1 <!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.01//EN"
2 "http://www.w3.org/TR/html4/strict.dtd">
5 <title>LLVM Alias Analysis Infrastructure</title>
6 <link rel="stylesheet" href="llvm.css" type="text/css">
10 <div class="doc_title">
11 LLVM Alias Analysis Infrastructure
15 <li><a href="#introduction">Introduction</a></li>
17 <li><a href="#overview"><tt>AliasAnalysis</tt> Class Overview</a>
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>
26 <li><a href="#writingnew">Writing a new <tt>AliasAnalysis</tt> Implementation</a>
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>
34 <li><a href="#limitations">Limitations</a></li>
38 <li><a href="#using">Using alias analysis results</a>
40 <li><a href="#memdep">Using the <tt>MemoryDependenceAnalysis</tt> Pass</a></li>
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>
46 <li><a href="#exist">Existing alias analysis implementations and clients</a>
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>
54 <li><a href="#memdep">Memory Dependence Analysis</a></li>
57 <div class="doc_author">
58 <p>Written by <a href="mailto:sabre@nondot.org">Chris Lattner</a></p>
61 <!-- *********************************************************************** -->
62 <div class="doc_section">
63 <a name="introduction">Introduction</a>
65 <!-- *********************************************************************** -->
67 <div class="doc_text">
69 <p>Alias Analysis (aka Pointer Analysis) is a class of techniques which attempt
70 to determine whether or not two pointers ever can point to the same object in
71 memory. There are many different algorithms for alias analysis and many
72 different ways of classifying them: flow-sensitive vs flow-insensitive,
73 context-sensitive vs context-insensitive, field-sensitive vs field-insensitive,
74 unification-based vs subset-based, etc. Traditionally, alias analyses respond
75 to a query with a <a href="#MustMayNo">Must, May, or No</a> alias response,
76 indicating that two pointers always point to the same object, might point to the
77 same object, or are known to never point to the same object.</p>
80 href="http://llvm.org/doxygen/classllvm_1_1AliasAnalysis.html"><tt>AliasAnalysis</tt></a>
81 class is the primary interface used by clients and implementations of alias
82 analyses in the LLVM system. This class is the common interface between clients
83 of alias analysis information and the implementations providing it, and is
84 designed to support a wide range of implementations and clients (but currently
85 all clients are assumed to be flow-insensitive). In addition to simple alias
86 analysis information, this class exposes Mod/Ref information from those
87 implementations which can provide it, allowing for powerful analyses and
88 transformations to work well together.</p>
90 <p>This document contains information necessary to successfully implement this
91 interface, use it, and to test both sides. It also explains some of the finer
92 points about what exactly results mean. If you feel that something is unclear
93 or should be added, please <a href="mailto:sabre@nondot.org">let me
98 <!-- *********************************************************************** -->
99 <div class="doc_section">
100 <a name="overview"><tt>AliasAnalysis</tt> Class Overview</a>
102 <!-- *********************************************************************** -->
104 <div class="doc_text">
107 href="http://llvm.org/doxygen/classllvm_1_1AliasAnalysis.html"><tt>AliasAnalysis</tt></a>
108 class defines the interface that the various alias analysis implementations
109 should support. This class exports two important enums: <tt>AliasResult</tt>
110 and <tt>ModRefResult</tt> which represent the result of an alias query or a
111 mod/ref query, respectively.</p>
113 <p>The <tt>AliasAnalysis</tt> interface exposes information about memory,
114 represented in several different ways. In particular, memory objects are
115 represented as a starting address and size, and function calls are represented
116 as the actual <tt>call</tt> or <tt>invoke</tt> instructions that performs the
117 call. The <tt>AliasAnalysis</tt> interface also exposes some helper methods
118 which allow you to get mod/ref information for arbitrary instructions.</p>
120 <p>All <tt>AliasAnalysis</tt> interfaces require that in queries involving
121 multiple values, values which are not
122 <a href="LangRef.html#constants">constants</a> are all defined within the
127 <!-- ======================================================================= -->
128 <div class="doc_subsection">
129 <a name="pointers">Representation of Pointers</a>
132 <div class="doc_text">
134 <p>Most importantly, the <tt>AliasAnalysis</tt> class provides several methods
135 which are used to query whether or not two memory objects alias, whether
136 function calls can modify or read a memory object, etc. For all of these
137 queries, memory objects are represented as a pair of their starting address (a
138 symbolic LLVM <tt>Value*</tt>) and a static size.</p>
140 <p>Representing memory objects as a starting address and a size is critically
141 important for correct Alias Analyses. For example, consider this (silly, but
142 possible) C code:</p>
144 <div class="doc_code">
150 for (i = 0; i != 10; ++i) {
151 C[0] = A[i]; /* One byte store */
152 C[1] = A[9-i]; /* One byte store */
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
159 locations one byte apart, and the accesses are each one byte. In this case, the
160 LICM pass can use store motion to remove the stores from the loop. In
