Gitiles
Code Review Sign In
gerrit-public.fairphone.software / fp2-dev / platform / external / llvm / 5de585a57e8c89ba0f01b6b90a66a413692cc9cd / . / docs / AliasAnalysis.html
blob: c6330a946956edd771204f0dade2b71a268889c9 [file] [log] [blame]
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]1<!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.01//EN"
2 "http://www.w3.org/TR/html4/strict.dtd">
3<html>
4<head>
5 <title>LLVM Alias Analysis Infrastructure</title>
6 <link rel="stylesheet" href="llvm.css" type="text/css">
7</head>
8<body>
9
10<div class="doc_title">
11 LLVM Alias Analysis Infrastructure
12</div>
13
14<ol>
15 <li><a href="#introduction">Introduction</a></li>
16
17 <li><a href="#overview"><tt>AliasAnalysis</tt> Class Overview</a>
18 <ul>
19 <li><a href="#pointers">Representation of Pointers</a></li>
20 <li><a href="#alias">The <tt>alias</tt> method</a></li>
21 <li><a href="#ModRefInfo">The <tt>getModRefInfo</tt> methods</a></li>
22 <li><a href="#OtherItfs">Other useful <tt>AliasAnalysis</tt> methods</a></li>
23 </ul>
24 </li>
25
26 <li><a href="#writingnew">Writing a new <tt>AliasAnalysis</tt> Implementation</a>
27 <ul>
28 <li><a href="#passsubclasses">Different Pass styles</a></li>
29 <li><a href="#requiredcalls">Required initialization calls</a></li>
30 <li><a href="#interfaces">Interfaces which may be specified</a></li>
31 <li><a href="#chaining"><tt>AliasAnalysis</tt> chaining behavior</a></li>
32 <li><a href="#updating">Updating analysis results for transformations</a></li>
33 <li><a href="#implefficiency">Efficiency Issues</a></li>
Dan Gohmanaa785442010-06-24 19:34:03 +0000[diff] [blame]34 <li><a href="#passmanager">Pass Manager Issues</a></li>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]35 </ul>
36 </li>
37
38 <li><a href="#using">Using alias analysis results</a>
39 <ul>
Chris Lattner72da5762009-04-25 21:11:37 +0000[diff] [blame]40 <li><a href="#memdep">Using the <tt>MemoryDependenceAnalysis</tt> Pass</a></li>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]41 <li><a href="#ast">Using the <tt>AliasSetTracker</tt> class</a></li>
42 <li><a href="#direct">Using the <tt>AliasAnalysis</tt> interface directly</a></li>
43 </ul>
44 </li>
45
46 <li><a href="#exist">Existing alias analysis implementations and clients</a>
47 <ul>
48 <li><a href="#impls">Available <tt>AliasAnalysis</tt> implementations</a></li>
49 <li><a href="#aliasanalysis-xforms">Alias analysis driven transformations</a></li>
50 <li><a href="#aliasanalysis-debug">Clients for debugging and evaluation of
51 implementations</a></li>
52 </ul>
53 </li>
Owen Anderson4f10b312007-10-02 00:44:20 +0000[diff] [blame]54 <li><a href="#memdep">Memory Dependence Analysis</a></li>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]55</ol>
56
57<div class="doc_author">
58 <p>Written by <a href="mailto:sabre@nondot.org">Chris Lattner</a></p>
59</div>
60
61<!-- *********************************************************************** -->
62<div class="doc_section">
63 <a name="introduction">Introduction</a>
64</div>
65<!-- *********************************************************************** -->
66
67<div class="doc_text">
68
69<p>Alias Analysis (aka Pointer Analysis) is a class of techniques which attempt
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
75to a query with a <a href="#MustMayNo">Must, May, or No</a> alias response,
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>
78
79<p>The LLVM <a
80href="http://llvm.org/doxygen/classllvm_1_1AliasAnalysis.html"><tt>AliasAnalysis</tt></a>
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>
89
90<p>This document contains information necessary to successfully implement this
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
93or should be added, please <a href="mailto:sabre@nondot.org">let me
94know</a>.</p>
95
96</div>
97
98<!-- *********************************************************************** -->
99<div class="doc_section">
100 <a name="overview"><tt>AliasAnalysis</tt> Class Overview</a>
101</div>
102<!-- *********************************************************************** -->
103
104<div class="doc_text">
105
106<p>The <a
107href="http://llvm.org/doxygen/classllvm_1_1AliasAnalysis.html"><tt>AliasAnalysis</tt></a>
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>
112
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>
119
Dan Gohman83a01ad22010-07-07 14:27:09 +0000[diff] [blame]120<p>All <tt>AliasAnalysis</tt> interfaces require that in queries involving
121multiple values, values which are not
122<a href="LangRef.html#constants">constants</a> are all defined within the
123same function.</p>
124
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]125</div>
126
127<!-- ======================================================================= -->
128<div class="doc_subsection">
129 <a name="pointers">Representation of Pointers</a>
130</div>
131
132<div class="doc_text">
133
134<p>Most importantly, the <tt>AliasAnalysis</tt> class provides several methods
135which are used to query whether or not two memory objects alias, whether
136function calls can modify or read a memory object, etc. For all of these
137queries, memory objects are represented as a pair of their starting address (a
138symbolic LLVM <tt>Value*</tt>) and a static size.</p>
139
