| blob | 347861e9638896a59a85879ee228860fae29fc25 |
3 <html>
4 <head>
7 </head>
8 <body>
11 LLVM Programmer's Manual
12 </div>
14 <ol>
17 <ul>
19 <!--
20 <li>The <tt>-time-passes</tt> option</li>
21 <li>How to use the LLVM Makefile system</li>
22 <li>How to write a regression test</li>
24 -->
25 </ul>
26 </li>
28 <ul>
32 option</a>
33 <ul>
36 </ul>
37 </li>
39 option</a></li>
40 <!--
41 <li>The <tt>InstVisitor</tt> template
42 <li>The general graph API
43 -->
45 </ul>
46 </li>
48 <ul>
50 <ul>
59 </ul></li>
61 <ul>
70 </ul></li>
72 <ul>
79 </ul></li>
80 </ul>
81 </li>
83 <ul>
85 <ul>
93 class pointer</a> </li>
95 complex example</a> </li>
97 the same way</a> </li>
99 use-def chains</a> </li>
100 </ul>
101 </li>
103 <ul>
109 </ul>
110 </li>
111 <!--
112 <li>Working with the Control Flow Graph
113 <ul>
114 <li>Accessing predecessors and successors of a <tt>BasicBlock</tt>
115 <li>
116 <li>
117 </ul>
118 -->
119 </ul>
120 </li>
123 <ul>
125 <ul>
130 </ul></li>
132 <li><a href="#SymbolTable">The <tt>ValueSymbolTable</tt> and <tt>TypeSymbolTable</tt> classes </a></li>
133 </ul></li>
136 <ul>
140 <ul>
142 <ul>
145 <ul>
147 <ul>
150 </ul>
151 </li>
152 </ul>
153 </li>
154 </ul>
155 </li>
158 </ul>
159 </li>
160 </ul>
161 </li>
162 </ol>
169 </div>
171 <!-- *********************************************************************** -->
174 </div>
175 <!-- *********************************************************************** -->
179 <p>This document is meant to highlight some of the important classes and
180 interfaces available in the LLVM source-base. This manual is not
181 intended to explain what LLVM is, how it works, and what LLVM code looks
182 like. It assumes that you know the basics of LLVM and are interested
183 in writing transformations or otherwise analyzing or manipulating the
184 code.</p>
186 <p>This document should get you oriented so that you can find your
187 way in the continuously growing source code that makes up the LLVM
188 infrastructure. Note that this manual is not intended to serve as a
189 replacement for reading the source code, so if you think there should be
190 a method in one of these classes to do something, but it's not listed,
192 are provided to make this as easy as possible.</p>
194 <p>The first section of this document describes general information that is
195 useful to know when working in the LLVM infrastructure, and the second describes
196 the Core LLVM classes. In the future this manual will be extended with
197 information describing how to use extension libraries, such as dominator
198 information, CFG traversal routines, and useful utilities like the <tt><a
201 </div>
203 <!-- *********************************************************************** -->
206 </div>
207 <!-- *********************************************************************** -->
211 <p>This section contains general information that is useful if you are working
212 in the LLVM source-base, but that isn't specific to any particular API.</p>
214 </div>
216 <!-- ======================================================================= -->
219 </div>
223 <p>LLVM makes heavy use of the C++ Standard Template Library (STL),
224 perhaps much more than you are used to, or have seen before. Because of
225 this, you might want to do a little background reading in the
226 techniques used and capabilities of the library. There are many good
227 pages that discuss the STL, and several books on the subject that you
228 can get, so it will not be discussed in this document.</p>
232 <ol>
235 reference</a> - an excellent reference for the STL and other parts of the
236 standard C++ library.</li>
239 O'Reilly book in the making. It has a decent
240 Standard Library
241 Reference that rivals Dinkumware's, and is unfortunately no longer free since the book has been
242 published.</li>
245 Questions</a></li>
248 Contains a useful <a
253 Page</a></li>
257 the book).</a></li>
259 </ol>
263 to write maintainable code more than where to put your curly braces.</p>
265 </div>
267 <!-- ======================================================================= -->
270 </div>
274 <ol>
276 Branch and Tag Primer</a></li>
278 static and shared libraries across platforms</a></li>
279 </ol>
281 </div>
283 <!-- *********************************************************************** -->
286 </div>
287 <!-- *********************************************************************** -->
291 <p>Here we highlight some LLVM APIs that are generally useful and good to
292 know about when writing transformations.</p>
294 </div>
296 <!-- ======================================================================= -->
300 </div>
304 <p>The LLVM source-base makes extensive use of a custom form of RTTI.
