37:49 +0000[diff] [blame]657
658<!-- *********************************************************************** -->
Misha Brukmancd11e452003-10-22 23:27:16 +0000[diff] [blame]659<div class="doc_section">
Chris Lattnerccc80652004-05-23 21:04:01 +0000[diff] [blame]660 <a name="exist">Existing alias analysis implementations and clients</a>
Misha Brukmancd11e452003-10-22 23:27:16 +0000[diff] [blame]661</div>
Chris Lattner9f648752003-03-04 19:37:49 +0000[diff] [blame]662<!-- *********************************************************************** -->
663
Misha Brukmancd11e452003-10-22 23:27:16 +0000[diff] [blame]664<div class="doc_text">
Misha Brukman3896be22003-10-24 18:06:11 +0000[diff] [blame]665
Chris Lattnerccc80652004-05-23 21:04:01 +0000[diff] [blame]666<p>If you're going to be working with the LLVM alias analysis infrastructure,
667you should know what clients and implementations of alias analysis are
668available. In particular, if you are implementing an alias analysis, you should
669be aware of the <a href="#aliasanalysis-debug">the clients</a> that are useful
670for monitoring and evaluating different implementations.</p>
Misha Brukman3896be22003-10-24 18:06:11 +0000[diff] [blame]671
Misha Brukmancd11e452003-10-22 23:27:16 +0000[diff] [blame]672</div>
Chris Lattner9f648752003-03-04 19:37:49 +0000[diff] [blame]673
674<!-- ======================================================================= -->
Misha Brukmancd11e452003-10-22 23:27:16 +0000[diff] [blame]675<div class="doc_subsection">
Chris Lattnerccc80652004-05-23 21:04:01 +0000[diff] [blame]676 <a name="impls">Available <tt>AliasAnalysis</tt> implementations</a>
677</div>
678
679<div class="doc_text">
680
681<p>This section lists the various implementations of the <tt>AliasAnalysis</tt>
682interface. With the exception of the <a href="#no-aa"><tt>-no-aa</tt></a> and
683<a href="#basic-aa"><tt>-basicaa</tt></a> implementations, all of these <a
Chris Lattnerfcd37252004-06-21 22:52:48 +0000[diff] [blame]684href="#chaining">chain</a> to other alias analysis implementations.</p>
Chris Lattnerccc80652004-05-23 21:04:01 +0000[diff] [blame]685
686</div>
687
688<!-- _______________________________________________________________________ -->
689<div class="doc_subsubsection">
Misha Brukmancd11e452003-10-22 23:27:16 +0000[diff] [blame]690 <a name="no-aa">The <tt>-no-aa</tt> pass</a>
691</div>
Chris Lattner9f648752003-03-04 19:37:49 +0000[diff] [blame]692
Misha Brukmancd11e452003-10-22 23:27:16 +0000[diff] [blame]693<div class="doc_text">
Misha Brukman3896be22003-10-24 18:06:11 +0000[diff] [blame]694
Chris Lattnerccc80652004-05-23 21:04:01 +0000[diff] [blame]695<p>The <tt>-no-aa</tt> pass is just like what it sounds: an alias analysis that
696never returns any useful information. This pass can be useful if you think that
697alias analysis is doing something wrong and are trying to narrow down a
698problem.</p>
699
700</div>
701
Chris Lattnerccc80652004-05-23 21:04:01 +0000[diff] [blame]702<!-- _______________________________________________________________________ -->
703<div class="doc_subsubsection">
704 <a name="basic-aa">The <tt>-basicaa</tt> pass</a>
705</div>
706
707<div class="doc_text">
708
709<p>The <tt>-basicaa</tt> pass is the default LLVM alias analysis. It is an
710aggressive local analysis that "knows" many important facts:</p>
711
712<ul>
713<li>Distinct globals, stack allocations, and heap allocations can never
714 alias.</li>
715<li>Globals, stack allocations, and heap allocations never alias the null
716 pointer.</li>
717<li>Different fields of a structure do not alias.</li>
718<li>Indexes into arrays with statically differing subscripts cannot alias.</li>
719<li>Many common standard C library functions <a
720 href="#simplemodref">never access memory or only read memory</a>.</li>
721<li>Pointers that obviously point to constant globals
722 "<tt>pointToConstantMemory</tt>".</li>
723<li>Function calls can not modify or references stack allocations if they never
724 escape from the function that allocates them (a common case for automatic
725 arrays).</li>
726</ul>
727
728</div>
729
Chris Lattner100a4f82004-06-28 19:19:47 +0000[diff] [blame]730<!-- _______________________________________________________________________ -->
731<div class="doc_subsubsection">
732 <a name="globalsmodref">The <tt>-globalsmodref-aa</tt> pass</a>
733</div>
734
735<div class="doc_text">
736
737<p>This pass implements a simple context-sensitive mod/ref and alias analysis
Chris Lattnerccb354b2004-07-27 07:50:07 +0000[diff] [blame]738for internal global variables that don't "have their address taken". If a
739global does not have its address taken, the pass knows that no pointers alias
740the global. This pass also keeps track of functions that it knows never access
Chris Lattner4f4365e2009-04-25 21:11:37 +0000[diff] [blame]741memory or never read memory. This allows certain optimizations (e.g. GVN) to
Chris Lattnerccb354b2004-07-27 07:50:07 +0000[diff] [blame]742eliminate call instructions entirely.