140<p>Representing memory objects as a starting address and a size is critically
141important for correct Alias Analyses. For example, consider this (silly, but
142possible) C code:</p>
143
144<div class="doc_code">
145<pre>
146int i;
147char C[2];
148char A[10];
149/* ... */
150for (i = 0; i != 10; ++i) {
151 C[0] = A[i]; /* One byte store */
152 C[1] = A[9-i]; /* One byte store */
153}
154</pre>
155</div>
156
157<p>In this case, the <tt>basicaa</tt> pass will disambiguate the stores to
158<tt>C[0]</tt> and <tt>C[1]</tt> because they are accesses to two distinct
159locations one byte apart, and the accesses are each one byte. In this case, the
160LICM pass can use store motion to remove the stores from the loop. In
161constrast, the following code:</p>
162
163<div class="doc_code">
164<pre>
165int i;
166char C[2];
167char A[10];
168/* ... */
169for (i = 0; i != 10; ++i) {
170 ((short*)C)[0] = A[i]; /* Two byte store! */
171 C[1] = A[9-i]; /* One byte store */
172}
173</pre>
174</div>
175
176<p>In this case, the two stores to C do alias each other, because the access to
177the <tt>&amp;C[0]</tt> element is a two byte access. If size information wasn't
178available in the query, even the first case would have to conservatively assume
179that the accesses alias.</p>
180
181</div>
182
183<!-- ======================================================================= -->
184<div class="doc_subsection">
185 <a name="alias">The <tt>alias</tt> method</a>
186</div>
187
188<div class="doc_text">
Dan Gohman83a01ad22010-07-07 14:27:09 +0000[diff] [blame]189<p>The <tt>alias</tt> method is the primary interface used to determine whether
190or not two memory objects alias each other. It takes two memory objects as
191input and returns MustAlias, MayAlias, or NoAlias as appropriate.</p>
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>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]196</div>
197
198<!-- _______________________________________________________________________ -->
199<div class="doc_subsubsection">
200 <a name="MustMayNo">Must, May, and No Alias Responses</a>
201</div>
202
203<div class="doc_text">
Dan Gohmanc8208442010-07-02 18:41:32 +0000[diff] [blame]204<p>The NoAlias response may be used when there is never an immediate dependence
205between any memory reference <i>based</i> on one pointer and any memory
206reference <i>based</i> the other. The most obvious example is when the two
207pointers point to non-overlapping memory ranges. Another is when the two
208pointers are only ever used for reading memory. Another is when the memory is
209freed and reallocated between accesses through one pointer and accesses through
210the other -- in this case, there is a dependence, but it's mediated by the free
211and reallocation.</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]212
Dan Gohman24fc36d2010-07-02 23:46:54 +0000[diff] [blame]213<p>As an exception to this is with the
Dan Gohman83a01ad22010-07-07 14:27:09 +0000[diff] [blame]214<a href="LangRef.html#noalias"><tt>noalias</tt></a> keyword; the "irrelevant"
215dependencies are ignored.</p>
Dan Gohman24fc36d2010-07-02 23:46:54 +0000[diff] [blame]216
Nick Lewycky74c87562008-12-14 21:08:48 +0000[diff] [blame]217<p>The MayAlias response is used whenever the two pointers might refer to the
218same object. If the two memory objects overlap, but do not start at the same
219location, return MayAlias.</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]220
Nick Lewycky74c87562008-12-14 21:08:48 +0000[diff] [blame]221<p>The MustAlias response may only be returned if the two memory objects are
222guaranteed to always start at exactly the same location. A MustAlias response
223implies that the pointers compare equal.</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]224
225</div>
226
227<!-- ======================================================================= -->
228<div class="doc_subsection">
229 <a name="ModRefInfo">The <tt>getModRefInfo</tt> methods</a>
230</div>
231
232<div class="doc_text">
233
234<p>The <tt>getModRefInfo</tt> methods return information about whether the
235execution of an instruction can read or modify a memory location. Mod/Ref
236information is always conservative: if an instruction <b>might</b> read or write
237a location, ModRef is returned.</p>
238
239<p>The <tt>AliasAnalysis</tt> class also provides a <tt>getModRefInfo</tt>
240method for testing dependencies between function calls. This method takes two
241call sites (CS1 &amp; CS2), returns NoModRef if the two calls refer to disjoint
242memory locations, Ref if CS1 reads memory written by CS2, Mod if CS1 writes to
243memory read or written by CS2, or ModRef if CS1 might read or write memory
Chris Lattner3b803752009-11-22 16:01:44 +0000[diff] [blame]244accessed by CS2. Note that this relation is not commutative.</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]245
246</div>
247
248
249<!-- ======================================================================= -->
250<div class="doc_subsection">
251 <a name="OtherItfs">Other useful <tt>AliasAnalysis</tt> methods</a>
252</div>
253
254<div class="doc_text">
255
256<p>
257Several other tidbits of information are often collected by various alias
258analysis implementations and can be put to good use by various clients.