161 constrast, the following code:</p>
163 <div class="doc_code">
169 for (i = 0; i != 10; ++i) {
170 ((short*)C)[0] = A[i]; /* Two byte store! */
171 C[1] = A[9-i]; /* One byte store */
176 <p>In this case, the two stores to C do alias each other, because the access to
177 the <tt>&C[0]</tt> element is a two byte access. If size information wasn't
178 available in the query, even the first case would have to conservatively assume
179 that the accesses alias.</p>
183 <!-- ======================================================================= -->
184 <div class="doc_subsection">
185 <a name="alias">The <tt>alias</tt> method</a>
188 <div class="doc_text">
189 <p>The <tt>alias</tt> method is the primary interface used to determine whether
190 or not two memory objects alias each other. It takes two memory objects as
191 input and returns MustAlias, PartialAlias, MayAlias, or NoAlias as
194 <p>Like all <tt>AliasAnalysis</tt> interfaces, the <tt>alias</tt> method requires
195 that either the two pointer values be defined within the same function, or at
196 least one of the values is a <a href="LangRef.html#constants">constant</a>.</p>
199 <!-- _______________________________________________________________________ -->
200 <div class="doc_subsubsection">
201 <a name="MustMayNo">Must, May, and No Alias Responses</a>
204 <div class="doc_text">
205 <p>The NoAlias response may be used when there is never an immediate dependence
206 between any memory reference <i>based</i> on one pointer and any memory
207 reference <i>based</i> the other. The most obvious example is when the two
208 pointers point to non-overlapping memory ranges. Another is when the two
209 pointers are only ever used for reading memory. Another is when the memory is
210 freed and reallocated between accesses through one pointer and accesses through
211 the other -- in this case, there is a dependence, but it's mediated by the free
212 and reallocation.</p>
214 <p>As an exception to this is with the
215 <a href="LangRef.html#noalias"><tt>noalias</tt></a> keyword; the "irrelevant"
216 dependencies are ignored.</p>
218 <p>The MayAlias response is used whenever the two pointers might refer to the
221 <p>The PartialAlias response is used when the two memory objects are known
222 to be overlapping in some way, but do not start at the same address.</p>
224 <p>The MustAlias response may only be returned if the two memory objects are
225 guaranteed to always start at exactly the same location. A MustAlias response
226 implies that the pointers compare equal.</p>
230 <!-- ======================================================================= -->
231 <div class="doc_subsection">
232 <a name="ModRefInfo">The <tt>getModRefInfo</tt> methods</a>
235 <div class="doc_text">
237 <p>The <tt>getModRefInfo</tt> methods return information about whether the
238 execution of an instruction can read or modify a memory location. Mod/Ref
239 information is always conservative: if an instruction <b>might</b> read or write
240 a location, ModRef is returned.</p>
242 <p>The <tt>AliasAnalysis</tt> class also provides a <tt>getModRefInfo</tt>
243 method for testing dependencies between function calls. This method takes two
244 call sites (CS1 & CS2), returns NoModRef if neither call writes to memory
245 read or written by the other, Ref if CS1 reads memory written by CS2, Mod if CS1
246 writes to memory read or written by CS2, or ModRef if CS1 might read or write
247 memory written to by CS2. Note that this relation is not commutative.</p>
252 <!-- ======================================================================= -->
253 <div class="doc_subsection">
254 <a name="OtherItfs">Other useful <tt>AliasAnalysis</tt> methods</a>
257 <div class="doc_text">
260 Several other tidbits of information are often collected by various alias
261 analysis implementations and can be put to good use by various clients.
266 <!-- _______________________________________________________________________ -->
267 <div class="doc_subsubsection">
268 The <tt>pointsToConstantMemory</tt> method
271 <div class="doc_text">
273 <p>The <tt>pointsToConstantMemory</tt> method returns true if and only if the
274 analysis can prove that the pointer only points to unchanging memory locations
275 (functions, constant global variables, and the null pointer). This information
276 can be used to refine mod/ref information: it is impossible for an unchanging
277 memory location to be modified.</p>
281 <!-- _______________________________________________________________________ -->
282 <div class="doc_subsubsection">
283 <a name="simplemodref">The <tt>doesNotAccessMemory</tt> and
284 <tt>onlyReadsMemory</tt> methods</a>
287 <div class="doc_text">
289 <p>These methods are used to provide very simple mod/ref information for
290 function calls. The <tt>doesNotAccessMemory</tt> method returns true for a
291 function if the analysis can prove that the function never reads or writes to
292 memory, or if the function only reads from constant memory. Functions with this
293 property are side-effect free and only depend on their input arguments, allowing
294 them to be eliminated if they form common subexpressions or be hoisted out of
295 loops. Many common functions behave this way (e.g., <tt>sin</tt> and
296 <tt>cos</tt>) but many others do not (e.g., <tt>acos</tt>, which modifies the
297 <tt>errno</tt> variable).</p>
299 <p>The <tt>onlyReadsMemory</tt> method returns true for a function if analysis
300 can prove that (at most) the function only reads from non-volatile memory.