Chris Lattner100a4f82004-06-28 19:19:47 +0000[diff] [blame]743</p>
744
745<p>The real power of this pass is that it provides context-sensitive mod/ref
746information for call instructions. This allows the optimizer to know that
747calls to a function do not clobber or read the value of the global, allowing
748loads and stores to be eliminated.</p>
749
750<p>Note that this pass is somewhat limited in its scope (only support
751non-address taken globals), but is very quick analysis.</p>
752</div>
Chris Lattnerccc80652004-05-23 21:04:01 +0000[diff] [blame]753
754<!-- _______________________________________________________________________ -->
755<div class="doc_subsubsection">
Chris Lattnerccc80652004-05-23 21:04:01 +0000[diff] [blame]756 <a name="steens-aa">The <tt>-steens-aa</tt> pass</a>
757</div>
758
759<div class="doc_text">
760
761<p>The <tt>-steens-aa</tt> pass implements a variation on the well-known
762"Steensgaard's algorithm" for interprocedural alias analysis. Steensgaard's
763algorithm is a unification-based, flow-insensitive, context-insensitive, and
764field-insensitive alias analysis that is also very scalable (effectively linear
765time).</p>
766
767<p>The LLVM <tt>-steens-aa</tt> pass implements a "speculatively
768field-<b>sensitive</b>" version of Steensgaard's algorithm using the Data
769Structure Analysis framework. This gives it substantially more precision than
770the standard algorithm while maintaining excellent analysis scalability.</p>
771
Chris Lattnerd6e0dd12007-07-03 04:41:50 +0000[diff] [blame]772<p>Note that <tt>-steens-aa</tt> is available in the optional "poolalloc"
773module, it is not part of the LLVM core.</p>
774
Chris Lattnerccc80652004-05-23 21:04:01 +0000[diff] [blame]775</div>
776
777<!-- _______________________________________________________________________ -->
778<div class="doc_subsubsection">
779 <a name="ds-aa">The <tt>-ds-aa</tt> pass</a>
780</div>
781
782<div class="doc_text">
783
784<p>The <tt>-ds-aa</tt> pass implements the full Data Structure Analysis
785algorithm. Data Structure Analysis is a modular unification-based,
786flow-insensitive, context-<b>sensitive</b>, and speculatively
787field-<b>sensitive</b> alias analysis that is also quite scalable, usually at
788O(n*log(n)).</p>
789
790<p>This algorithm is capable of responding to a full variety of alias analysis
791queries, and can provide context-sensitive mod/ref information as well. The
792only major facility not implemented so far is support for must-alias
793information.</p>
Misha Brukman3896be22003-10-24 18:06:11 +0000[diff] [blame]794
Chris Lattnerd6e0dd12007-07-03 04:41:50 +0000[diff] [blame]795<p>Note that <tt>-ds-aa</tt> is available in the optional "poolalloc"
796module, it is not part of the LLVM core.</p>
797
Misha Brukmancd11e452003-10-22 23:27:16 +0000[diff] [blame]798</div>
Chris Lattner9f648752003-03-04 19:37:49 +0000[diff] [blame]799
Dan Gohman85dfca62010-06-28 22:09:52 +0000[diff] [blame]800<!-- _______________________________________________________________________ -->
801<div class="doc_subsubsection">
802 <a name="scev-aa">The <tt>-scev-aa</tt> pass</a>
803</div>
804
805<div class="doc_text">
806
807<p>The <tt>-scev-aa</tt> pass implements AliasAnalysis queries by
808translating them into ScalarEvolution queries. This gives it a
809more complete understanding of <tt>getelementptr</tt> instructions
810and loop induction variables than other alias analyses have.</p>
811
812</div>
Chris Lattner9f648752003-03-04 19:37:49 +0000[diff] [blame]813
814<!-- ======================================================================= -->
Misha Brukmancd11e452003-10-22 23:27:16 +0000[diff] [blame]815<div class="doc_subsection">
Chris Lattnerccc80652004-05-23 21:04:01 +0000[diff] [blame]816 <a name="aliasanalysis-xforms">Alias analysis driven transformations</a>
817</div>
818
819<div class="doc_text">
820LLVM includes several alias-analysis driven transformations which can be used
821with any of the implementations above.