306 operator, but they don't have some drawbacks (primarily stemming from
308 have a v-table). Because they are used so often, you must know what they
309 do and how they work. All of these templates are defined in the <a
311 file (note that you very rarely have to include this file directly).</p>
313 <dl>
317 "<tt>instanceof</tt>" operator. It returns true or false depending on whether
318 a reference or pointer points to an instance of the specified class. This can
319 be very useful for constraint checking of various sorts (example below).</p>
320 </dd>
325 converts a pointer or reference from a base class to a derived cast, causing
326 an assertion failure if it is not really an instance of the right type. This
327 should be used in cases where you have some information that makes you believe
332 <pre>
334 if (isa<<a href="#Constant">Constant</a>>(V) || isa<<a href="#Argument">Argument</a>>(V) || isa<<a href="#GlobalValue">GlobalValue</a>>(V))
335 return true;
339 }
340 </pre>
341 </div>
345 operator.</p>
347 </dd>
352 It checks to see if the operand is of the specified type, and if so, returns a
353 pointer to it (this operator does not work with references). If the operand is
354 not of the correct type, a null pointer is returned. Thus, this works very
358 statement like this:</p>
361 <pre>
362 if (<a href="#AllocationInst">AllocationInst</a> *AI = dyn_cast<<a href="#AllocationInst">AllocationInst</a>>(Val)) {
364 }
365 </pre>
366 </div>
370 statement, which is very convenient.</p>
375 blocks to check for lots of different variants of classes. If you find
376 yourself wanting to do this, it is much cleaner and more efficient to use the
379 </dd>
385 argument (which it then propagates). This can sometimes be useful, allowing
386 you to combine several null checks into one.</p></dd>
392 as an argument (which it then propagates). This can sometimes be useful,
393 allowing you to combine several null checks into one.</p></dd>
395 </dl>
397 <p>These five templates can be used with any classes, whether they have a
398 v-table or not. To add support for these templates, you simply need to add
400 to. Describing this is currently outside the scope of this document, but there
401 are lots of examples in the LLVM source base.</p>
403 </div>
405 <!-- ======================================================================= -->
408 </div>
412 <p>Often when working on your pass you will put a bunch of debugging printouts
413 and other code into your pass. After you get it working, you want to remove
414 it, but you may need it again in the future (to work out new bugs that you run
415 across).</p>
417 <p> Naturally, because of this, you don't want to delete the debug printouts,
418 but you don't want them to always be noisy. A standard compromise is to comment
419 them out, allowing you to enable them if you need them in the future.</p>
423 this problem. Basically, you can put arbitrary code into the argument of the
428 <pre>
430 </pre>
431 </div>
436 <pre>
440 I am here!
441 </pre>
442 </div>
445 to not have to create "yet another" command line option for the debug output for
447 so they do not cause a performance impact at all (for the same reason, they
448 should also not contain side-effects!).</p>
451 enable or disable it directly in gdb. Just use "<tt>set DebugFlag=0</tt>" or
452 "<tt>set DebugFlag=1</tt>" from the gdb if the program is running. If the
453 program hasn't been started yet, you can always just run it with
456 </div>
458 <!-- _______________________________________________________________________ -->
462 </div>
468 generator). If you want to enable debug information with more fine-grained
470 option as follows:</p>
473 <pre>
475 #undef DEBUG_TYPE
476 #define DEBUG_TYPE "foo"
478 #undef DEBUG_TYPE
479 #define DEBUG_TYPE "bar"
481 #undef DEBUG_TYPE
482 #define DEBUG_TYPE ""
484 </pre>
485 </div>
490 <pre>
494 No debug type
495 'foo' debug type
496 'bar' debug type
497 No debug type (2)
499 'foo' debug type
501 'bar' debug type
502 </pre>
503 </div>
506 a file, to specify the debug type for the entire module (if you do this before
509 "bar", because there is no system in place to ensure that names do not
510 conflict. If two different modules use the same string, they will all be turned
511 on when the name is specified. This allows, for example, all debug information
513 even if the source lives in multiple files.</p>
515 </div>
517 <!-- ======================================================================= -->
520 option</a>
521 </div>
528 keep track of what the LLVM compiler is doing and how effective various
529 optimizations are. It is useful to see what optimizations are contributing to
530 making a particular program run faster.</p>
532 <p>Often you may run your pass on some big program, and you're interested to see
533 how many times it makes a certain transformation. Although you can do this with
534 hand inspection, or some ad-hoc method, this is a real pain and not very useful
536 keep track of this information, and the calculated information is presented in a
537 uniform manner with the rest of the passes being executed.</p>
540 it are as follows:</p>
542 <ol>
546 <pre>
547 #define <a href="#DEBUG_TYPE">DEBUG_TYPE</a> "mypassname" <i>// This goes before any #includes.</i>
548 STATISTIC(NumXForms, "The # of times I did stuff");
549 </pre>
550 </div>
553 specified by the first argument. The pass name is taken from the DEBUG_TYPE
554 macro, and the description is taken from the second argument. The variable
560 <pre>
562 </pre>
563 </div>
565 </li>
566 </ol>
572 <pre>
575 </pre>
576 </div>
579 suite, it gives a report that looks like this:</p>
582 <pre>
583 7646 bytecodewriter - Number of normal instructions
584 725 bytecodewriter - Number of oversized instructions
585 129996 bytecodewriter - Number of bytecode bytes written
586 2817 raise - Number of insts DCEd or constprop'd
587 3213 raise - Number of cast-of-self removed
588 5046 raise - Number of expression trees converted
589 75 raise - Number of other getelementptr's formed
590 138 raise - Number of load/store peepholes
591 42 deadtypeelim - Number of unused typenames removed from symtab
592 392 funcresolve - Number of varargs functions resolved
593 27 globaldce - Number of global variables removed
594 2 adce - Number of basic blocks removed
595 134 cee - Number of branches revectored
596 49 cee - Number of setcc instruction eliminated
597 532 gcse - Number of loads removed
598 2919 gcse - Number of instructions removed
599 86 indvars - Number of canonical indvars added
600 87 indvars - Number of aux indvars removed
601 25 instcombine - Number of dead inst eliminate
602 434 instcombine - Number of insts combined
