| blob | a1726aa0c7780dff8d08c8ef2f5c046b9bf8ffb8 |
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>
35 </ul>
36 </li>
39 <ul>
43 </ul>
44 </li>
47 <ul>
51 implementations</a></li>
52 </ul>
53 </li>
55 </ol>
59 </div>
61 <!-- *********************************************************************** -->
64 </div>
65 <!-- *********************************************************************** -->
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
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>
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
96 </div>
98 <!-- *********************************************************************** -->
101 </div>
102 <!-- *********************************************************************** -->
108 class defines the interface that the various alias analysis implementations
111 mod/ref query, respectively.</p>
114 represented in several different ways. In particular, memory objects are
115 represented as a starting address and size, and function calls are represented
118 which allow you to get mod/ref information for arbitrary instructions.</p>
121 multiple values, values which are not
123 same function.</p>
125 </div>
127 <!-- ======================================================================= -->
130 </div>
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
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>
145 <pre>
146 int i;
147 char C[2];
148 char A[10];
149 /* ... */
151 C[0] = A[i]; /* One byte store */
153 }
154 </pre>
155 </div>
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>
164 <pre>
165 int i;
166 char C[2];
167 char A[10];
168 /* ... */
170 ((short*)C)[0] = A[i]; /* Two byte store! */
172 }
173 </pre>
174 </div>
176 <p>In this case, the two stores to C do alias each other, because the access to
178 available in the query, even the first case would have to conservatively assume
179 that the accesses alias.</p>
181 </div>
183 <!-- ======================================================================= -->
186 </div>
190 or not two memory objects alias each other. It takes two memory objects as
191 input and returns MustAlias, MayAlias, or NoAlias as appropriate.</p>
194 that either the two pointer values be defined within the same function, or at
196 </div>
198 <!-- _______________________________________________________________________ -->
201 </div>
204 <p>The NoAlias response may be used when there is never an immediate dependence
207 pointers point to non-overlapping memory ranges. Another is when the two
208 pointers are only ever used for reading memory. Another is when the memory is
209 freed and reallocated between accesses through one pointer and accesses through
210 the other -- in this case, there is a dependence, but it's mediated by the free
211 and reallocation.</p>
213 <p>As an exception to this is with the
215 dependencies are ignored.</p>
217 <p>The MayAlias response is used whenever the two pointers might refer to the
218 same object. If the two memory objects overlap, but do not start at the same
219 location, return MayAlias.</p>
221 <p>The MustAlias response may only be returned if the two memory objects are
222 guaranteed to always start at exactly the same location. A MustAlias response
223 implies that the pointers compare equal.</p>
225 </div>
227 <!-- ======================================================================= -->
230 </div>
235 execution of an instruction can read or modify a memory location. Mod/Ref
237 a location, ModRef is returned.</p>
240 method for testing dependencies between function calls. This method takes two
241 call sites (CS1 & CS2), returns NoModRef if neither call writes to memory
242 read or written by the other, Ref if CS1 reads memory written by CS2, Mod if CS1
243 writes to memory read or written by CS2, or ModRef if CS1 might read or write
244 memory written to by CS2. Note that this relation is not commutative.</p>
246 </div>
249 <!-- ======================================================================= -->
252 </div>
256 <p>
257 Several other tidbits of information are often collected by various alias
258 analysis implementations and can be put to good use by various clients.
259 </p>
261 </div>
263 <!-- _______________________________________________________________________ -->
266 </div>
271 analysis can prove that the pointer only points to unchanging memory locations
272 (functions, constant global variables, and the null pointer). This information
273 can be used to refine mod/ref information: it is impossible for an unchanging
274 memory location to be modified.</p>
276 </div>
278 <!-- _______________________________________________________________________ -->
282 </div>
286 <p>These methods are used to provide very simple mod/ref information for
288 function if the analysis can prove that the function never reads or writes to
289 memory, or if the function only reads from constant memory. Functions with this
290 property are side-effect free and only depend on their input arguments, allowing
291 them to be eliminated if they form common subexpressions or be hoisted out of
297 can prove that (at most) the function only reads from non-volatile memory.
298 Functions with this property are side-effect free, only depending on their input
299 arguments and the state of memory when they are called. This property allows
300 calls to these functions to be eliminated and moved around, as long as there is
301 no store instruction that changes the contents of memory. Note that all
305 </div>
307 <!-- *********************************************************************** -->
310 </div>
311 <!-- *********************************************************************** -->
315 <p>Writing a new alias analysis implementation for LLVM is quite
316 straight-forward. There are already several implementations that you can use
317 for examples, and the following information should help fill in any details.