822</div>
823
824<!-- _______________________________________________________________________ -->
825<div class="doc_subsubsection">
826 <a name="adce">The <tt>-adce</tt> pass</a>
827</div>
828
829<div class="doc_text">
830
831<p>The <tt>-adce</tt> pass, which implements Aggressive Dead Code Elimination
832uses the <tt>AliasAnalysis</tt> interface to delete calls to functions that do
833not have side-effects and are not used.</p>
834
835</div>
836
837
838<!-- _______________________________________________________________________ -->
839<div class="doc_subsubsection">
840 <a name="licm">The <tt>-licm</tt> pass</a>
841</div>
842
843<div class="doc_text">
844
845<p>The <tt>-licm</tt> pass implements various Loop Invariant Code Motion related
846transformations. It uses the <tt>AliasAnalysis</tt> interface for several
847different transformations:</p>
848
849<ul>
850<li>It uses mod/ref information to hoist or sink load instructions out of loops
851if there are no instructions in the loop that modifies the memory loaded.</li>
852
853<li>It uses mod/ref information to hoist function calls out of loops that do not
854write to memory and are loop-invariant.</li>
855
856<li>If uses alias information to promote memory objects that are loaded and
857stored to in loops to live in a register instead. It can do this if there are
858no may aliases to the loaded/stored memory location.</li>
859</ul>
860
861</div>
862
863<!-- _______________________________________________________________________ -->
864<div class="doc_subsubsection">
865 <a name="argpromotion">The <tt>-argpromotion</tt> pass</a>
866</div>
867
868<div class="doc_text">
869<p>
870The <tt>-argpromotion</tt> pass promotes by-reference arguments to be passed in
871by-value instead. In particular, if pointer arguments are only loaded from it
872passes in the value loaded instead of the address to the function. This pass
873uses alias information to make sure that the value loaded from the argument
874pointer is not modified between the entry of the function and any load of the
875pointer.</p>
876</div>
877
878<!-- _______________________________________________________________________ -->
879<div class="doc_subsubsection">
Chris Lattner4f4365e2009-04-25 21:11:37 +0000[diff] [blame]880 <a name="gvn">The <tt>-gvn</tt>, <tt>-memcpyopt</tt>, and <tt>-dse</tt>
881 passes</a>
Chris Lattnerccc80652004-05-23 21:04:01 +0000[diff] [blame]882</div>
883
884<div class="doc_text">
Misha Brukman294611a2004-07-28 22:18:33 +0000[diff] [blame]885
Chris Lattner4f4365e2009-04-25 21:11:37 +0000[diff] [blame]886<p>These passes use AliasAnalysis information to reason about loads and stores.