301 Functions with this property are side-effect free, only depending on their input
302 arguments and the state of memory when they are called. This property allows
303 calls to these functions to be eliminated and moved around, as long as there is
304 no store instruction that changes the contents of memory. Note that all
305 functions that satisfy the <tt>doesNotAccessMemory</tt> method also satisfies
306 <tt>onlyReadsMemory</tt>.</p>
310 <!-- *********************************************************************** -->
311 <div class="doc_section">
312 <a name="writingnew">Writing a new <tt>AliasAnalysis</tt> Implementation</a>
314 <!-- *********************************************************************** -->
316 <div class="doc_text">
318 <p>Writing a new alias analysis implementation for LLVM is quite
319 straight-forward. There are already several implementations that you can use
320 for examples, and the following information should help fill in any details.
321 For a examples, take a look at the <a href="#impls">various alias analysis
322 implementations</a> included with LLVM.</p>
326 <!-- ======================================================================= -->
327 <div class="doc_subsection">
328 <a name="passsubclasses">Different Pass styles</a>
331 <div class="doc_text">
333 <p>The first step to determining what type of <a
334 href="WritingAnLLVMPass.html">LLVM pass</a> you need to use for your Alias
335 Analysis. As is the case with most other analyses and transformations, the
336 answer should be fairly obvious from what type of problem you are trying to
340 <li>If you require interprocedural analysis, it should be a
342 <li>If you are a function-local analysis, subclass <tt>FunctionPass</tt>.</li>
343 <li>If you don't need to look at the program at all, subclass
344 <tt>ImmutablePass</tt>.</li>
347 <p>In addition to the pass that you subclass, you should also inherit from the
348 <tt>AliasAnalysis</tt> interface, of course, and use the
349 <tt>RegisterAnalysisGroup</tt> template to register as an implementation of
350 <tt>AliasAnalysis</tt>.</p>
354 <!-- ======================================================================= -->
355 <div class="doc_subsection">
356 <a name="requiredcalls">Required initialization calls</a>
359 <div class="doc_text">
361 <p>Your subclass of <tt>AliasAnalysis</tt> is required to invoke two methods on
362 the <tt>AliasAnalysis</tt> base class: <tt>getAnalysisUsage</tt> and
363 <tt>InitializeAliasAnalysis</tt>. In particular, your implementation of
364 <tt>getAnalysisUsage</tt> should explicitly call into the
365 <tt>AliasAnalysis::getAnalysisUsage</tt> method in addition to doing any
366 declaring any pass dependencies your pass has. Thus you should have something
369 <div class="doc_code">
371 void getAnalysisUsage(AnalysisUsage &AU) const {
372 AliasAnalysis::getAnalysisUsage(AU);
373 <i>// declare your dependencies here.</i>
378 <p>Additionally, your must invoke the <tt>InitializeAliasAnalysis</tt> method
379 from your analysis run method (<tt>run</tt> for a <tt>Pass</tt>,
380 <tt>runOnFunction</tt> for a <tt>FunctionPass</tt>, or <tt>InitializePass</tt>
381 for an <tt>ImmutablePass</tt>). For example (as part of a <tt>Pass</tt>):</p>
383 <div class="doc_code">
385 bool run(Module &M) {
386 InitializeAliasAnalysis(this);
387 <i>// Perform analysis here...</i>
395 <!-- ======================================================================= -->
396 <div class="doc_subsection">
397 <a name="interfaces">Interfaces which may be specified</a>
400 <div class="doc_text">
403 href="/doxygen/classllvm_1_1AliasAnalysis.html"><tt>AliasAnalysis</tt></a>
404 virtual methods default to providing <a href="#chaining">chaining</a> to another
405 alias analysis implementation, which ends up returning conservatively correct
406 information (returning "May" Alias and "Mod/Ref" for alias and mod/ref queries
407 respectively). Depending on the capabilities of the analysis you are
408 implementing, you just override the interfaces you can improve.</p>
414 <!-- ======================================================================= -->
415 <div class="doc_subsection">
416 <a name="chaining"><tt>AliasAnalysis</tt> chaining behavior</a>
419 <div class="doc_text">
421 <p>With only two special exceptions (the <tt><a
422 href="#basic-aa">basicaa</a></tt> and <a href="#no-aa"><tt>no-aa</tt></a>
423 passes) every alias analysis pass chains to another alias analysis
424 implementation (for example, the user can specify "<tt>-basicaa -ds-aa
425 -licm</tt>" to get the maximum benefit from both alias
426 analyses). The alias analysis class automatically takes care of most of this
427 for methods that you don't override. For methods that you do override, in code
428 paths that return a conservative MayAlias or Mod/Ref result, simply return