321 </div>
323 <!-- ======================================================================= -->
326 </div>
332 Analysis. As is the case with most other analyses and transformations, the
333 answer should be fairly obvious from what type of problem you are trying to
334 solve:</p>
336 <ol>
337 <li>If you require interprocedural analysis, it should be a
340 <li>If you don't need to look at the program at all, subclass
342 </ol>
344 <p>In addition to the pass that you subclass, you should also inherit from the
349 </div>
351 <!-- ======================================================================= -->
354 </div>
363 declaring any pass dependencies your pass has. Thus you should have something
364 like this:</p>
367 <pre>
368 void getAnalysisUsage(AnalysisUsage &AU) const {
369 AliasAnalysis::getAnalysisUsage(AU);
371 }
372 </pre>
373 </div>
381 <pre>
382 bool run(Module &M) {
383 InitializeAliasAnalysis(this);
385 return false;
386 }
387 </pre>
388 </div>
390 </div>
392 <!-- ======================================================================= -->
395 </div>
402 alias analysis implementation, which ends up returning conservatively correct
404 respectively). Depending on the capabilities of the analysis you are
405 implementing, you just override the interfaces you can improve.</p>
407 </div>
411 <!-- ======================================================================= -->
414 </div>
420 passes) every alias analysis pass chains to another alias analysis
421 implementation (for example, the user can specify "<tt>-basicaa -ds-aa
422 -licm</tt>" to get the maximum benefit from both alias
423 analyses). The alias analysis class automatically takes care of most of this
424 for methods that you don't override. For methods that you do override, in code
425 paths that return a conservative MayAlias or Mod/Ref result, simply return
426 whatever the superclass computes. For example:</p>
429 <pre>
430 AliasAnalysis::AliasResult alias(const Value *V1, unsigned V1Size,
431 const Value *V2, unsigned V2Size) {
432 if (...)
433 return NoAlias;
434 ...
437 return AliasAnalysis::alias(V1, V1Size, V2, V2Size);
438 }
439 </pre>
440 </div>
442 <p>In addition to analysis queries, you must make sure to unconditionally pass
444 well if you override them, which allows all alias analyses in a change to be
445 updated.</p>
447 </div>
450 <!-- ======================================================================= -->
453 </div>
456 <p>
457 Alias analysis information is initially computed for a static snapshot of the
458 program, but clients will use this information to make transformations to the
459 code. All but the most trivial forms of alias analysis will need to have their
460 analysis results updated to reflect the changes made by these transformations.
461 </p>
463 <p>
465 communicate program changes from the clients to the analysis implementations.
466 Various alias analysis implementations should use these methods to ensure that
467 their internal data structures are kept up-to-date as the program changes (for
468 example, when an instruction is deleted), and clients of alias analysis must be
469 sure to call these interfaces appropriately.
470 </p>
471 </div>
473 <!-- _______________________________________________________________________ -->
478 instruction or any other value from the program (including values that do not
479 use pointers). Typically alias analyses keep data structures that have entries
480 for each value in the program. When this method is called, they should remove
481 any entries for the specified value, if they exist.
482 </div>
484 <!-- _______________________________________________________________________ -->
489 program. There is no way to introduce a value into the program that did not
490 exist before (this doesn't make sense for a safe compiler transformation), so
491 this is the only way to introduce a new value. This method indicates that the
492 new value has exactly the same properties as the value being copied.
493 </div>
495 <!-- _______________________________________________________________________ -->
499 This method is a simple helper method that is provided to make clients easier to
500 use. It is implemented by copying the old analysis information to the new
501 value, then deleting the old value. This method cannot be overridden by alias
502 analysis implementations.
503 </div>
505 <!-- ======================================================================= -->
508 </div>
512 <p>From the LLVM perspective, the only thing you need to do to provide an
514 serviced quickly. The actual calculation of the alias analysis results (the
515 "run" method) is only performed once, but many (perhaps duplicate) queries may
516 be performed. Because of this, try to move as much computation to the run
517 method as possible (within reason).</p>
519 </div>
521 <!-- ======================================================================= -->
524 </div>
528 <p>PassManager support for alternative AliasAnalysis implementation
529 has some issues.</p>
531 <p>There is no way to override the default alias analysis. It would
532 be very useful to be able to do something like "opt -my-aa -O2" and
533 have it use -my-aa for all passes which need AliasAnalysis, but there
534 is currently no support for that, short of changing the source code
535 and recompiling. Similarly, there is also no way of setting a chain
536 of analyses as the default.</p>
538 <p>There is no way for transform passes to declare that they preserve
541 which are intended to allow a pass to keep an AliasAnalysis consistent,
542 however there's no way for a pass to declare in its
545 actually have any effect.</tt>
550 these utilities. If you try to use them, the pass manager will
551 silently route alias analysis queries directly to
556 alias analysis passes.</p>
558 </div>
560 <!-- *********************************************************************** -->
563 </div>
564 <!-- *********************************************************************** -->
568 <p>There are several different ways to use alias analysis results. In order of
569 preference, these are...</p>
571 </div>
573 <!-- ======================================================================= -->
576 </div>
581 information about memory-using instructions. This will tell you which store
582 feeds into a load, for example. It uses caching and other techniques to be
583 efficient, and is used by Dead Store Elimination, GVN, and memcpy optimizations.