887</p>
Chris Lattnerccc80652004-05-23 21:04:01 +0000[diff] [blame]888
889</div>
890
Chris Lattnerccc80652004-05-23 21:04:01 +0000[diff] [blame]891<!-- ======================================================================= -->
892<div class="doc_subsection">
Misha Brukman294611a2004-07-28 22:18:33 +0000[diff] [blame]893 <a name="aliasanalysis-debug">Clients for debugging and evaluation of
894 implementations</a>
Chris Lattnerccc80652004-05-23 21:04:01 +0000[diff] [blame]895</div>
896
Misha Brukman294611a2004-07-28 22:18:33 +0000[diff] [blame]897<div class="doc_text">
Chris Lattnerccc80652004-05-23 21:04:01 +0000[diff] [blame]898
Misha Brukman294611a2004-07-28 22:18:33 +0000[diff] [blame]899<p>These passes are useful for evaluating the various alias analysis
Chris Lattnera9cf1962010-03-01 19:24:17 +0000[diff] [blame]900implementations. You can use them with commands like '<tt>opt -ds-aa
Misha Brukman294611a2004-07-28 22:18:33 +0000[diff] [blame]901-aa-eval foo.bc -disable-output -stats</tt>'.</p>
902
903</div>
Chris Lattnerccc80652004-05-23 21:04:01 +0000[diff] [blame]904
905<!-- _______________________________________________________________________ -->
906<div class="doc_subsubsection">
Misha Brukmancd11e452003-10-22 23:27:16 +0000[diff] [blame]907 <a name="print-alias-sets">The <tt>-print-alias-sets</tt> pass</a>
908</div>
Chris Lattner9f648752003-03-04 19:37:49 +0000[diff] [blame]909
Misha Brukmancd11e452003-10-22 23:27:16 +0000[diff] [blame]910<div class="doc_text">
Misha Brukman3896be22003-10-24 18:06:11 +0000[diff] [blame]911
912<p>The <tt>-print-alias-sets</tt> pass is exposed as part of the
Chris Lattner05d5c9d2006-01-03 06:04:48 +0000[diff] [blame]913<tt>opt</tt> tool to print out the Alias Sets formed by the <a
Chris Lattner9f648752003-03-04 19:37:49 +0000[diff] [blame]914href="#ast"><tt>AliasSetTracker</tt></a> class. This is useful if you're using
Chris Lattner05d5c9d2006-01-03 06:04:48 +0000[diff] [blame]915the <tt>AliasSetTracker</tt> class. To use it, use something like:</p>
916
917<div class="doc_code">
918<pre>
919% opt -ds-aa -print-alias-sets -disable-output
920</pre>
921</div>
Misha Brukman3896be22003-10-24 18:06:11 +0000[diff] [blame]922
Misha Brukmancd11e452003-10-22 23:27:16 +0000[diff] [blame]923</div>
Chris Lattner9f648752003-03-04 19:37:49 +0000[diff] [blame]924
Chris Lattnerccc80652004-05-23 21:04:01 +0000[diff] [blame]925
926<!-- _______________________________________________________________________ -->
927<div class="doc_subsubsection">
Misha Brukmancd11e452003-10-22 23:27:16 +0000[diff] [blame]928 <a name="count-aa">The <tt>-count-aa</tt> pass</a>
929</div>
Chris Lattner9f648752003-03-04 19:37:49 +0000[diff] [blame]930
Misha Brukmancd11e452003-10-22 23:27:16 +0000[diff] [blame]931<div class="doc_text">
Misha Brukman3896be22003-10-24 18:06:11 +0000[diff] [blame]932
933<p>The <tt>-count-aa</tt> pass is useful to see how many queries a particular
Misha Brukman294611a2004-07-28 22:18:33 +0000[diff] [blame]934pass is making and what responses are returned by the alias analysis. As an
935example,</p>
Misha Brukman3896be22003-10-24 18:06:11 +0000[diff] [blame]936
Misha Brukman294611a2004-07-28 22:18:33 +0000[diff] [blame]937<div class="doc_code">
Chris Lattner9f648752003-03-04 19:37:49 +0000[diff] [blame]938<pre>
Misha Brukman294611a2004-07-28 22:18:33 +0000[diff] [blame]939% opt -basicaa -count-aa -ds-aa -count-aa -licm
Chris Lattner9f648752003-03-04 19:37:49 +0000[diff] [blame]940</pre>
Misha Brukman294611a2004-07-28 22:18:33 +0000[diff] [blame]941</div>
Misha Brukman3896be22003-10-24 18:06:11 +0000[diff] [blame]942
Misha Brukman294611a2004-07-28 22:18:33 +0000[diff] [blame]943<p>will print out how many queries (and what responses are returned) by the