603 248 licm - Number of load insts hoisted
604 1298 licm - Number of insts hoisted to a loop pre-header
605 3 licm - Number of insts hoisted to multiple loop preds (bad, no loop pre-header)
606 75 mem2reg - Number of alloca's promoted
607 1444 cfgsimplify - Number of blocks simplified
608 </pre>
609 </div>
611 <p>Obviously, with so many optimizations, having a unified framework for this
612 stuff is very nice. Making your pass fit well into the framework makes it more
613 maintainable and useful.</p>
615 </div>
617 <!-- ======================================================================= -->
620 </div>
624 <p>Several of the important data structures in LLVM are graphs: for example
628 DAGs</a>. In many cases, while debugging various parts of the compiler, it is
629 nice to instantly visualize these graphs.</p>
631 <p>LLVM provides several callbacks that are available in a debug build to do
633 the current LLVM tool will pop up a window containing the CFG for the function
634 where each basic block is a node in the graph, and each node contains the
635 instructions in the block. Similarly, there also exists
640 up a window. Alternatively, you can sprinkle calls to these functions in your
641 code in places you want to debug.</p>
643 <p>Getting this to work requires a small amount of configuration. On Unix
645 toolkit, and make sure 'dot' and 'gv' are in your path. If you are running on
646 Mac OS/X, download and install the Mac OS/X <a
649 it) to your path. Once in your system and path are set up, rerun the LLVM
650 configure script and rebuild LLVM to enable this functionality.</p>
656 specified color (choices of colors can be found at <a
658 complex node attributes can be provided with <tt>call
661 Attributes</a>.) If you want to restart and clear all the current graph
664 </div>
666 <!-- *********************************************************************** -->
669 </div>
670 <!-- *********************************************************************** -->
675 and we commonly use STL data structures. This section describes the trade-offs
676 you should consider when you pick one.</p>
678 <p>
679 The first step is a choose your own adventure: do you want a sequential
680 container, a set-like container, or a map-like container? The most important
681 thing when choosing a container is the algorithmic properties of how you plan to
682 access the container. Based on that, you should use:</p>
684 <ul>
686 of an value based on another value. Map-like containers also support
687 efficient queries for containment (whether a key is in the map). Map-like
688 containers generally do not support efficient reverse mapping (values to
689 keys). If you need that, use two maps. Some map-like containers also
690 support efficient iteration through the keys in sorted order. Map-like
691 containers are the most expensive sort, only use them if you need one of
692 these capabilities.</li>
695 stuff into a container that automatically eliminates duplicates. Some
696 set-like containers support efficient iteration through the elements in
697 sorted order. Set-like containers are more expensive than sequential
698 containers.
699 </li>
702 the most efficient way to add elements and keeps track of the order they are
703 added to the collection. They permit duplicates and support efficient
704 iteration, but do not support efficient look-up based on a key.
705 </li>
707 </ul>
709 <p>
710 Once the proper category of container is determined, you can fine tune the
711 memory use, constant factors, and cache behaviors of access by intelligently
712 picking a member of the category. Note that constant factors and cache behavior
713 can be a big deal. If you have a vector that usually only contains a few
714 elements (but could contain many), for example, it's much better to use
716 . Doing so avoids (relatively) expensive malloc/free calls, which dwarf the
717 cost of adding the elements to the container. </p>
719 </div>
721 <!-- ======================================================================= -->
724 </div>
727 There are a variety of sequential containers available for you, based on your
728 needs. Pick the first in this section that will do what you want.
729 </div>
731 <!-- _______________________________________________________________________ -->
734 </div>
737 <p>Fixed size arrays are very simple and very fast. They are good if you know
738 exactly how many elements you have, or you have a (low) upper bound on how many
739 you have.</p>
740 </div>
742 <!-- _______________________________________________________________________ -->
745 </div>
748 <p>Heap allocated arrays (new[] + delete[]) are also simple. They are good if
749 the number of elements is variable, if you know how many elements you will need
750 before the array is allocated, and if the array is usually large (if not,
752 allocated array is the cost of the new/delete (aka malloc/free). Also note that
753 if you are allocating an array of a type with a constructor, the constructor and
754 destructors will be run for every element in the array (re-sizable vectors only
755 construct those elements actually used).</p>
756 </div>
758 <!-- _______________________________________________________________________ -->
761 </div>
766 it supports efficient iteration, lays out elements in memory order (so you can
767 do pointer arithmetic between elements), supports efficient push_back/pop_back
768 operations, supports efficient random access to its elements, etc.</p>
770 <p>The advantage of SmallVector is that it allocates space for
772 the SmallVector is dynamically smaller than N, no malloc is performed. This can
773 be a big win in cases where the malloc/free call is far more expensive than the
774 code that fiddles around with the elements.</p>
777 predecessors/successors of a block is usually less than 8). On the other hand,
778 this makes the size of the SmallVector itself large, so you don't want to
779 allocate lots of them (doing so will waste a lot of space). As such,
780 SmallVectors are most useful when on the stack.</p>
782 <p>SmallVector also provides a nice portable and efficient replacement for
785 </div>
787 <!-- _______________________________________________________________________ -->
790 </div>
793 <p>
794 std::vector is well loved and respected. It is useful when SmallVector isn't:
795 when the size of the vector is often large (thus the small optimization will
796 rarely be a benefit) or if you will be allocating many instances of the vector
797 itself (which would waste space for elements that aren't in the container).
798 vector is also useful when interfacing with code that expects vectors :).