584 </p>
586 </div>
588 <!-- ======================================================================= -->
591 </div>
596 in some scope, rather than information about pairwise aliasing. The <tt><a
598 is used to efficiently build these Alias Sets from the pairwise alias analysis
602 to add information about various potentially aliasing instructions in the scope
603 you are interested in. Once all of the alias sets are completed, your pass
604 should simply iterate through the constructed alias sets, using the
608 to be disjoint, calculate mod/ref information and volatility for the set, and
609 keep track of whether or not all of the pointers in the set are Must aliases.
610 The AliasSetTracker also makes sure that sets are properly folded due to call
611 instructions, and can provide a list of pointers in each set.</p>
616 then all load instructions from that set may be hoisted out of the loop. If any
618 sunk to outside of the loop, promoting the memory location to a register for the
619 duration of the loop nest. Both of these transformations only apply if the
620 pointer argument is loop-invariant.</p>
622 </div>
624 <!-- _______________________________________________________________________ -->
626 The AliasSetTracker implementation
627 </div>
631 <p>The AliasSetTracker class is implemented to be as efficient as possible. It
632 uses the union-find algorithm to efficiently merge AliasSets when a pointer is
633 inserted into the AliasSetTracker that aliases multiple sets. The primary data
634 structure is a hash table mapping pointers to the AliasSet they are in.</p>
636 <p>The AliasSetTracker class must maintain a list of all of the LLVM Value*'s
637 that are in each AliasSet. Since the hash table already has entries for each
638 LLVM Value* of interest, the AliasesSets thread the linked list through these
639 hash-table nodes to avoid having to allocate memory unnecessarily, and to make
640 merging alias sets extremely efficient (the linked list merge is constant time).
641 </p>
643 <p>You shouldn't need to understand these details if you are just a client of
644 the AliasSetTracker, but if you look at the code, hopefully this brief
645 description will help make sense of why things are designed the way they
646 are.</p>
648 </div>
650 <!-- ======================================================================= -->
653 </div>
657 <p>If neither of these utility class are what your pass needs, you should use
659 the higher-level methods when possible (e.g., use mod/ref information instead of
661 best precision and efficiency.</p>
663 </div>
665 <!-- *********************************************************************** -->
668 </div>
669 <!-- *********************************************************************** -->
673 <p>If you're going to be working with the LLVM alias analysis infrastructure,
674 you should know what clients and implementations of alias analysis are
675 available. In particular, if you are implementing an alias analysis, you should
677 for monitoring and evaluating different implementations.</p>
679 </div>
681 <!-- ======================================================================= -->
684 </div>
693 </div>
695 <!-- _______________________________________________________________________ -->
698 </div>
703 never returns any useful information. This pass can be useful if you think that
704 alias analysis is doing something wrong and are trying to narrow down a
705 problem.</p>
707 </div>
709 <!-- _______________________________________________________________________ -->
712 </div>
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>
728 <li>Pointers that obviously point to constant globals
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>
735 </div>
737 <!-- _______________________________________________________________________ -->
740 </div>
744 <p>This pass implements a simple context-sensitive mod/ref and alias analysis
745 for internal global variables that don't "have their address taken". If a
746 global does not have its address taken, the pass knows that no pointers alias
747 the global. This pass also keeps track of functions that it knows never access
748 memory or never read memory. This allows certain optimizations (e.g. GVN) to
749 eliminate call instructions entirely.