259</p>
260
261</div>
262
263<!-- _______________________________________________________________________ -->
264<div class="doc_subsubsection">
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]265 The <tt>pointsToConstantMemory</tt> method
266</div>
267
268<div class="doc_text">
269
270<p>The <tt>pointsToConstantMemory</tt> method returns true if and only if the
271analysis can prove that the pointer only points to unchanging memory locations
272(functions, constant global variables, and the null pointer). This information
273can be used to refine mod/ref information: it is impossible for an unchanging
274memory location to be modified.</p>
275
276</div>
277
278<!-- _______________________________________________________________________ -->
279<div class="doc_subsubsection">
280 <a name="simplemodref">The <tt>doesNotAccessMemory</tt> and
281 <tt>onlyReadsMemory</tt> methods</a>
282</div>
283
284<div class="doc_text">
285
286<p>These methods are used to provide very simple mod/ref information for
287function calls. The <tt>doesNotAccessMemory</tt> method returns true for a
288function if the analysis can prove that the function never reads or writes to
289memory, or if the function only reads from constant memory. Functions with this
290property are side-effect free and only depend on their input arguments, allowing
291them to be eliminated if they form common subexpressions or be hoisted out of
292loops. Many common functions behave this way (e.g., <tt>sin</tt> and
293<tt>cos</tt>) but many others do not (e.g., <tt>acos</tt>, which modifies the
294<tt>errno</tt> variable).</p>
295
296<p>The <tt>onlyReadsMemory</tt> method returns true for a function if analysis
297can prove that (at most) the function only reads from non-volatile memory.
298Functions with this property are side-effect free, only depending on their input
299arguments and the state of memory when they are called. This property allows
300calls to these functions to be eliminated and moved around, as long as there is
301no store instruction that changes the contents of memory. Note that all
302functions that satisfy the <tt>doesNotAccessMemory</tt> method also satisfies
303<tt>onlyReadsMemory</tt>.</p>
304
305</div>
306
307<!-- *********************************************************************** -->
308<div class="doc_section">
309 <a name="writingnew">Writing a new <tt>AliasAnalysis</tt> Implementation</a>
310</div>
311<!-- *********************************************************************** -->
312
313<div class="doc_text">
314
315<p>Writing a new alias analysis implementation for LLVM is quite
316straight-forward. There are already several implementations that you can use
317for examples, and the following information should help fill in any details.
318For a examples, take a look at the <a href="#impls">various alias analysis
319implementations</a> included with LLVM.</p>
320
321</div>
322
323<!-- ======================================================================= -->
324<div class="doc_subsection">
325 <a name="passsubclasses">Different Pass styles</a>
326</div>
327
328<div class="doc_text">
329
330<p>The first step to determining what type of <a
331href="WritingAnLLVMPass.html">LLVM pass</a> you need to use for your Alias
332Analysis. As is the case with most other analyses and transformations, the
333answer should be fairly obvious from what type of problem you are trying to
334solve:</p>
335
336<ol>
337 <li>If you require interprocedural analysis, it should be a
338 <tt>Pass</tt>.</li>
339 <li>If you are a function-local analysis, subclass <tt>FunctionPass</tt>.</li>
340 <li>If you don't need to look at the program at all, subclass
341 <tt>ImmutablePass</tt>.</li>
342</ol>
343
344<p>In addition to the pass that you subclass, you should also inherit from the
345<tt>AliasAnalysis</tt> interface, of course, and use the
346<tt>RegisterAnalysisGroup</tt> template to register as an implementation of
347<tt>AliasAnalysis</tt>.</p>
348
349</div>
350
351<!-- ======================================================================= -->
352<div class="doc_subsection">
353 <a name="requiredcalls">Required initialization calls</a>
354</div>
355
356<div class="doc_text">
357
358<p>Your subclass of <tt>AliasAnalysis</tt> is required to invoke two methods on
359the <tt>AliasAnalysis</tt> base class: <tt>getAnalysisUsage</tt> and
360<tt>InitializeAliasAnalysis</tt>. In particular, your implementation of
361<tt>getAnalysisUsage</tt> should explicitly call into the
362<tt>AliasAnalysis::getAnalysisUsage</tt> method in addition to doing any
363declaring any pass dependencies your pass has. Thus you should have something
364like this:</p>
365
366<div class="doc_code">
367<pre>
368void getAnalysisUsage(AnalysisUsage &amp;AU) const {
369 AliasAnalysis::getAnalysisUsage(AU);
370 <i>// declare your dependencies here.</i>
371}
372</pre>
373</div>
374
375<p>Additionally, your must invoke the <tt>InitializeAliasAnalysis</tt> method
376from your analysis run method (<tt>run</tt> for a <tt>Pass</tt>,
377<tt>runOnFunction</tt> for a <tt>FunctionPass</tt>, or <tt>InitializePass</tt>
378for an <tt>ImmutablePass</tt>). For example (as part of a <tt>Pass</tt>):</p>
379