944<tt>-licm</tt> pass (of the <tt>-ds-aa</tt> pass) and how many queries are made
945of the <tt>-basicaa</tt> pass by the <tt>-ds-aa</tt> pass. This can be useful
946when debugging a transformation or an alias analysis implementation.</p>
Misha Brukman3896be22003-10-24 18:06:11 +0000[diff] [blame]947
Misha Brukmancd11e452003-10-22 23:27:16 +0000[diff] [blame]948</div>
Chris Lattner9f648752003-03-04 19:37:49 +0000[diff] [blame]949
Chris Lattnerccc80652004-05-23 21:04:01 +0000[diff] [blame]950<!-- _______________________________________________________________________ -->
951<div class="doc_subsubsection">
Misha Brukmancd11e452003-10-22 23:27:16 +0000[diff] [blame]952 <a name="aa-eval">The <tt>-aa-eval</tt> pass</a>
953</div>
Chris Lattner9f648752003-03-04 19:37:49 +0000[diff] [blame]954
Misha Brukmancd11e452003-10-22 23:27:16 +0000[diff] [blame]955<div class="doc_text">
Misha Brukman3896be22003-10-24 18:06:11 +0000[diff] [blame]956
957<p>The <tt>-aa-eval</tt> pass simply iterates through all pairs of pointers in a
Chris Lattner9f648752003-03-04 19:37:49 +0000[diff] [blame]958function and asks an alias analysis whether or not the pointers alias. This
959gives an indication of the precision of the alias analysis. Statistics are
Chris Lattnerccc80652004-05-23 21:04:01 +0000[diff] [blame]960printed indicating the percent of no/may/must aliases found (a more precise
961algorithm will have a lower number of may aliases).</p>
Misha Brukman3896be22003-10-24 18:06:11 +0000[diff] [blame]962
Misha Brukmancd11e452003-10-22 23:27:16 +0000[diff] [blame]963</div>
Chris Lattner9f648752003-03-04 19:37:49 +0000[diff] [blame]964
Chris Lattner9f648752003-03-04 19:37:49 +0000[diff] [blame]965<!-- *********************************************************************** -->
Owen Anderson5a726b82007-10-02 00:43:25 +0000[diff] [blame]966<div class="doc_section">
967 <a name="memdep">Memory Dependence Analysis</a>
968</div>
969<!-- *********************************************************************** -->
970
971<div class="doc_text">
972
973<p>If you're just looking to be a client of alias analysis information, consider
974using the Memory Dependence Analysis interface instead. MemDep is a lazy,
975caching layer on top of alias analysis that is able to answer the question of
976what preceding memory operations a given instruction depends on, either at an
977intra- or inter-block level. Because of its laziness and caching
978policy, using MemDep can be a significant performance win over accessing alias
979analysis directly.</p>
980
981</div>
982
983<!-- *********************************************************************** -->
Chris Lattner9f648752003-03-04 19:37:49 +0000[diff] [blame]984
Misha Brukman3896be22003-10-24 18:06:11 +0000[diff] [blame]985<hr>
Misha Brukman915cab22003-11-22 01:26:21 +0000[diff] [blame]986<address>
Misha Brukman38847d52003-12-21 22:53:21 +0000[diff] [blame]987 <a href="http://jigsaw.w3.org/css-validator/check/referer"><img
Misha Brukman44408702008-12-11 17:34:48 +0000[diff] [blame]988 src="http://jigsaw.w3.org/css-validator/images/vcss-blue" alt="Valid CSS"></a>
Misha Brukman915cab22003-11-22 01:26:21 +0000[diff] [blame]989 <a href="http://validator.w3.org/check/referer"><img
Misha Brukman44408702008-12-11 17:34:48 +0000[diff] [blame]990 src="http://www.w3.org/Icons/valid-html401-blue" alt="Valid HTML 4.01"></a>
Misha Brukman915cab22003-11-22 01:26:21 +0000[diff] [blame]991
992 <a href="mailto:sabre@nondot.org">Chris Lattner</a><br>
Reid Spencer05fe4b02006-03-14 05:39:39 +0000[diff] [blame]993 <a href="http://llvm.org">LLVM Compiler Infrastructure</a><br>
Misha Brukman915cab22003-11-22 01:26:21 +0000[diff] [blame]994 Last modified: $Date$
995</address>
Misha Brukmancd11e452003-10-22 23:27:16 +0000[diff] [blame]996
997</body>
998</html>