| blob | 5b4eb937a52fa8c7a9d94410f95e42d5d855fb9e |
3 <html>
4 <head>
7 </head>
8 <body>
11 LLVM Alias Analysis Infrastructure
12 </div>
14 <ol>
18 <ul>
23 </ul>
24 </li>
27 <ul>
34 </ul>
35 </li>
38 <ul>
42 </ul>
43 </li>
46 <ul>
50 implementations</a></li>
51 </ul>
52 </li>
54 </ol>
58 </div>
60 <!-- *********************************************************************** -->
63 </div>
64 <!-- *********************************************************************** -->
68 <p>Alias Analysis (aka Pointer Analysis) is a class of techniques which attempt
69 to determine whether or not two pointers ever can point to the same object in
70 memory. There are many different algorithms for alias analysis and many
71 different ways of classifying them: flow-sensitive vs flow-insensitive,
72 context-sensitive vs context-insensitive, field-sensitive vs field-insensitive,
73 unification-based vs subset-based, etc. Traditionally, alias analyses respond
75 indicating that two pointers always point to the same object, might point to the
76 same object, or are known to never point to the same object.</p>
80 class is the primary interface used by clients and implementations of alias
81 analyses in the LLVM system. This class is the common interface between clients
82 of alias analysis information and the implementations providing it, and is
83 designed to support a wide range of implementations and clients (but currently
84 all clients are assumed to be flow-insensitive). In addition to simple alias
85 analysis information, this class exposes Mod/Ref information from those
86 implementations which can provide it, allowing for powerful analyses and
87 transformations to work well together.</p>
89 <p>This document contains information necessary to successfully implement this
90 interface, use it, and to test both sides. It also explains some of the finer
91 points about what exactly results mean. If you feel that something is unclear
95 </div>
97 <!-- *********************************************************************** -->
100 </div>
101 <!-- *********************************************************************** -->
107 class defines the interface that the various alias analysis implementations
110 mod/ref query, respectively.</p>
113 represented in several different ways. In particular, memory objects are
114 represented as a starting address and size, and function calls are represented
117 which allow you to get mod/ref information for arbitrary instructions.</p>
119 </div>
121 <!-- ======================================================================= -->
124 </div>
129 which are used to query whether or not two memory objects alias, whether
130 function calls can modify or read a memory object, etc. For all of these
131 queries, memory objects are represented as a pair of their starting address (a
134 <p>Representing memory objects as a starting address and a size is critically
135 important for correct Alias Analyses. For example, consider this (silly, but
136 possible) C code:</p>
139 <pre>
140 int i;
141 char C[2];
142 char A[10];
143 /* ... */
145 C[0] = A[i]; /* One byte store */
147 }
148 </pre>
149 </div>
153 locations one byte apart, and the accesses are each one byte. In this case, the
154 LICM pass can use store motion to remove the stores from the loop. In
155 constrast, the following code:</p>
158 <pre>
159 int i;
160 char C[2];
161 char A[10];
162 /* ... */
164 ((short*)C)[0] = A[i]; /* Two byte store! */
166 }
167 </pre>
168 </div>
170 <p>In this case, the two stores to C do alias each other, because the access to
172 available in the query, even the first case would have to conservatively assume
173 that the accesses alias.</p>
175 </div>
177 <!-- ======================================================================= -->
180 </div>
184 not two memory objects alias each other. It takes two memory objects as input
185 and returns MustAlias, MayAlias, or NoAlias as appropriate.
186 </div>
188 <!-- _______________________________________________________________________ -->
191 </div>
194 <p>The NoAlias response is used when the two pointers refer to distinct objects,
195 regardless of whether the pointers compare equal. For example, freed pointers
196 don't alias any pointers that were allocated afterwards. As a degenerate case,
197 pointers returned by malloc(0) have no bytes for an object, and are considered
198 NoAlias even when malloc returns the same pointer. The same rule applies to
199 NULL pointers.</p>
201 <p>The MayAlias response is used whenever the two pointers might refer to the
202 same object. If the two memory objects overlap, but do not start at the same
203 location, return MayAlias.</p>
205 <p>The MustAlias response may only be returned if the two memory objects are
206 guaranteed to always start at exactly the same location. A MustAlias response
207 implies that the pointers compare equal.</p>
209 </div>
211 <!-- ======================================================================= -->
214 </div>
219 execution of an instruction can read or modify a memory location. Mod/Ref
221 a location, ModRef is returned.</p>
224 method for testing dependencies between function calls. This method takes two
225 call sites (CS1 & CS2), returns NoModRef if the two calls refer to disjoint
226 memory locations, Ref if CS1 reads memory written by CS2, Mod if CS1 writes to
227 memory read or written by CS2, or ModRef if CS1 might read or write memory
228 accessed by CS2. Note that this relation is not commutative.</p>
230 </div>
233 <!-- ======================================================================= -->
236 </div>
240 <p>
241 Several other tidbits of information are often collected by various alias
242 analysis implementations and can be put to good use by various clients.