429 whatever the superclass computes. For example:</p>
431 <div class="doc_code">
433 AliasAnalysis::AliasResult alias(const Value *V1, unsigned V1Size,
434 const Value *V2, unsigned V2Size) {
439 <i>// Couldn't determine a must or no-alias result.</i>
440 return AliasAnalysis::alias(V1, V1Size, V2, V2Size);
445 <p>In addition to analysis queries, you must make sure to unconditionally pass
446 LLVM <a href="#updating">update notification</a> methods to the superclass as
447 well if you override them, which allows all alias analyses in a change to be
453 <!-- ======================================================================= -->
454 <div class="doc_subsection">
455 <a name="updating">Updating analysis results for transformations</a>
458 <div class="doc_text">
460 Alias analysis information is initially computed for a static snapshot of the
461 program, but clients will use this information to make transformations to the
462 code. All but the most trivial forms of alias analysis will need to have their
463 analysis results updated to reflect the changes made by these transformations.
467 The <tt>AliasAnalysis</tt> interface exposes two methods which are used to
468 communicate program changes from the clients to the analysis implementations.
469 Various alias analysis implementations should use these methods to ensure that
470 their internal data structures are kept up-to-date as the program changes (for
471 example, when an instruction is deleted), and clients of alias analysis must be
472 sure to call these interfaces appropriately.
476 <!-- _______________________________________________________________________ -->
477 <div class="doc_subsubsection">The <tt>deleteValue</tt> method</div>
479 <div class="doc_text">
480 The <tt>deleteValue</tt> method is called by transformations when they remove an
481 instruction or any other value from the program (including values that do not
482 use pointers). Typically alias analyses keep data structures that have entries
483 for each value in the program. When this method is called, they should remove
484 any entries for the specified value, if they exist.
487 <!-- _______________________________________________________________________ -->
488 <div class="doc_subsubsection">The <tt>copyValue</tt> method</div>
490 <div class="doc_text">
491 The <tt>copyValue</tt> method is used when a new value is introduced into the
492 program. There is no way to introduce a value into the program that did not
493 exist before (this doesn't make sense for a safe compiler transformation), so
494 this is the only way to introduce a new value. This method indicates that the
495 new value has exactly the same properties as the value being copied.
498 <!-- _______________________________________________________________________ -->
499 <div class="doc_subsubsection">The <tt>replaceWithNewValue</tt> method</div>
501 <div class="doc_text">
502 This method is a simple helper method that is provided to make clients easier to
503 use. It is implemented by copying the old analysis information to the new
504 value, then deleting the old value. This method cannot be overridden by alias
505 analysis implementations.
508 <!-- ======================================================================= -->
509 <div class="doc_subsection">
510 <a name="implefficiency">Efficiency Issues</a>
513 <div class="doc_text">
515 <p>From the LLVM perspective, the only thing you need to do to provide an
516 efficient alias analysis is to make sure that alias analysis <b>queries</b> are
517 serviced quickly. The actual calculation of the alias analysis results (the
518 "run" method) is only performed once, but many (perhaps duplicate) queries may
519 be performed. Because of this, try to move as much computation to the run
520 method as possible (within reason).</p>
524 <!-- ======================================================================= -->
525 <div class="doc_subsection">
526 <a name="limitations">Limitations</a>
529 <div class="doc_text">
531 <p>The AliasAnalysis infrastructure has several limitations which make
532 writing a new <tt>AliasAnalysis</tt> implementation difficult.</p>
534 <p>There is no way to override the default alias analysis. It would
535 be very useful to be able to do something like "opt -my-aa -O2" and
536 have it use -my-aa for all passes which need AliasAnalysis, but there
537 is currently no support for that, short of changing the source code
538 and recompiling. Similarly, there is also no way of setting a chain
539 of analyses as the default.</p>
541 <p>There is no way for transform passes to declare that they preserve
542 <tt>AliasAnalysis</tt> implementations. The <tt>AliasAnalysis</tt>
543 interface includes <tt>deleteValue</tt> and <tt>copyValue</tt> methods
544 which are intended to allow a pass to keep an AliasAnalysis consistent,
545 however there's no way for a pass to declare in its