799 </p>
804 <pre>
805 for ( ... ) {
807 use V;
808 }
809 </pre>
810 </div>
815 <pre>
817 for ( ... ) {
818 use V;
819 V.clear();
820 }
821 </pre>
822 </div>
824 <p>Doing so will save (at least) one heap allocation and free per iteration of
825 the loop.</p>
827 </div>
829 <!-- _______________________________________________________________________ -->
832 </div>
835 <p>std::deque is, in some senses, a generalized version of std::vector. Like
836 std::vector, it provides constant time random access and other similar
837 properties, but it also provides efficient access to the front of the list. It
838 does not guarantee continuity of elements within memory.</p>
840 <p>In exchange for this extra flexibility, std::deque has significantly higher
841 constant factor costs than std::vector. If possible, use std::vector or
842 something cheaper.</p>
843 </div>
845 <!-- _______________________________________________________________________ -->
848 </div>
851 <p>std::list is an extremely inefficient class that is rarely useful.
852 It performs a heap allocation for every element inserted into it, thus having an
853 extremely high constant factor, particularly for small data types. std::list
854 also only supports bidirectional iteration, not random access iteration.</p>
856 <p>In exchange for this high cost, std::list supports efficient access to both
857 ends of the list (like std::deque, but unlike std::vector or SmallVector). In
858 addition, the iterator invalidation characteristics of std::list are stronger
859 than that of a vector class: inserting or removing an element into the list does
860 not invalidate iterator or pointers to other elements in the list.</p>
861 </div>
863 <!-- _______________________________________________________________________ -->
866 </div>
870 intrusive, because it requires the element to store and provide access to the
871 prev/next pointers for the list.</p>
873 <p>ilist has the same drawbacks as std::list, and additionally requires an
874 ilist_traits implementation for the element type, but it provides some novel
875 characteristics. In particular, it can efficiently store polymorphic objects,
876 the traits class is informed when an element is inserted or removed from the
877 list, and ilists are guaranteed to support a constant-time splice operation.
878 </p>
880 <p>These properties are exactly what we want for things like Instructions and
881 basic blocks, which is why these are implemented with ilists.</p>
882 </div>
884 <!-- _______________________________________________________________________ -->
887 </div>
892 <p>There are also various STL adapter classes such as std::queue,
893 std::priority_queue, std::stack, etc. These provide simplified access to an
894 underlying container but don't affect the cost of the container itself.</p>
896 </div>
899 <!-- ======================================================================= -->
902 </div>
906 <p>Set-like containers are useful when you need to canonicalize multiple values
907 into a single representation. There are several different choices for how to do
908 this, providing various trade-offs.</p>
910 </div>
913 <!-- _______________________________________________________________________ -->
916 </div>
920 <p>If you intend to insert a lot of elements, then do a lot of queries, a
921 great approach is to use a vector (or other sequential container) with
922 std::sort+std::unique to remove duplicates. This approach works really well if
923 your usage pattern has these two distinct phases (insert then query), and can be
925 </p>
927 <p>
928 This combination provides the several nice properties: the result data is
929 contiguous in memory (good for cache locality), has few allocations, is easy to
930 address (iterators in the final vector are just indices or pointers), and can be
931 efficiently queried with a standard binary or radix search.</p>
933 </div>
935 <!-- _______________________________________________________________________ -->
938 </div>
942 <p>If you have a set-like data structure that is usually small and whose elements
944 has space for N elements in place (thus, if the set is dynamically smaller than
945 N, no malloc traffic is required) and accesses them with a simple linear search.
946 When the set grows beyond 'N' elements, it allocates a more expensive representation that
947 guarantees efficient access (for most types, it falls back to std::set, but for
948 pointers it uses something far better, <a
951 <p>The magic of this class is that it handles small sets extremely efficiently,
952 but gracefully handles extremely large sets without loss of efficiency. The
953 drawback is that the interface is quite small: it supports insertion, queries
954 and erasing, but does not support iteration.</p>
956 </div>
958 <!-- _______________________________________________________________________ -->
961 </div>
965 <p>SmallPtrSet has all the advantages of SmallSet (and a SmallSet of pointers is
966 transparently implemented with a SmallPtrSet), but also supports iterators. If
967 more than 'N' insertions are performed, a single quadratically
968 probed hash table is allocated and grows as needed, providing extremely
969 efficient access (constant time insertion/deleting/queries with low constant
970 factors) and is very stingy with malloc traffic.</p>
972 <p>Note that, unlike std::set, the iterators of SmallPtrSet are invalidated
973 whenever an insertion occurs. Also, the values visited by the iterators are not
974 visited in sorted order.</p>
976 </div>
978 <!-- _______________________________________________________________________ -->
981 </div>
985 <p>
986 FoldingSet is an aggregate class that is really good at uniquing
987 expensive-to-create or polymorphic objects. It is a combination of a chained
988 hash table with intrusive links (uniqued objects are required to inherit from
990 its ID process.</p>
993 a complex object (for example, a node in the code generator). The client has a
994 description of *what* it wants to generate (it knows the opcode and all the
995 operands), but we don't want to 'new' a node, then try inserting it into a set
996 only to find out it already exists, at which point we would have to delete it
997 and return the node that already exists.
998 </p>
1000 <p>To support this style of client, FoldingSet perform a query with a
1001 FoldingSetNodeID (which wraps SmallVector) that can be used to describe the
1002 element that we want to query for. The query either returns the element
1003 matching the ID or it returns an opaque ID that indicates where insertion should
1004 take place. Construction of the ID usually does not require heap traffic.</p>
1006 <p>Because FoldingSet uses intrusive links, it can support polymorphic objects
1007 in the set (for example, you can have SDNode instances mixed with LoadSDNodes).