380<div class="doc_code">
381<pre>
382bool run(Module &amp;M) {
383 InitializeAliasAnalysis(this);
384 <i>// Perform analysis here...</i>
385 return false;
386}
387</pre>
388</div>
389
390</div>
391
392<!-- ======================================================================= -->
393<div class="doc_subsection">
394 <a name="interfaces">Interfaces which may be specified</a>
395</div>
396
397<div class="doc_text">
398
399<p>All of the <a
400href="/doxygen/classllvm_1_1AliasAnalysis.html"><tt>AliasAnalysis</tt></a>
401virtual methods default to providing <a href="#chaining">chaining</a> to another
402alias analysis implementation, which ends up returning conservatively correct
403information (returning "May" Alias and "Mod/Ref" for alias and mod/ref queries
404respectively). Depending on the capabilities of the analysis you are
405implementing, you just override the interfaces you can improve.</p>
406
407</div>
408
409
410
411<!-- ======================================================================= -->
412<div class="doc_subsection">
413 <a name="chaining"><tt>AliasAnalysis</tt> chaining behavior</a>
414</div>
415
416<div class="doc_text">
417
418<p>With only two special exceptions (the <tt><a
419href="#basic-aa">basicaa</a></tt> and <a href="#no-aa"><tt>no-aa</tt></a>
420passes) every alias analysis pass chains to another alias analysis
421implementation (for example, the user can specify "<tt>-basicaa -ds-aa
Chris Lattner5dd92052010-03-01 19:24:17 +0000[diff] [blame]422-licm</tt>" to get the maximum benefit from both alias
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]423analyses). The alias analysis class automatically takes care of most of this
424for methods that you don't override. For methods that you do override, in code
425paths that return a conservative MayAlias or Mod/Ref result, simply return
426whatever the superclass computes. For example:</p>
427
428<div class="doc_code">
429<pre>
430AliasAnalysis::AliasResult alias(const Value *V1, unsigned V1Size,
431 const Value *V2, unsigned V2Size) {
432 if (...)
433 return NoAlias;
434 ...
435
436 <i>// Couldn't determine a must or no-alias result.</i>
437 return AliasAnalysis::alias(V1, V1Size, V2, V2Size);
438}
439</pre>
440</div>
441
442<p>In addition to analysis queries, you must make sure to unconditionally pass
443LLVM <a href="#updating">update notification</a> methods to the superclass as
444well if you override them, which allows all alias analyses in a change to be
445updated.</p>
446
447</div>
448
449
450<!-- ======================================================================= -->
451<div class="doc_subsection">
452 <a name="updating">Updating analysis results for transformations</a>
453</div>
454
455<div class="doc_text">
456<p>
457Alias analysis information is initially computed for a static snapshot of the
458program, but clients will use this information to make transformations to the
459code. All but the most trivial forms of alias analysis will need to have their
460analysis results updated to reflect the changes made by these transformations.
461</p>
462
463<p>
464The <tt>AliasAnalysis</tt> interface exposes two methods which are used to
465communicate program changes from the clients to the analysis implementations.
466Various alias analysis implementations should use these methods to ensure that
467their internal data structures are kept up-to-date as the program changes (for
468example, when an instruction is deleted), and clients of alias analysis must be
469sure to call these interfaces appropriately.
470</p>
471</div>
472
473<!-- _______________________________________________________________________ -->
474<div class="doc_subsubsection">The <tt>deleteValue</tt> method</div>
475
476<div class="doc_text">
477The <tt>deleteValue</tt> method is called by transformations when they remove an
478instruction or any other value from the program (including values that do not
479use pointers). Typically alias analyses keep data structures that have entries
480for each value in the program. When this method is called, they should remove
481any entries for the specified value, if they exist.
482</div>
483
484<!-- _______________________________________________________________________ -->
485<div class="doc_subsubsection">The <tt>copyValue</tt> method</div>
486
487<div class="doc_text">
488The <tt>copyValue</tt> method is used when a new value is introduced into the
489program. There is no way to introduce a value into the program that did not
490exist before (this doesn't make sense for a safe compiler transformation), so
491this is the only way to introduce a new value. This method indicates that the
492new value has exactly the same properties as the value being copied.
493</div>
494
495<!-- _______________________________________________________________________ -->
496<div class="doc_subsubsection">The <tt>replaceWithNewValue</tt> method</div>
497
498<div class="doc_text">
499This method is a simple helper method that is provided to make clients easier to
500use. It is implemented by copying the old analysis information to the new
501value, then deleting the old value. This method cannot be overridden by alias
502analysis implementations.