243 </p>
245 </div>
247 <!-- _______________________________________________________________________ -->
250 </div>
255 analysis can prove that the pointer only points to unchanging memory locations
256 (functions, constant global variables, and the null pointer). This information
257 can be used to refine mod/ref information: it is impossible for an unchanging
258 memory location to be modified.</p>
260 </div>
262 <!-- _______________________________________________________________________ -->
266 </div>
270 <p>These methods are used to provide very simple mod/ref information for
272 function if the analysis can prove that the function never reads or writes to
273 memory, or if the function only reads from constant memory. Functions with this
274 property are side-effect free and only depend on their input arguments, allowing
275 them to be eliminated if they form common subexpressions or be hoisted out of
281 can prove that (at most) the function only reads from non-volatile memory.
282 Functions with this property are side-effect free, only depending on their input
283 arguments and the state of memory when they are called. This property allows
284 calls to these functions to be eliminated and moved around, as long as there is
285 no store instruction that changes the contents of memory. Note that all
289 </div>
291 <!-- *********************************************************************** -->
294 </div>
295 <!-- *********************************************************************** -->
299 <p>Writing a new alias analysis implementation for LLVM is quite
300 straight-forward. There are already several implementations that you can use
301 for examples, and the following information should help fill in any details.
305 </div>
307 <!-- ======================================================================= -->
310 </div>
316 Analysis. As is the case with most other analyses and transformations, the
317 answer should be fairly obvious from what type of problem you are trying to
318 solve:</p>
320 <ol>
321 <li>If you require interprocedural analysis, it should be a
324 <li>If you don't need to look at the program at all, subclass
326 </ol>
328 <p>In addition to the pass that you subclass, you should also inherit from the
333 </div>
335 <!-- ======================================================================= -->
338 </div>
347 declaring any pass dependencies your pass has. Thus you should have something
348 like this:</p>
351 <pre>
352 void getAnalysisUsage(AnalysisUsage &AU) const {
353 AliasAnalysis::getAnalysisUsage(AU);
355 }
356 </pre>
357 </div>
365 <pre>
366 bool run(Module &M) {
367 InitializeAliasAnalysis(this);
369 return false;
370 }
371 </pre>
372 </div>
374 </div>
376 <!-- ======================================================================= -->
379 </div>
386 alias analysis implementation, which ends up returning conservatively correct
388 respectively). Depending on the capabilities of the analysis you are
389 implementing, you just override the interfaces you can improve.</p>
391 </div>
395 <!-- ======================================================================= -->
398 </div>
404 passes) every alias analysis pass chains to another alias analysis
405 implementation (for example, the user can specify "<tt>-basicaa -ds-aa
406 -licm</tt>" to get the maximum benefit from both alias
407 analyses). The alias analysis class automatically takes care of most of this
408 for methods that you don't override. For methods that you do override, in code
409 paths that return a conservative MayAlias or Mod/Ref result, simply return
410 whatever the superclass computes. For example:</p>
413 <pre>
414 AliasAnalysis::AliasResult alias(const Value *V1, unsigned V1Size,
415 const Value *V2, unsigned V2Size) {
416 if (...)
417 return NoAlias;
418 ...
421 return AliasAnalysis::alias(V1, V1Size, V2, V2Size);
422 }
423 </pre>
424 </div>
426 <p>In addition to analysis queries, you must make sure to unconditionally pass
428 well if you override them, which allows all alias analyses in a change to be
429 updated.</p>
431 </div>
434 <!-- ======================================================================= -->
437 </div>
440 <p>
441 Alias analysis information is initially computed for a static snapshot of the
442 program, but clients will use this information to make transformations to the
443 code. All but the most trivial forms of alias analysis will need to have their
444 analysis results updated to reflect the changes made by these transformations.
445 </p>
447 <p>
449 communicate program changes from the clients to the analysis implementations.
450 Various alias analysis implementations should use these methods to ensure that
451 their internal data structures are kept up-to-date as the program changes (for
452 example, when an instruction is deleted), and clients of alias analysis must be
453 sure to call these interfaces appropriately.
454 </p>
455 </div>
457 <!-- _______________________________________________________________________ -->
462 instruction or any other value from the program (including values that do not
463 use pointers). Typically alias analyses keep data structures that have entries
464 for each value in the program. When this method is called, they should remove
465 any entries for the specified value, if they exist.
466 </div>
468 <!-- _______________________________________________________________________ -->
473 program. There is no way to introduce a value into the program that did not
474 exist before (this doesn't make sense for a safe compiler transformation), so
475 this is the only way to introduce a new value. This method indicates that the
476 new value has exactly the same properties as the value being copied.