1008 Because the elements are individually allocated, pointers to the elements are
1009 stable: inserting or removing elements does not invalidate any pointers to other
1010 elements.
1011 </p>
1013 </div>
1015 <!-- _______________________________________________________________________ -->
1018 </div>
1023 many things but great at nothing. std::set allocates memory for each element
1024 inserted (thus it is very malloc intensive) and typically stores three pointers
1025 per element in the set (thus adding a large amount of per-element space
1026 overhead). It offers guaranteed log(n) performance, which is not particularly
1027 fast from a complexity standpoint (particularly if the elements of the set are
1028 expensive to compare, like strings), and has extremely high constant factors for
1029 lookup, insertion and removal.</p>
1031 <p>The advantages of std::set are that its iterators are stable (deleting or
1032 inserting an element from the set does not affect iterators or pointers to other
1033 elements) and that iteration over the set is guaranteed to be in sorted order.
1034 If the elements in the set are large, then the relative overhead of the pointers
1035 and malloc traffic is not a big deal, but if the elements of the set are small,
1036 std::set is almost never a good choice.</p>
1038 </div>
1040 <!-- _______________________________________________________________________ -->
1043 </div>
1048 Container</a>. The important property
1049 that this provides is efficient insertion with uniquing (duplicate elements are
1050 ignored) with iteration support. It implements this by inserting elements into
1051 both a set-like container and the sequential container, using the set-like
1052 container for uniquing and the sequential container for iteration.
1053 </p>
1055 <p>The difference between SetVector and other sets is that the order of
1056 iteration is guaranteed to match the order of insertion into the SetVector.
1057 This property is really important for things like sets of pointers. Because
1058 pointer values are non-deterministic (e.g. vary across runs of the program on
1059 different machines), iterating over the pointers in the set will
1060 not be in a well-defined order.</p>
1062 <p>
1063 The drawback of SetVector is that it requires twice as much space as a normal
1064 set and has the sum of constant factors from the set-like container and the
1065 sequential container that it uses. Use it *only* if you need to iterate over
1066 the elements in a deterministic order. SetVector is also expensive to delete
1067 elements out of (linear time), unless you use it's "pop_back" method, which is
1068 faster.
1069 </p>
1071 <p>SetVector is an adapter class that defaults to using std::vector and std::set
1072 for the underlying containers, so it is quite expensive. However,
1074 defaults to using a SmallVector and SmallSet of a specified size. If you use
1075 this, and if your sets are dynamically smaller than N, you will save a lot of
1076 heap traffic.</p>
1078 </div>
1080 <!-- _______________________________________________________________________ -->
1083 </div>
1087 <p>
1089 retains a unique ID for each element inserted into the set. It internally
1090 contains a map and a vector, and it assigns a unique ID for each value inserted
1091 into the set.</p>
1093 <p>UniqueVector is very expensive: its cost is the sum of the cost of
1094 maintaining both the map and vector, it has high complexity, high constant
1095 factors, and produces a lot of malloc traffic. It should be avoided.</p>
1097 </div>
1100 <!-- _______________________________________________________________________ -->
1103 </div>
1107 <p>
1108 The STL provides several other options, such as std::multiset and the various
1111 <p>std::multiset is useful if you're not interested in elimination of
1112 duplicates, but has all the drawbacks of std::set. A sorted vector (where you
1113 don't delete duplicate entries) or some other approach is almost always
1114 better.</p>
1116 <p>The various hash_set implementations (exposed portably by
1117 "llvm/ADT/hash_set") is a simple chained hashtable. This algorithm is as malloc
1118 intensive as std::set (performing an allocation for each element inserted,
1119 thus having really high constant factors) but (usually) provides O(1)
1120 insertion/deletion of elements. This can be useful if your elements are large
1121 (thus making the constant-factor cost relatively low) or if comparisons are
1122 expensive. Element iteration does not visit elements in a useful order.</p>
1124 </div>
1126 <!-- ======================================================================= -->
1129 </div>
1132 Map-like containers are useful when you want to associate data to a key. As
1133 usual, there are a lot of different ways to do this. :)
1134 </div>
1136 <!-- _______________________________________________________________________ -->
1139 </div>
1143 <p>
1144 If your usage pattern follows a strict insert-then-query approach, you can
1146 for set-like containers</a>. The only difference is that your query function
1147 (which uses std::lower_bound to get efficient log(n) lookup) should only compare
1148 the key, not both the key and value. This yields the same advantages as sorted
1149 vectors for sets.
1150 </p>
1151 </div>
1153 <!-- _______________________________________________________________________ -->
1156 </div>
1160 <p>
1161 Strings are commonly used as keys in maps, and they are difficult to support
1162 efficiently: they are variable length, inefficient to hash and compare when
1163 long, expensive to copy, etc. StringMap is a specialized container designed to
1164 cope with these issues. It supports mapping an arbitrary range of bytes to an
1165 arbitrary other object.</p>
1167 <p>The StringMap implementation uses a quadratically-probed hash table, where
1168 the buckets store a pointer to the heap allocated entries (and some other
1169 stuff). The entries in the map must be heap allocated because the strings are
1170 variable length. The string data (key) and the element object (value) are
1171 stored in the same allocation with the string data immediately after the element
1172 object. This container guarantees the "<tt>(char*)(&Value+1)</tt>" points
1173 to the key string for a value.</p>
1175 <p>The StringMap is very fast for several reasons: quadratic probing is very
1176 cache efficient for lookups, the hash value of strings in buckets is not
1177 recomputed when lookup up an element, StringMap rarely has to touch the
1178 memory for unrelated objects when looking up a value (even when hash collisions
1179 happen), hash table growth does not recompute the hash values for strings
1180 already in the table, and each pair in the map is store in a single allocation
1181 (the string data is stored in the same allocation as the Value of a pair).</p>
1183 <p>StringMap also provides query methods that take byte ranges, so it only ever
1184 copies a string if a value is inserted into the table.</p>
1185 </div>
1187 <!-- _______________________________________________________________________ -->
1190 </div>
1193 <p>
1194 IndexedMap is a specialized container for mapping small dense integers (or
1195 values that can be mapped to small dense integers) to some other type. It is
1196 internally implemented as a vector with a mapping function that maps the keys to
1197 the dense integer range.