503</div>
504
505<!-- ======================================================================= -->
506<div class="doc_subsection">
507 <a name="implefficiency">Efficiency Issues</a>
508</div>
509
510<div class="doc_text">
511
512<p>From the LLVM perspective, the only thing you need to do to provide an
513efficient alias analysis is to make sure that alias analysis <b>queries</b> are
514serviced quickly. The actual calculation of the alias analysis results (the
515"run" method) is only performed once, but many (perhaps duplicate) queries may
516be performed. Because of this, try to move as much computation to the run
517method as possible (within reason).</p>
518
519</div>
520
Dan Gohmanaa785442010-06-24 19:34:03 +0000[diff] [blame]521<!-- ======================================================================= -->
522<div class="doc_subsection">
523 <a name="passmanager">Pass Manager Issues</a>
524</div>
525
526<div class="doc_text">
527
528<p>PassManager support for alternative AliasAnalysis implementation
529has some issues.</p>
530
531<p>There is no way to override the default alias analysis. It would
532be very useful to be able to do something like "opt -my-aa -O2" and
533have it use -my-aa for all passes which need AliasAnalysis, but there
534is currently no support for that, short of changing the source code
535and recompiling. Similarly, there is also no way of setting a chain
536of analyses as the default.</p>
537
538<p>There is no way for transform passes to declare that they preserve
539<tt>AliasAnalysis</tt> implementations. The <tt>AliasAnalysis</tt>
540interface includes <tt>deleteValue</tt> and <tt>copyValue</tt> methods
541which are intended to allow a pass to keep an AliasAnalysis consistent,
542however there's no way for a pass to declare in its
543<tt>getAnalysisUsage</tt> that it does so. Some passes attempt to use
544<tt>AU.addPreserved&lt;AliasAnalysis&gt;</tt>, however this doesn't
545actually have any effect.</tt>
546
547<p><tt>AliasAnalysisCounter</tt> (<tt>-count-aa</tt>) and <tt>AliasDebugger</tt>
548(<tt>-debug-aa</tt>) are implemented as <tt>ModulePass</tt> classes, so if your
549alias analysis uses <tt>FunctionPass</tt>, it won't be able to use
550these utilities. If you try to use them, the pass manager will
551silently route alias analysis queries directly to
552<tt>BasicAliasAnalysis</tt> instead.</p>
553
554<p>Similarly, the <tt>opt -p</tt> option introduces <tt>ModulePass</tt>
555passes between each pass, which prevents the use of <tt>FunctionPass</tt>
556alias analysis passes.</p>
557
558</div>
559
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]560<!-- *********************************************************************** -->
561<div class="doc_section">
562 <a name="using">Using alias analysis results</a>
563</div>
564<!-- *********************************************************************** -->
565
566<div class="doc_text">
567
568<p>There are several different ways to use alias analysis results. In order of
569preference, these are...</p>
570
571</div>
572
573<!-- ======================================================================= -->
574<div class="doc_subsection">
Chris Lattner72da5762009-04-25 21:11:37 +0000[diff] [blame]575 <a name="memdep">Using the <tt>MemoryDependenceAnalysis</tt> Pass</a>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]576</div>
577
578<div class="doc_text">
579
Chris Lattner72da5762009-04-25 21:11:37 +0000[diff] [blame]580<p>The <tt>memdep</tt> pass uses alias analysis to provide high-level dependence
581information about memory-using instructions. This will tell you which store
582feeds into a load, for example. It uses caching and other techniques to be
583efficient, and is used by Dead Store Elimination, GVN, and memcpy optimizations.
584</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]585
586</div>
587
588<!-- ======================================================================= -->
589<div class="doc_subsection">
590 <a name="ast">Using the <tt>AliasSetTracker</tt> class</a>
591</div>
592
593<div class="doc_text">
594
595<p>Many transformations need information about alias <b>sets</b> that are active
596in some scope, rather than information about pairwise aliasing. The <tt><a
597href="/doxygen/classllvm_1_1AliasSetTracker.html">AliasSetTracker</a></tt> class
598is used to efficiently build these Alias Sets from the pairwise alias analysis
599information provided by the <tt>AliasAnalysis</tt> interface.</p>
600
601<p>First you initialize the AliasSetTracker by using the "<tt>add</tt>" methods
602to add information about various potentially aliasing instructions in the scope
603you are interested in. Once all of the alias sets are completed, your pass
604should simply iterate through the constructed alias sets, using the
605<tt>AliasSetTracker</tt> <tt>begin()</tt>/<tt>end()</tt> methods.</p>
606
607<p>The <tt>AliasSet</tt>s formed by the <tt>AliasSetTracker</tt> are guaranteed
608to be disjoint, calculate mod/ref information and volatility for the set, and
609keep track of whether or not all of the pointers in the set are Must aliases.