546 <tt>getAnalysisUsage</tt> that it does so. Some passes attempt to use
547 <tt>AU.addPreserved<AliasAnalysis></tt>, however this doesn't
548 actually have any effect.</tt>
550 <p><tt>AliasAnalysisCounter</tt> (<tt>-count-aa</tt>) and <tt>AliasDebugger</tt>
551 (<tt>-debug-aa</tt>) are implemented as <tt>ModulePass</tt> classes, so if your
552 alias analysis uses <tt>FunctionPass</tt>, it won't be able to use
553 these utilities. If you try to use them, the pass manager will
554 silently route alias analysis queries directly to
555 <tt>BasicAliasAnalysis</tt> instead.</p>
557 <p>Similarly, the <tt>opt -p</tt> option introduces <tt>ModulePass</tt>
558 passes between each pass, which prevents the use of <tt>FunctionPass</tt>
559 alias analysis passes.</p>
561 <p>The <tt>AliasAnalysis</tt> API does have functions for notifying
562 implementations when values are deleted or copied, however these
563 aren't sufficient. There are many other ways that LLVM IR can be
564 modified which could be relevant to <tt>AliasAnalysis</tt>
565 implementations which can not be expressed.</p>
567 <p>The <tt>AliasAnalysisDebugger</tt> utility seems to suggest that
568 <tt>AliasAnalysis</tt> implementations can expect that they will be
569 informed of any relevant <tt>Value</tt> before it appears in an
570 alias query. However, popular clients such as <tt>GVN</tt> don't
571 support this, and are known to trigger errors when run with the
572 <tt>AliasAnalysisDebugger</tt>.</p>
574 <p>Due to several of the above limitations, the most obvious use for
575 the <tt>AliasAnalysisCounter</tt> utility, collecting stats on all
576 alias queries in a compilation, doesn't work, even if the
577 <tt>AliasAnalysis</tt> implementations don't use <tt>FunctionPass</tt>.
578 There's no way to set a default, much less a default sequence,
579 and there's no way to preserve it.</p>
581 <p>The <tt>AliasSetTracker</tt> class (which is used by <tt>LICM</tt>
582 makes a non-deterministic number of alias queries. This can cause stats
583 collected by <tt>AliasAnalysisCounter</tt> to have fluctuations among
584 identical runs, for example. Another consequence is that debugging
585 techniques involving pausing execution after a predetermined number
586 of queries can be unreliable.</p>
588 <p>Many alias queries can be reformulated in terms of other alias
589 queries. When multiple <tt>AliasAnalysis</tt> queries are chained together,
590 it would make sense to start those queries from the beginning of the chain,
591 with care taken to avoid infinite looping, however currently an
592 implementation which wants to do this can only start such queries
597 <!-- *********************************************************************** -->
598 <div class="doc_section">
599 <a name="using">Using alias analysis results</a>
601 <!-- *********************************************************************** -->
603 <div class="doc_text">
605 <p>There are several different ways to use alias analysis results. In order of
606 preference, these are...</p>
610 <!-- ======================================================================= -->
611 <div class="doc_subsection">
612 <a name="memdep">Using the <tt>MemoryDependenceAnalysis</tt> Pass</a>
615 <div class="doc_text">
617 <p>The <tt>memdep</tt> pass uses alias analysis to provide high-level dependence
618 information about memory-using instructions. This will tell you which store
619 feeds into a load, for example. It uses caching and other techniques to be
620 efficient, and is used by Dead Store Elimination, GVN, and memcpy optimizations.
625 <!-- ======================================================================= -->
626 <div class="doc_subsection">
627 <a name="ast">Using the <tt>AliasSetTracker</tt> class</a>
630 <div class="doc_text">
632 <p>Many transformations need information about alias <b>sets</b> that are active
633 in some scope, rather than information about pairwise aliasing. The <tt><a
634 href="/doxygen/classllvm_1_1AliasSetTracker.html">AliasSetTracker</a></tt> class
635 is used to efficiently build these Alias Sets from the pairwise alias analysis
636 information provided by the <tt>AliasAnalysis</tt> interface.</p>
638 <p>First you initialize the AliasSetTracker by using the "<tt>add</tt>" methods
639 to add information about various potentially aliasing instructions in the scope
640 you are interested in. Once all of the alias sets are completed, your pass
641 should simply iterate through the constructed alias sets, using the
642 <tt>AliasSetTracker</tt> <tt>begin()</tt>/<tt>end()</tt> methods.</p>
644 <p>The <tt>AliasSet</tt>s formed by the <tt>AliasSetTracker</tt> are guaranteed
645 to be disjoint, calculate mod/ref information and volatility for the set, and
646 keep track of whether or not all of the pointers in the set are Must aliases.