1198 </p>
1200 <p>
1201 This is useful for cases like virtual registers in the LLVM code generator: they
1202 have a dense mapping that is offset by a compile-time constant (the first
1203 virtual register ID).</p>
1205 </div>
1207 <!-- _______________________________________________________________________ -->
1210 </div>
1214 <p>
1215 DenseMap is a simple quadratically probed hash table. It excels at supporting
1216 small keys and values: it uses a single allocation to hold all of the pairs that
1217 are currently inserted in the map. DenseMap is a great way to map pointers to
1218 pointers, or map other small types to each other.
1219 </p>
1221 <p>
1222 There are several aspects of DenseMap that you should be aware of, however. The
1223 iterators in a densemap are invalidated whenever an insertion occurs, unlike
1224 map. Also, because DenseMap allocates space for a large number of key/value
1225 pairs (it starts with 64 by default), it will waste a lot of space if your keys
1226 or values are large. Finally, you must implement a partial specialization of
1227 DenseMapKeyInfo for the key that you want, if it isn't already supported. This
1228 is required to tell DenseMap about two special marker values (which can never be
1229 inserted into the map) that it needs internally.</p>
1231 </div>
1233 <!-- _______________________________________________________________________ -->
1236 </div>
1240 <p>
1242 a single allocation per pair inserted into the map, it offers log(n) lookup with
1243 an extremely large constant factor, imposes a space penalty of 3 pointers per
1244 pair in the map, etc.</p>
1246 <p>std::map is most useful when your keys or values are very large, if you need
1247 to iterate over the collection in sorted order, or if you need stable iterators
1248 into the map (i.e. they don't get invalidated if an insertion or deletion of
1249 another element takes place).</p>
1251 </div>
1253 <!-- _______________________________________________________________________ -->
1256 </div>
1260 <p>
1261 The STL provides several other options, such as std::multimap and the various
1264 <p>std::multimap is useful if you want to map a key to multiple values, but has
1265 all the drawbacks of std::map. A sorted vector or some other approach is almost
1266 always better.</p>
1268 <p>The various hash_map implementations (exposed portably by
1269 "llvm/ADT/hash_map") are simple chained hash tables. This algorithm is as
1270 malloc intensive as std::map (performing an allocation for each element
1271 inserted, thus having really high constant factors) but (usually) provides O(1)
1272 insertion/deletion of elements. This can be useful if your elements are large
1273 (thus making the constant-factor cost relatively low) or if comparisons are
1274 expensive. Element iteration does not visit elements in a useful order.</p>
1276 </div>
1279 <!-- *********************************************************************** -->
1282 </div>
1283 <!-- *********************************************************************** -->
1287 <p>This section describes how to perform some very simple transformations of
1288 LLVM code. This is meant to give examples of common idioms used, showing the
1290 you should also read about the main classes that you will be working with. The
1292 and descriptions of the main classes that you should know about.</p>
1294 </div>
1296 <!-- NOTE: this section should be heavy on example code -->
1297 <!-- ======================================================================= -->
1300 </div>
1304 <p>The LLVM compiler infrastructure have many different data structures that may
1305 be traversed. Following the example of the C++ standard template library, the
1306 techniques used to traverse these various data structures are all basically the
1309 function returns an iterator pointing to one past the last valid element of the
1311 between the two operations.</p>
1313 <p>Because the pattern for iteration is common across many different aspects of
1314 the program representation, the standard template library algorithms may be used
1315 on them, and it is easier to remember how to iterate. First we show a few common
1316 examples of the data structures that need to be traversed. Other data
1317 structures are traversed in very similar ways.</p>
1319 </div>
1321 <!-- _______________________________________________________________________ -->
1326 </div>
1331 transform in some way; in particular, you'd like to manipulate its
1338 <pre>
1345 </pre>
1346 </div>
1348 <p>Note that i can be used as if it were a pointer for the purposes of
1350 because the indirection operator is overloaded for the iterator
1354 </div>
1356 <!-- _______________________________________________________________________ -->
1361 </div>
1366 easy to iterate over the individual instructions that make up
1371 <pre>
1377 </pre>
1378 </div>
1380 <p>However, this isn't really the best way to print out the contents of a
1382 anything you'll care about, you could have just invoked the print routine on the
1385 </div>
1387 <!-- _______________________________________________________________________ -->
1392 </div>
1401 small example that shows how to dump all instructions in a function to the standard error stream:<p>
1404 <pre>
1408 for (inst_iterator i = inst_begin(F), e = inst_end(F); i != e; ++i)
1410 </pre>
1411 </div>
1414 work list with its initial contents. For example, if you wanted to
1416 F, all you would need to do is something like:</p>
1419 <pre>
1421 worklist.insert(inst_begin(F), inst_end(F));
1422 </pre>
1423 </div>
1428 </div>
1430 <!-- _______________________________________________________________________ -->
1433 vice-versa)</a>
1434 </div>
1438 <p>Sometimes, it'll be useful to grab a reference (or pointer) to a class
1439 instance when all you've got at hand is an iterator. Well, extracting
1440 a reference or a pointer from an iterator is very straight-forward.