610The AliasSetTracker also makes sure that sets are properly folded due to call
611instructions, and can provide a list of pointers in each set.</p>
612
613<p>As an example user of this, the <a href="/doxygen/structLICM.html">Loop
614Invariant Code Motion</a> pass uses <tt>AliasSetTracker</tt>s to calculate alias
615sets for each loop nest. If an <tt>AliasSet</tt> in a loop is not modified,
616then all load instructions from that set may be hoisted out of the loop. If any
617alias sets are stored to <b>and</b> are must alias sets, then the stores may be
618sunk to outside of the loop, promoting the memory location to a register for the
619duration of the loop nest. Both of these transformations only apply if the
620pointer argument is loop-invariant.</p>
621
622</div>
623
624<!-- _______________________________________________________________________ -->
625<div class="doc_subsubsection">
626 The AliasSetTracker implementation
627</div>
628
629<div class="doc_text">
630
631<p>The AliasSetTracker class is implemented to be as efficient as possible. It
632uses the union-find algorithm to efficiently merge AliasSets when a pointer is
633inserted into the AliasSetTracker that aliases multiple sets. The primary data
634structure is a hash table mapping pointers to the AliasSet they are in.</p>
635
636<p>The AliasSetTracker class must maintain a list of all of the LLVM Value*'s
637that are in each AliasSet. Since the hash table already has entries for each
638LLVM Value* of interest, the AliasesSets thread the linked list through these
639hash-table nodes to avoid having to allocate memory unnecessarily, and to make
640merging alias sets extremely efficient (the linked list merge is constant time).
641</p>
642
643<p>You shouldn't need to understand these details if you are just a client of
644the AliasSetTracker, but if you look at the code, hopefully this brief
645description will help make sense of why things are designed the way they
646are.</p>
647
648</div>
649
650<!-- ======================================================================= -->
651<div class="doc_subsection">
652 <a name="direct">Using the <tt>AliasAnalysis</tt> interface directly</a>
653</div>
654
655<div class="doc_text">
656
657<p>If neither of these utility class are what your pass needs, you should use
658the interfaces exposed by the <tt>AliasAnalysis</tt> class directly. Try to use
659the higher-level methods when possible (e.g., use mod/ref information instead of
660the <a href="#alias"><tt>alias</tt></a> method directly if possible) to get the
661best precision and efficiency.</p>
662
663</div>
664
665<!-- *********************************************************************** -->
666<div class="doc_section">
667 <a name="exist">Existing alias analysis implementations and clients</a>
668</div>
669<!-- *********************************************************************** -->
670
671<div class="doc_text">
672
673<p>If you're going to be working with the LLVM alias analysis infrastructure,
674you should know what clients and implementations of alias analysis are
675available. In particular, if you are implementing an alias analysis, you should
676be aware of the <a href="#aliasanalysis-debug">the clients</a> that are useful
677for monitoring and evaluating different implementations.</p>
678
679</div>
680
681<!-- ======================================================================= -->
682<div class="doc_subsection">
683 <a name="impls">Available <tt>AliasAnalysis</tt> implementations</a>
684</div>
685
686<div class="doc_text">
687
688<p>This section lists the various implementations of the <tt>AliasAnalysis</tt>
689interface. With the exception of the <a href="#no-aa"><tt>-no-aa</tt></a> and
690<a href="#basic-aa"><tt>-basicaa</tt></a> implementations, all of these <a
691href="#chaining">chain</a> to other alias analysis implementations.</p>
692
693</div>
694
695<!-- _______________________________________________________________________ -->
696<div class="doc_subsubsection">
697 <a name="no-aa">The <tt>-no-aa</tt> pass</a>
698</div>
699
700<div class="doc_text">
701
702<p>The <tt>-no-aa</tt> pass is just like what it sounds: an alias analysis that
703never returns any useful information. This pass can be useful if you think that
704alias analysis is doing something wrong and are trying to narrow down a
705problem.</p>
706
707</div>
708
709<!-- _______________________________________________________________________ -->
710<div class="doc_subsubsection">
711 <a name="basic-aa">The <tt>-basicaa</tt> pass</a>
712</div>
713
714<div class="doc_text">
715
716<p>The <tt>-basicaa</tt> pass is the default LLVM alias analysis. It is an
717aggressive local analysis that "knows" many important facts:</p>
718
719<ul>
720<li>Distinct globals, stack allocations, and heap allocations can never
721 alias.</li>
722<li>Globals, stack allocations, and heap allocations never alias the null
723 pointer.</li>
724<li>Different fields of a structure do not alias.</li>
725<li>Indexes into arrays with statically differing subscripts cannot alias.</li>
726<li>Many common standard C library functions <a
727 href="#simplemodref">never access memory or only read memory</a>.</li>
728<li>Pointers that obviously point to constant globals
729 "<tt>pointToConstantMemory</tt>".</li>
730<li>Function calls can not modify or references stack allocations if they never
731 escape from the function that allocates them (a common case for automatic
732 arrays).</li>
733</ul>
734
735</div>
736
737<!-- _______________________________________________________________________ -->
738<div class="doc_subsubsection">
739 <a name="globalsmodref">The <tt>-globalsmodref-aa</tt> pass</a>
740</div>
741
742<div class="doc_text">
743
744<p>This pass implements a simple context-sensitive mod/ref and alias analysis
745for internal global variables that don't "have their address taken". If a
746global does not have its address taken, the pass knows that no pointers alias
747the global. This pass also keeps track of functions that it knows never access
Chris Lattner72da5762009-04-25 21:11:37 +0000[diff] [blame]748memory or never read memory. This allows certain optimizations (e.g. GVN) to
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]749eliminate call instructions entirely.