647 The AliasSetTracker also makes sure that sets are properly folded due to call
648 instructions, and can provide a list of pointers in each set.</p>
650 <p>As an example user of this, the <a href="/doxygen/structLICM.html">Loop
651 Invariant Code Motion</a> pass uses <tt>AliasSetTracker</tt>s to calculate alias
652 sets for each loop nest. If an <tt>AliasSet</tt> in a loop is not modified,
653 then all load instructions from that set may be hoisted out of the loop. If any
654 alias sets are stored to <b>and</b> are must alias sets, then the stores may be
655 sunk to outside of the loop, promoting the memory location to a register for the
656 duration of the loop nest. Both of these transformations only apply if the
657 pointer argument is loop-invariant.</p>
661 <!-- _______________________________________________________________________ -->
662 <div class="doc_subsubsection">
663 The AliasSetTracker implementation
666 <div class="doc_text">
668 <p>The AliasSetTracker class is implemented to be as efficient as possible. It
669 uses the union-find algorithm to efficiently merge AliasSets when a pointer is
670 inserted into the AliasSetTracker that aliases multiple sets. The primary data
671 structure is a hash table mapping pointers to the AliasSet they are in.</p>
673 <p>The AliasSetTracker class must maintain a list of all of the LLVM Value*'s
674 that are in each AliasSet. Since the hash table already has entries for each
675 LLVM Value* of interest, the AliasesSets thread the linked list through these
676 hash-table nodes to avoid having to allocate memory unnecessarily, and to make
677 merging alias sets extremely efficient (the linked list merge is constant time).
680 <p>You shouldn't need to understand these details if you are just a client of
681 the AliasSetTracker, but if you look at the code, hopefully this brief
682 description will help make sense of why things are designed the way they
687 <!-- ======================================================================= -->
688 <div class="doc_subsection">
689 <a name="direct">Using the <tt>AliasAnalysis</tt> interface directly</a>
692 <div class="doc_text">
694 <p>If neither of these utility class are what your pass needs, you should use
695 the interfaces exposed by the <tt>AliasAnalysis</tt> class directly. Try to use
696 the higher-level methods when possible (e.g., use mod/ref information instead of
697 the <a href="#alias"><tt>alias</tt></a> method directly if possible) to get the
698 best precision and efficiency.</p>
702 <!-- *********************************************************************** -->
703 <div class="doc_section">
704 <a name="exist">Existing alias analysis implementations and clients</a>
706 <!-- *********************************************************************** -->
708 <div class="doc_text">
710 <p>If you're going to be working with the LLVM alias analysis infrastructure,
711 you should know what clients and implementations of alias analysis are
712 available. In particular, if you are implementing an alias analysis, you should
713 be aware of the <a href="#aliasanalysis-debug">the clients</a> that are useful
714 for monitoring and evaluating different implementations.</p>
718 <!-- ======================================================================= -->
719 <div class="doc_subsection">
720 <a name="impls">Available <tt>AliasAnalysis</tt> implementations</a>
723 <div class="doc_text">
725 <p>This section lists the various implementations of the <tt>AliasAnalysis</tt>
726 interface. With the exception of the <a href="#no-aa"><tt>-no-aa</tt></a> and
727 <a href="#basic-aa"><tt>-basicaa</tt></a> implementations, all of these <a
728 href="#chaining">chain</a> to other alias analysis implementations.</p>
732 <!-- _______________________________________________________________________ -->
733 <div class="doc_subsubsection">
734 <a name="no-aa">The <tt>-no-aa</tt> pass</a>
737 <div class="doc_text">
739 <p>The <tt>-no-aa</tt> pass is just like what it sounds: an alias analysis that
740 never returns any useful information. This pass can be useful if you think that
741 alias analysis is doing something wrong and are trying to narrow down a
746 <!-- _______________________________________________________________________ -->
747 <div class="doc_subsubsection">
748 <a name="basic-aa">The <tt>-basicaa</tt> pass</a>
751 <div class="doc_text">
753 <p>The <tt>-basicaa</tt> pass is an aggressive local analysis that "knows"
754 many important facts:</p>
757 <li>Distinct globals, stack allocations, and heap allocations can never
759 <li>Globals, stack allocations, and heap allocations never alias the null
761 <li>Different fields of a structure do not alias.</li>
762 <li>Indexes into arrays with statically differing subscripts cannot alias.</li>
763 <li>Many common standard C library functions <a
764 href="#simplemodref">never access memory or only read memory</a>.</li>
765 <li>Pointers that obviously point to constant globals
766 "<tt>pointToConstantMemory</tt>".</li>
767 <li>Function calls can not modify or references stack allocations if they never
768 escape from the function that allocates them (a common case for automatic
774 <!-- _______________________________________________________________________ -->
775 <div class="doc_subsubsection">
776 <a name="globalsmodref">The <tt>-globalsmodref-aa</tt> pass</a>
779 <div class="doc_text">
781 <p>This pass implements a simple context-sensitive mod/ref and alias analysis
782 for internal global variables that don't "have their address taken". If a
783 global does not have its address taken, the pass knows that no pointers alias
784 the global. This pass also keeps track of functions that it knows never access
785 memory or never read memory. This allows certain optimizations (e.g. GVN) to
786 eliminate call instructions entirely.