1445 <pre>
1448 const Instruction& inst = *j;
1449 </pre>
1450 </div>
1452 <p>However, the iterators you'll be working with in the LLVM framework are
1453 special: they will automatically convert to a ptr-to-instance type whenever they
1454 need to. Instead of dereferencing the iterator and then taking the address of
1455 the result, you can simply assign the iterator to the proper pointer type and
1456 you get the dereference and address-of operation as a result of the assignment
1457 (behind the scenes, this is a result of overloading casting mechanisms). Thus
1458 the last line of the last example,</p>
1461 <pre>
1462 Instruction* pinst = &*i;
1463 </pre>
1464 </div>
1469 <pre>
1470 Instruction* pinst = i;
1471 </pre>
1472 </div>
1474 <p>It's also possible to turn a class pointer into the corresponding iterator,
1475 and this is a constant time operation (very efficient). The following code
1476 snippet illustrates use of the conversion constructors provided by LLVM
1477 iterators. By using these, you can explicitly grab the iterator of something
1478 without actually obtaining it via iteration over some structure:</p>
1481 <pre>
1482 void printNextInstruction(Instruction* inst) {
1483 BasicBlock::iterator it(inst);
1486 }
1487 </pre>
1488 </div>
1490 </div>
1492 <!--_______________________________________________________________________-->
1495 example</a>
1496 </div>
1500 <p>Say that you're writing a FunctionPass and would like to count all the
1504 much more straight-forward manner, but this example will allow us to explore how
1506 is what we want to do:</p>
1509 <pre>
1510 initialize callCounter to zero
1511 for each Function f in the Module
1512 for each BasicBlock b in f
1513 for each Instruction i in b
1514 if (i is a CallInst and calls the given function)
1515 increment callCounter
1516 </pre>
1517 </div>
1519 <p>And the actual code is (remember, because we're writing a
1524 <pre>
1525 Function* targetFunc = ...;
1527 class OurFunctionPass : public FunctionPass {
1528 public:
1529 OurFunctionPass(): callCounter(0) { }
1531 virtual runOnFunction(Function& F) {
1532 for (Function::iterator b = F.begin(), be = F.end(); b != be; ++b) {
1540 if (callInst->getCalledFunction() == targetFunc)
1541 ++callCounter;
1542 }
1543 }
1544 }
1545 }
1547 private:
1548 unsigned callCounter;
1549 };
1550 </pre>
1551 </div>
1553 </div>
1555 <!--_______________________________________________________________________-->
1558 </div>
1562 <p>You may have noticed that the previous example was a bit oversimplified in
1563 that it did not deal with call sites generated by 'invoke' instructions. In
1564 this, and in other situations, you may find that you want to treat
1567 less closely-related things. For these cases, LLVM provides a handy wrapper
1568 class called <a
1575 reference and it should not be dynamically allocated or deallocated using
1577 assignable and constructable, with costs equivalents to that of a bare pointer.
1578 If you look at its definition, it has only a single pointer member.</p>
1580 </div>
1582 <!--_______________________________________________________________________-->
1585 </div>
1599 <pre>
1600 Function* F = ...;
1606 }
1607 </pre>
1608 </div>
1615 all of the values that a particular instruction uses (that is, the operands of
1619 <pre>
1620 Instruction* pi = ...;
1623 Value* v = *i;
1625 }
1626 </pre>
1627 </div>
1629 <!--
1630 def-use chains ("finding all users of"): Value::use_begin/use_end
1631 use-def chains ("finding all values used"): User::op_begin/op_end [op=operand]
1632 -->
1634 </div>
1636 <!-- ======================================================================= -->
1639 </div>
1643 <p>There are some primitive transformation operations present in the LLVM
1644 infrastructure that are worth knowing about. When performing
1645 transformations, it's fairly common to manipulate the contents of basic
1646 blocks. This section describes some of the common methods for doing so
1647 and gives example code.</p>
1649 </div>
1651 <!--_______________________________________________________________________-->
1655 </div>
1662 constructor for the kind of instruction to instantiate and provide the necessary
1667 <pre>
1668 AllocaInst* ai = new AllocaInst(Type::IntTy);
1669 </pre>
1670 </div>
1674 subclass is likely to have varying default parameters which change the semantics
1675 of the instruction, so refer to the <a
1681 <p>It is very useful to name the values of instructions when you're able to, as
1682 this facilitates the debugging of your transformations. If you end up looking
1683 at generated LLVM machine code, you definitely want to have logical names
1684 associated with the results of instructions! By supplying a value for the
1686 associate a logical name with the result of the instruction's execution at
1687 run time. For example, say that I'm writing a transformation that dynamically
1688 allocates space for an integer on the stack, and that integer is going to be
1689 used as some kind of index by some other code. To accomplish this, I place an
1695 <pre>
1697 </pre>
1698 </div>
1701 execution value, which is a pointer to an integer on the run time stack.</p>
1708 <ul>
1709 <li>Insertion into an explicit instruction list
1716 <pre>
1717 BasicBlock *pb = ...;
1718 Instruction *pi = ...;
1719 Instruction *newInst = new Instruction(...);
1722 </pre>
1723 </div>
1727 classes provide constructors which take a pointer to a