750</p>
751
752<p>The real power of this pass is that it provides context-sensitive mod/ref
753information for call instructions. This allows the optimizer to know that
754calls to a function do not clobber or read the value of the global, allowing
755loads and stores to be eliminated.</p>
756
757<p>Note that this pass is somewhat limited in its scope (only support
758non-address taken globals), but is very quick analysis.</p>
759</div>
760
761<!-- _______________________________________________________________________ -->
762<div class="doc_subsubsection">
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]763 <a name="steens-aa">The <tt>-steens-aa</tt> pass</a>
764</div>
765
766<div class="doc_text">
767
768<p>The <tt>-steens-aa</tt> pass implements a variation on the well-known
769"Steensgaard's algorithm" for interprocedural alias analysis. Steensgaard's
770algorithm is a unification-based, flow-insensitive, context-insensitive, and
771field-insensitive alias analysis that is also very scalable (effectively linear
772time).</p>
773
774<p>The LLVM <tt>-steens-aa</tt> pass implements a "speculatively
775field-<b>sensitive</b>" version of Steensgaard's algorithm using the Data
776Structure Analysis framework. This gives it substantially more precision than
777the standard algorithm while maintaining excellent analysis scalability.</p>
778
779<p>Note that <tt>-steens-aa</tt> is available in the optional "poolalloc"
780module, it is not part of the LLVM core.</p>
781
782</div>
783
784<!-- _______________________________________________________________________ -->
785<div class="doc_subsubsection">
786 <a name="ds-aa">The <tt>-ds-aa</tt> pass</a>
787</div>
788
789<div class="doc_text">
790
791<p>The <tt>-ds-aa</tt> pass implements the full Data Structure Analysis
792algorithm. Data Structure Analysis is a modular unification-based,
793flow-insensitive, context-<b>sensitive</b>, and speculatively
794field-<b>sensitive</b> alias analysis that is also quite scalable, usually at
795O(n*log(n)).</p>
796
797<p>This algorithm is capable of responding to a full variety of alias analysis
798queries, and can provide context-sensitive mod/ref information as well. The
799only major facility not implemented so far is support for must-alias
800information.</p>
801
802<p>Note that <tt>-ds-aa</tt> is available in the optional "poolalloc"
803module, it is not part of the LLVM core.</p>
804
805</div>
806
Dan Gohman3e6ba362010-06-28 22:09:52 +0000[diff] [blame]807<!-- _______________________________________________________________________ -->
808<div class="doc_subsubsection">
809 <a name="scev-aa">The <tt>-scev-aa</tt> pass</a>
810</div>
811
812<div class="doc_text">
813
814<p>The <tt>-scev-aa</tt> pass implements AliasAnalysis queries by
815translating them into ScalarEvolution queries. This gives it a
816more complete understanding of <tt>getelementptr</tt> instructions
817and loop induction variables than other alias analyses have.</p>
818
819</div>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]820
821<!-- ======================================================================= -->
822<div class="doc_subsection">
823 <a name="aliasanalysis-xforms">Alias analysis driven transformations</a>
824</div>
825
826<div class="doc_text">
827LLVM includes several alias-analysis driven transformations which can be used
828with any of the implementations above.
829</div>
830
831<!-- _______________________________________________________________________ -->
832<div class="doc_subsubsection">
833 <a name="adce">The <tt>-adce</tt> pass</a>
834</div>
835
836<div class="doc_text">
837
838<p>The <tt>-adce</tt> pass, which implements Aggressive Dead Code Elimination
839uses the <tt>AliasAnalysis</tt> interface to delete calls to functions that do
840not have side-effects and are not used.</p>
841
842</div>
843
844
845<!-- _______________________________________________________________________ -->
846<div class="doc_subsubsection">
847 <a name="licm">The <tt>-licm</tt> pass</a>
848</div>
849
850<div class="doc_text">
851
852<p>The <tt>-licm</tt> pass implements various Loop Invariant Code Motion related
853transformations. It uses the <tt>AliasAnalysis</tt> interface for several
854different transformations:</p>
855
856<ul>
857<li>It uses mod/ref information to hoist or sink load instructions out of loops
858if there are no instructions in the loop that modifies the memory loaded.</li>
859
860<li>It uses mod/ref information to hoist function calls out of loops that do not
861write to memory and are loop-invariant.</li>
862
863<li>If uses alias information to promote memory objects that are loaded and
864stored to in loops to live in a register instead. It can do this if there are
865no may aliases to the loaded/stored memory location.</li>
866</ul>
867
868</div>
869
870<!-- _______________________________________________________________________ -->
871<div class="doc_subsubsection">
872 <a name="argpromotion">The <tt>-argpromotion</tt> pass</a>
873</div>
874
875<div class="doc_text">
876<p>
877The <tt>-argpromotion</tt> pass promotes by-reference arguments to be passed in
878by-value instead. In particular, if pointer arguments are only loaded from it
879passes in the value loaded instead of the address to the function. This pass
880uses alias information to make sure that the value loaded from the argument
881pointer is not modified between the entry of the function and any load of the
882pointer.</p>
883</div>
884
885<!-- _______________________________________________________________________ -->
886<div class="doc_subsubsection">
Chris Lattner72da5762009-04-25 21:11:37 +0000[diff] [blame]887 <a name="gvn">The <tt>-gvn</tt>, <tt>-memcpyopt</tt>, and <tt>-dse</tt>
888 passes</a>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]889</div>
890
891<div class="doc_text">
892
Chris Lattner72da5762009-04-25 21:11:37 +0000[diff] [blame]893<p>These passes use AliasAnalysis information to reason about loads and stores.