789 <p>The real power of this pass is that it provides context-sensitive mod/ref
790 information for call instructions. This allows the optimizer to know that
791 calls to a function do not clobber or read the value of the global, allowing
792 loads and stores to be eliminated.</p>
794 <p>Note that this pass is somewhat limited in its scope (only support
795 non-address taken globals), but is very quick analysis.</p>
798 <!-- _______________________________________________________________________ -->
799 <div class="doc_subsubsection">
800 <a name="steens-aa">The <tt>-steens-aa</tt> pass</a>
803 <div class="doc_text">
805 <p>The <tt>-steens-aa</tt> pass implements a variation on the well-known
806 "Steensgaard's algorithm" for interprocedural alias analysis. Steensgaard's
807 algorithm is a unification-based, flow-insensitive, context-insensitive, and
808 field-insensitive alias analysis that is also very scalable (effectively linear
811 <p>The LLVM <tt>-steens-aa</tt> pass implements a "speculatively
812 field-<b>sensitive</b>" version of Steensgaard's algorithm using the Data
813 Structure Analysis framework. This gives it substantially more precision than
814 the standard algorithm while maintaining excellent analysis scalability.</p>
816 <p>Note that <tt>-steens-aa</tt> is available in the optional "poolalloc"
817 module, it is not part of the LLVM core.</p>
821 <!-- _______________________________________________________________________ -->
822 <div class="doc_subsubsection">
823 <a name="ds-aa">The <tt>-ds-aa</tt> pass</a>
826 <div class="doc_text">
828 <p>The <tt>-ds-aa</tt> pass implements the full Data Structure Analysis
829 algorithm. Data Structure Analysis is a modular unification-based,
830 flow-insensitive, context-<b>sensitive</b>, and speculatively
831 field-<b>sensitive</b> alias analysis that is also quite scalable, usually at
834 <p>This algorithm is capable of responding to a full variety of alias analysis
835 queries, and can provide context-sensitive mod/ref information as well. The
836 only major facility not implemented so far is support for must-alias
839 <p>Note that <tt>-ds-aa</tt> is available in the optional "poolalloc"
840 module, it is not part of the LLVM core.</p>
844 <!-- _______________________________________________________________________ -->
845 <div class="doc_subsubsection">
846 <a name="scev-aa">The <tt>-scev-aa</tt> pass</a>
849 <div class="doc_text">
851 <p>The <tt>-scev-aa</tt> pass implements AliasAnalysis queries by
852 translating them into ScalarEvolution queries. This gives it a
853 more complete understanding of <tt>getelementptr</tt> instructions
854 and loop induction variables than other alias analyses have.</p>
858 <!-- ======================================================================= -->
859 <div class="doc_subsection">
860 <a name="aliasanalysis-xforms">Alias analysis driven transformations</a>
863 <div class="doc_text">
864 LLVM includes several alias-analysis driven transformations which can be used
865 with any of the implementations above.
868 <!-- _______________________________________________________________________ -->
869 <div class="doc_subsubsection">
870 <a name="adce">The <tt>-adce</tt> pass</a>
873 <div class="doc_text">
875 <p>The <tt>-adce</tt> pass, which implements Aggressive Dead Code Elimination
876 uses the <tt>AliasAnalysis</tt> interface to delete calls to functions that do
877 not have side-effects and are not used.</p>
882 <!-- _______________________________________________________________________ -->
883 <div class="doc_subsubsection">
884 <a name="licm">The <tt>-licm</tt> pass</a>
887 <div class="doc_text">
889 <p>The <tt>-licm</tt> pass implements various Loop Invariant Code Motion related
890 transformations. It uses the <tt>AliasAnalysis</tt> interface for several
891 different transformations:</p>
894 <li>It uses mod/ref information to hoist or sink load instructions out of loops
895 if there are no instructions in the loop that modifies the memory loaded.</li>
897 <li>It uses mod/ref information to hoist function calls out of loops that do not
898 write to memory and are loop-invariant.</li>
900 <li>If uses alias information to promote memory objects that are loaded and
901 stored to in loops to live in a register instead. It can do this if there are
902 no may aliases to the loaded/stored memory location.</li>
907 <!-- _______________________________________________________________________ -->
908 <div class="doc_subsubsection">
909 <a name="argpromotion">The <tt>-argpromotion</tt> pass</a>
912 <div class="doc_text">
914 The <tt>-argpromotion</tt> pass promotes by-reference arguments to be passed in
915 by-value instead. In particular, if pointer arguments are only loaded from it
916 passes in the value loaded instead of the address to the function. This pass
917 uses alias information to make sure that the value loaded from the argument
918 pointer is not modified between the entry of the function and any load of the
922 <!-- _______________________________________________________________________ -->
923 <div class="doc_subsubsection">
924 <a name="gvn">The <tt>-gvn</tt>, <tt>-memcpyopt</tt>, and <tt>-dse</tt>
928 <div class="doc_text">
930 <p>These passes use AliasAnalysis information to reason about loads and stores.