477 </div>
479 <!-- _______________________________________________________________________ -->
483 This method is a simple helper method that is provided to make clients easier to
484 use. It is implemented by copying the old analysis information to the new
485 value, then deleting the old value. This method cannot be overridden by alias
486 analysis implementations.
487 </div>
489 <!-- ======================================================================= -->
492 </div>
496 <p>From the LLVM perspective, the only thing you need to do to provide an
498 serviced quickly. The actual calculation of the alias analysis results (the
499 "run" method) is only performed once, but many (perhaps duplicate) queries may
500 be performed. Because of this, try to move as much computation to the run
501 method as possible (within reason).</p>
503 </div>
505 <!-- *********************************************************************** -->
508 </div>
509 <!-- *********************************************************************** -->
513 <p>There are several different ways to use alias analysis results. In order of
514 preference, these are...</p>
516 </div>
518 <!-- ======================================================================= -->
521 </div>
526 information about memory-using instructions. This will tell you which store
527 feeds into a load, for example. It uses caching and other techniques to be
528 efficient, and is used by Dead Store Elimination, GVN, and memcpy optimizations.
529 </p>
531 </div>
533 <!-- ======================================================================= -->
536 </div>
541 in some scope, rather than information about pairwise aliasing. The <tt><a
543 is used to efficiently build these Alias Sets from the pairwise alias analysis
547 to add information about various potentially aliasing instructions in the scope
548 you are interested in. Once all of the alias sets are completed, your pass
549 should simply iterate through the constructed alias sets, using the
553 to be disjoint, calculate mod/ref information and volatility for the set, and
554 keep track of whether or not all of the pointers in the set are Must aliases.
555 The AliasSetTracker also makes sure that sets are properly folded due to call
556 instructions, and can provide a list of pointers in each set.</p>
561 then all load instructions from that set may be hoisted out of the loop. If any
563 sunk to outside of the loop, promoting the memory location to a register for the
564 duration of the loop nest. Both of these transformations only apply if the
565 pointer argument is loop-invariant.</p>
567 </div>
569 <!-- _______________________________________________________________________ -->
571 The AliasSetTracker implementation
572 </div>
576 <p>The AliasSetTracker class is implemented to be as efficient as possible. It
577 uses the union-find algorithm to efficiently merge AliasSets when a pointer is
578 inserted into the AliasSetTracker that aliases multiple sets. The primary data
579 structure is a hash table mapping pointers to the AliasSet they are in.</p>
581 <p>The AliasSetTracker class must maintain a list of all of the LLVM Value*'s
582 that are in each AliasSet. Since the hash table already has entries for each
583 LLVM Value* of interest, the AliasesSets thread the linked list through these
584 hash-table nodes to avoid having to allocate memory unnecessarily, and to make
585 merging alias sets extremely efficient (the linked list merge is constant time).
586 </p>
588 <p>You shouldn't need to understand these details if you are just a client of
589 the AliasSetTracker, but if you look at the code, hopefully this brief
590 description will help make sense of why things are designed the way they
591 are.</p>
593 </div>
595 <!-- ======================================================================= -->
598 </div>
602 <p>If neither of these utility class are what your pass needs, you should use
604 the higher-level methods when possible (e.g., use mod/ref information instead of
606 best precision and efficiency.</p>
608 </div>
610 <!-- *********************************************************************** -->
613 </div>
614 <!-- *********************************************************************** -->
618 <p>If you're going to be working with the LLVM alias analysis infrastructure,
619 you should know what clients and implementations of alias analysis are
620 available. In particular, if you are implementing an alias analysis, you should
622 for monitoring and evaluating different implementations.</p>
624 </div>
626 <!-- ======================================================================= -->
629 </div>
638 </div>
640 <!-- _______________________________________________________________________ -->
643 </div>
648 never returns any useful information. This pass can be useful if you think that
649 alias analysis is doing something wrong and are trying to narrow down a
650 problem.</p>
652 </div>
654 <!-- _______________________________________________________________________ -->
657 </div>
664 <ul>
665 <li>Distinct globals, stack allocations, and heap allocations can never
666 alias.</li>
667 <li>Globals, stack allocations, and heap allocations never alias the null
668 pointer.</li>
673 <li>Pointers that obviously point to constant globals
675 <li>Function calls can not modify or references stack allocations if they never
676 escape from the function that allocates them (a common case for automatic
677 arrays).</li>
678 </ul>
680 </div>
682 <!-- _______________________________________________________________________ -->
685 </div>
689 <p>This pass implements a simple context-sensitive mod/ref and alias analysis
690 for internal global variables that don't "have their address taken". If a
691 global does not have its address taken, the pass knows that no pointers alias
692 the global. This pass also keeps track of functions that it knows never access
693 memory or never read memory. This allows certain optimizations (e.g. GVN) to
694 eliminate call instructions entirely.