894</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]895
896</div>
897
898<!-- ======================================================================= -->
899<div class="doc_subsection">
900 <a name="aliasanalysis-debug">Clients for debugging and evaluation of
901 implementations</a>
902</div>
903
904<div class="doc_text">
905
906<p>These passes are useful for evaluating the various alias analysis
Chris Lattner5dd92052010-03-01 19:24:17 +0000[diff] [blame]907implementations. You can use them with commands like '<tt>opt -ds-aa
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]908-aa-eval foo.bc -disable-output -stats</tt>'.</p>
909
910</div>
911
912<!-- _______________________________________________________________________ -->
913<div class="doc_subsubsection">
914 <a name="print-alias-sets">The <tt>-print-alias-sets</tt> pass</a>
915</div>
916
917<div class="doc_text">
918
919<p>The <tt>-print-alias-sets</tt> pass is exposed as part of the
920<tt>opt</tt> tool to print out the Alias Sets formed by the <a
921href="#ast"><tt>AliasSetTracker</tt></a> class. This is useful if you're using
922the <tt>AliasSetTracker</tt> class. To use it, use something like:</p>
923
924<div class="doc_code">
925<pre>
926% opt -ds-aa -print-alias-sets -disable-output
927</pre>
928</div>
929
930</div>
931
932
933<!-- _______________________________________________________________________ -->
934<div class="doc_subsubsection">
935 <a name="count-aa">The <tt>-count-aa</tt> pass</a>
936</div>
937
938<div class="doc_text">
939
940<p>The <tt>-count-aa</tt> pass is useful to see how many queries a particular
941pass is making and what responses are returned by the alias analysis. As an
942example,</p>
943
944<div class="doc_code">
945<pre>
946% opt -basicaa -count-aa -ds-aa -count-aa -licm
947</pre>
948</div>
949
950<p>will print out how many queries (and what responses are returned) by the
951<tt>-licm</tt> pass (of the <tt>-ds-aa</tt> pass) and how many queries are made
952of the <tt>-basicaa</tt> pass by the <tt>-ds-aa</tt> pass. This can be useful
953when debugging a transformation or an alias analysis implementation.</p>
954
955</div>
956
957<!-- _______________________________________________________________________ -->
958<div class="doc_subsubsection">
959 <a name="aa-eval">The <tt>-aa-eval</tt> pass</a>
960</div>
961
962<div class="doc_text">
963
964<p>The <tt>-aa-eval</tt> pass simply iterates through all pairs of pointers in a
965function and asks an alias analysis whether or not the pointers alias. This
966gives an indication of the precision of the alias analysis. Statistics are
967printed indicating the percent of no/may/must aliases found (a more precise
968algorithm will have a lower number of may aliases).</p>
969
970</div>
971
972<!-- *********************************************************************** -->
Owen Anderson05e080f2007-10-02 00:43:25 +0000[diff] [blame]973<div class="doc_section">
974 <a name="memdep">Memory Dependence Analysis</a>
975</div>
976<!-- *********************************************************************** -->
977
978<div class="doc_text">
979
980<p>If you're just looking to be a client of alias analysis information, consider
981using the Memory Dependence Analysis interface instead. MemDep is a lazy,
982caching layer on top of alias analysis that is able to answer the question of
983what preceding memory operations a given instruction depends on, either at an
984intra- or inter-block level. Because of its laziness and caching
985policy, using MemDep can be a significant performance win over accessing alias
986analysis directly.</p>
987
988</div>
989
990<!-- *********************************************************************** -->
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]991
992<hr>
993<address>
994 <a href="http://jigsaw.w3.org/css-validator/check/referer"><img
Misha Brukman947321d2008-12-11 17:34:48 +0000[diff] [blame]995 src="http://jigsaw.w3.org/css-validator/images/vcss-blue" alt="Valid CSS"></a>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]996 <a href="http://validator.w3.org/check/referer"><img
Misha Brukman947321d2008-12-11 17:34:48 +0000[diff] [blame]997 src="http://www.w3.org/Icons/valid-html401-blue" alt="Valid HTML 4.01"></a>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]998
999 <a href="mailto:sabre@nondot.org">Chris Lattner</a><br>
1000 <a href="http://llvm.org">LLVM Compiler Infrastructure</a><br>
1001 Last modified: $Date$
1002</address>
1003
1004</body>
1005</html>