750 </p>
752 <p>The real power of this pass is that it provides context-sensitive mod/ref
753 information for call instructions. This allows the optimizer to know that
754 calls to a function do not clobber or read the value of the global, allowing
755 loads and stores to be eliminated.</p>
757 <p>Note that this pass is somewhat limited in its scope (only support
758 non-address taken globals), but is very quick analysis.</p>
759 </div>
761 <!-- _______________________________________________________________________ -->
764 </div>
769 "Steensgaard's algorithm" for interprocedural alias analysis. Steensgaard's
770 algorithm is a unification-based, flow-insensitive, context-insensitive, and
771 field-insensitive alias analysis that is also very scalable (effectively linear
772 time).</p>
775 field-<b>sensitive</b>" version of Steensgaard's algorithm using the Data
776 Structure Analysis framework. This gives it substantially more precision than
777 the standard algorithm while maintaining excellent analysis scalability.</p>
780 module, it is not part of the LLVM core.</p>
782 </div>
784 <!-- _______________________________________________________________________ -->
787 </div>
792 algorithm. Data Structure Analysis is a modular unification-based,
795 O(n*log(n)).</p>
797 <p>This algorithm is capable of responding to a full variety of alias analysis
798 queries, and can provide context-sensitive mod/ref information as well. The
799 only major facility not implemented so far is support for must-alias
800 information.</p>
803 module, it is not part of the LLVM core.</p>
805 </div>
807 <!-- _______________________________________________________________________ -->
810 </div>
815 translating them into ScalarEvolution queries. This gives it a
817 and loop induction variables than other alias analyses have.</p>
819 </div>
821 <!-- ======================================================================= -->
824 </div>
827 LLVM includes several alias-analysis driven transformations which can be used
828 with any of the implementations above.
829 </div>
831 <!-- _______________________________________________________________________ -->
834 </div>
840 not have side-effects and are not used.</p>
842 </div>
845 <!-- _______________________________________________________________________ -->
848 </div>
854 different transformations:</p>
856 <ul>
857 <li>It uses mod/ref information to hoist or sink load instructions out of loops
858 if there are no instructions in the loop that modifies the memory loaded.</li>
860 <li>It uses mod/ref information to hoist function calls out of loops that do not
861 write to memory and are loop-invariant.</li>
863 <li>If uses alias information to promote memory objects that are loaded and
864 stored to in loops to live in a register instead. It can do this if there are
865 no may aliases to the loaded/stored memory location.</li>
866 </ul>
868 </div>
870 <!-- _______________________________________________________________________ -->
873 </div>
876 <p>
878 by-value instead. In particular, if pointer arguments are only loaded from it
879 passes in the value loaded instead of the address to the function. This pass
880 uses alias information to make sure that the value loaded from the argument
881 pointer is not modified between the entry of the function and any load of the
882 pointer.</p>
883 </div>
885 <!-- _______________________________________________________________________ -->
888 passes</a>
889 </div>
893 <p>These passes use AliasAnalysis information to reason about loads and stores.
894 </p>
896 </div>
898 <!-- ======================================================================= -->
901 implementations</a>
902 </div>
906 <p>These passes are useful for evaluating the various alias analysis
907 implementations. You can use them with commands like '<tt>opt -ds-aa
910 </div>
912 <!-- _______________________________________________________________________ -->
915 </div>
925 <pre>
926 % opt -ds-aa -print-alias-sets -disable-output
927 </pre>
928 </div>
930 </div>
933 <!-- _______________________________________________________________________ -->
936 </div>
941 pass is making and what responses are returned by the alias analysis. As an
942 example,</p>
945 <pre>
946 % opt -basicaa -count-aa -ds-aa -count-aa -licm
947 </pre>
948 </div>
950 <p>will print out how many queries (and what responses are returned) by the
953 when debugging a transformation or an alias analysis implementation.</p>
955 </div>
957 <!-- _______________________________________________________________________ -->
960 </div>
965 function and asks an alias analysis whether or not the pointers alias. This
966 gives an indication of the precision of the alias analysis. Statistics are
967 printed indicating the percent of no/may/must aliases found (a more precise
968 algorithm will have a lower number of may aliases).</p>
970 </div>
972 <!-- *********************************************************************** -->
975 </div>
976 <!-- *********************************************************************** -->
980 <p>If you're just looking to be a client of alias analysis information, consider
981 using the Memory Dependence Analysis interface instead. MemDep is a lazy,
982 caching layer on top of alias analysis that is able to answer the question of
983 what preceding memory operations a given instruction depends on, either at an
984 intra- or inter-block level. Because of its laziness and caching
985 policy, using MemDep can be a significant performance win over accessing alias
986 analysis directly.</p>
988 </div>
990 <!-- *********************************************************************** -->
992 <hr>
993 <address>
1001 Last modified: $Date$
1002 </address>
1004 </body>
1005 </html>