695 </p>
697 <p>The real power of this pass is that it provides context-sensitive mod/ref
698 information for call instructions. This allows the optimizer to know that
699 calls to a function do not clobber or read the value of the global, allowing
700 loads and stores to be eliminated.</p>
702 <p>Note that this pass is somewhat limited in its scope (only support
703 non-address taken globals), but is very quick analysis.</p>
704 </div>
706 <!-- _______________________________________________________________________ -->
709 </div>
714 "Steensgaard's algorithm" for interprocedural alias analysis. Steensgaard's
715 algorithm is a unification-based, flow-insensitive, context-insensitive, and
716 field-insensitive alias analysis that is also very scalable (effectively linear
717 time).</p>
720 field-<b>sensitive</b>" version of Steensgaard's algorithm using the Data
721 Structure Analysis framework. This gives it substantially more precision than
722 the standard algorithm while maintaining excellent analysis scalability.</p>
725 module, it is not part of the LLVM core.</p>
727 </div>
729 <!-- _______________________________________________________________________ -->
732 </div>
737 algorithm. Data Structure Analysis is a modular unification-based,
740 O(n*log(n)).</p>
742 <p>This algorithm is capable of responding to a full variety of alias analysis
743 queries, and can provide context-sensitive mod/ref information as well. The
744 only major facility not implemented so far is support for must-alias
745 information.</p>
748 module, it is not part of the LLVM core.</p>
750 </div>
753 <!-- ======================================================================= -->
756 </div>
759 LLVM includes several alias-analysis driven transformations which can be used
760 with any of the implementations above.
761 </div>
763 <!-- _______________________________________________________________________ -->
766 </div>
772 not have side-effects and are not used.</p>
774 </div>
777 <!-- _______________________________________________________________________ -->
780 </div>
786 different transformations:</p>
788 <ul>
789 <li>It uses mod/ref information to hoist or sink load instructions out of loops
790 if there are no instructions in the loop that modifies the memory loaded.</li>
792 <li>It uses mod/ref information to hoist function calls out of loops that do not
793 write to memory and are loop-invariant.</li>
795 <li>If uses alias information to promote memory objects that are loaded and
796 stored to in loops to live in a register instead. It can do this if there are
797 no may aliases to the loaded/stored memory location.</li>
798 </ul>
800 </div>
802 <!-- _______________________________________________________________________ -->
805 </div>
808 <p>
810 by-value instead. In particular, if pointer arguments are only loaded from it
811 passes in the value loaded instead of the address to the function. This pass
812 uses alias information to make sure that the value loaded from the argument
813 pointer is not modified between the entry of the function and any load of the
814 pointer.</p>
815 </div>
817 <!-- _______________________________________________________________________ -->
820 passes</a>
821 </div>
825 <p>These passes use AliasAnalysis information to reason about loads and stores.
826 </p>
828 </div>
830 <!-- ======================================================================= -->
833 implementations</a>
834 </div>
838 <p>These passes are useful for evaluating the various alias analysis
839 implementations. You can use them with commands like '<tt>opt -ds-aa
842 </div>
844 <!-- _______________________________________________________________________ -->
847 </div>
857 <pre>
858 % opt -ds-aa -print-alias-sets -disable-output
859 </pre>
860 </div>
862 </div>
865 <!-- _______________________________________________________________________ -->
868 </div>
873 pass is making and what responses are returned by the alias analysis. As an
874 example,</p>
877 <pre>
878 % opt -basicaa -count-aa -ds-aa -count-aa -licm
879 </pre>
880 </div>
882 <p>will print out how many queries (and what responses are returned) by the
885 when debugging a transformation or an alias analysis implementation.</p>
887 </div>
889 <!-- _______________________________________________________________________ -->
892 </div>
897 function and asks an alias analysis whether or not the pointers alias. This
898 gives an indication of the precision of the alias analysis. Statistics are
899 printed indicating the percent of no/may/must aliases found (a more precise
900 algorithm will have a lower number of may aliases).</p>
902 </div>
904 <!-- *********************************************************************** -->
907 </div>
908 <!-- *********************************************************************** -->
912 <p>If you're just looking to be a client of alias analysis information, consider
913 using the Memory Dependence Analysis interface instead. MemDep is a lazy,
914 caching layer on top of alias analysis that is able to answer the question of
915 what preceding memory operations a given instruction depends on, either at an
916 intra- or inter-block level. Because of its laziness and caching
917 policy, using MemDep can be a significant performance win over accessing alias
918 analysis directly.</p>
920 </div>
922 <!-- *********************************************************************** -->
924 <hr>
925 <address>
933 Last modified: $Date$
934 </address>
936 </body>
937 </html>