935 <!-- ======================================================================= -->
936 <div class="doc_subsection">
937 <a name="aliasanalysis-debug">Clients for debugging and evaluation of
941 <div class="doc_text">
943 <p>These passes are useful for evaluating the various alias analysis
944 implementations. You can use them with commands like '<tt>opt -ds-aa
945 -aa-eval foo.bc -disable-output -stats</tt>'.</p>
949 <!-- _______________________________________________________________________ -->
950 <div class="doc_subsubsection">
951 <a name="print-alias-sets">The <tt>-print-alias-sets</tt> pass</a>
954 <div class="doc_text">
956 <p>The <tt>-print-alias-sets</tt> pass is exposed as part of the
957 <tt>opt</tt> tool to print out the Alias Sets formed by the <a
958 href="#ast"><tt>AliasSetTracker</tt></a> class. This is useful if you're using
959 the <tt>AliasSetTracker</tt> class. To use it, use something like:</p>
961 <div class="doc_code">
963 % opt -ds-aa -print-alias-sets -disable-output
970 <!-- _______________________________________________________________________ -->
971 <div class="doc_subsubsection">
972 <a name="count-aa">The <tt>-count-aa</tt> pass</a>
975 <div class="doc_text">
977 <p>The <tt>-count-aa</tt> pass is useful to see how many queries a particular
978 pass is making and what responses are returned by the alias analysis. As an
981 <div class="doc_code">
983 % opt -basicaa -count-aa -ds-aa -count-aa -licm
987 <p>will print out how many queries (and what responses are returned) by the
988 <tt>-licm</tt> pass (of the <tt>-ds-aa</tt> pass) and how many queries are made
989 of the <tt>-basicaa</tt> pass by the <tt>-ds-aa</tt> pass. This can be useful
990 when debugging a transformation or an alias analysis implementation.</p>
994 <!-- _______________________________________________________________________ -->
995 <div class="doc_subsubsection">
996 <a name="aa-eval">The <tt>-aa-eval</tt> pass</a>
999 <div class="doc_text">
1001 <p>The <tt>-aa-eval</tt> pass simply iterates through all pairs of pointers in a
1002 function and asks an alias analysis whether or not the pointers alias. This
1003 gives an indication of the precision of the alias analysis. Statistics are
1004 printed indicating the percent of no/may/must aliases found (a more precise
1005 algorithm will have a lower number of may aliases).</p>
1009 <!-- *********************************************************************** -->
1010 <div class="doc_section">
1011 <a name="memdep">Memory Dependence Analysis</a>
1013 <!-- *********************************************************************** -->
1015 <div class="doc_text">
1017 <p>If you're just looking to be a client of alias analysis information, consider
1018 using the Memory Dependence Analysis interface instead. MemDep is a lazy,
1019 caching layer on top of alias analysis that is able to answer the question of
1020 what preceding memory operations a given instruction depends on, either at an
1021 intra- or inter-block level. Because of its laziness and caching
1022 policy, using MemDep can be a significant performance win over accessing alias
1023 analysis directly.</p>
1027 <!-- *********************************************************************** -->
1031 <a href="http://jigsaw.w3.org/css-validator/check/referer"><img
1032 src="http://jigsaw.w3.org/css-validator/images/vcss-blue" alt="Valid CSS"></a>
1033 <a href="http://validator.w3.org/check/referer"><img
1034 src="http://www.w3.org/Icons/valid-html401-blue" alt="Valid HTML 4.01"></a>
1036 <a href="mailto:sabre@nondot.org">Chris Lattner</a><br>
1037 <a href="http://llvm.org">LLVM Compiler Infrastructure</a><br>
1038 Last modified: $Date$