1729 looked like: </p>
1732 <pre>
1733 BasicBlock *pb = ...;
1734 Instruction *newInst = new Instruction(...);
1737 </pre>
1738 </div>
1743 <pre>
1744 BasicBlock *pb = ...;
1745 Instruction *newInst = new Instruction(..., pb);
1746 </pre>
1747 </div>
1749 <p>which is much cleaner, especially if you are creating
1750 long instruction streams.</p></li>
1752 <li>Insertion into an implicit instruction list
1755 are implicitly associated with an existing instruction list: the instruction
1756 list of the enclosing basic block. Thus, we could have accomplished the same
1758 </p>
1761 <pre>
1762 Instruction *pi = ...;
1763 Instruction *newInst = new Instruction(...);
1766 </pre>
1767 </div>
1769 <p>In fact, this sequence of steps occurs so frequently that the
1771 constructors which take (as a default parameter) a pointer to an
1775 provided instruction, immediately before that instruction. Using an
1777 parameter, the above code becomes:</p>
1780 <pre>
1781 Instruction* pi = ...;
1782 Instruction* newInst = new Instruction(..., pi);
1783 </pre>
1784 </div>
1786 <p>which is much cleaner, especially if you're creating a lot of
1788 </ul>
1790 </div>
1792 <!--_______________________________________________________________________-->
1795 </div>
1799 <p>Deleting an instruction from an existing sequence of instructions that form a
1801 you must have a pointer to the instruction that you wish to delete. Second, you
1802 need to obtain the pointer to that instruction's basic block. You use the
1803 pointer to the basic block to get its list of instructions and then use the
1804 erase function to remove your instruction. For example:</p>
1807 <pre>
1811 BB->getInstList().erase(I);
1812 </pre>
1813 </div>
1815 </div>
1817 <!--_______________________________________________________________________-->
1821 </div>
1827 <p>Including "<a href="/doxygen/BasicBlockUtils_8h-source.html">llvm/Transforms/Utils/BasicBlockUtils.h</a>"
1833 <ul>
1836 <p>This function replaces all uses (within a basic block) of a given
1837 instruction with a value, and then removes the original instruction. The
1838 following example illustrates the replacement of the result of a particular
1840 pointer to an integer.</p>
1843 <pre>
1844 AllocaInst* instToReplace = ...;
1845 BasicBlock::iterator ii(instToReplace);
1848 Constant::getNullValue(PointerType::get(Type::IntTy)));
1849 </pre></div></li>
1853 <p>This function replaces a particular instruction with another
1854 instruction. The following example illustrates the replacement of one
1858 <pre>
1859 AllocaInst* instToReplace = ...;
1860 BasicBlock::iterator ii(instToReplace);
1864 </pre></div></li>
1865 </ul>
1873 information.</p>
1875 <!-- Value::replaceAllUsesWith User::replaceUsesOfWith Point out:
1876 include/llvm/Transforms/Utils/ especially BasicBlockUtils.h with:
1877 ReplaceInstWithValue, ReplaceInstWithInst -->
1879 </div>
1881 <!-- *********************************************************************** -->
1884 </div>
1885 <!-- *********************************************************************** -->
1888 <p>
1889 This section describes some of the advanced or obscure API's that most clients
1890 do not need to be aware of. These API's tend manage the inner workings of the
1891 LLVM system, and only need to be accessed in unusual circumstances.
1892 </p>
1893 </div>
1895 <!-- ======================================================================= -->
1898 </div>
1902 <p>
1903 The LLVM type system has a very simple goal: allow clients to compare types for
1904 structural equality with a simple pointer comparison (aka a shallow compare).
1905 This goal makes clients much simpler and faster, and is used throughout the LLVM
1906 system.
1907 </p>
1909 <p>
1910 Unfortunately achieving this goal is not a simple matter. In particular,
1911 recursive types and late resolution of opaque types makes the situation very
1912 difficult to handle. Fortunately, for the most part, our implementation makes
1913 most clients able to be completely unaware of the nasty internal details. The
1914 primary case where clients are exposed to the inner workings of it are when
1915 building a recursive type. In addition to this case, the LLVM bytecode reader,
1916 assembly parser, and linker also have to be aware of the inner workings of this
1917 system.
1918 </p>
1920 <p>
1921 For our purposes below, we need three concepts. First, an "Opaque Type" is
1924 opaque type as part of its type graph (for example "<tt>{ opaque, i32 }</tt>").
1925 Third, a concrete type is a type that is not an abstract type (e.g. "<tt>{ i32,
1926 float }</tt>").
1927 </p>
1929 </div>
1931 <!-- ______________________________________________________________________ -->
1934 </div>
1938 <p>
1939 Because the most common question is "how do I build a recursive type with LLVM",
1940 we answer it now and explain it as we go. Here we include enough to cause this
1941 to be emitted to an output .ll file:
1942 </p>
1945 <pre>
1946 %mylist = type { %mylist*, i32 }
1947 </pre>
1948 </div>
1950 <p>
1951 To build this, use the following LLVM APIs:
1952 </p>
1955 <pre>
1959 Elts.push_back(PointerType::get(StructTy));
1960 Elts.push_back(Type::IntTy);
1961 StructType *NewSTy = StructType::get(Elts);
1965 cast<OpaqueType>(StructTy.get())-><a href="#refineAbstractTypeTo">refineAbstractTypeTo</a>(NewSTy);
1967 // <i>NewSTy is potentially invalidated, but StructTy (a <a href="#PATypeHolder">PATypeHolder</a>) is</i>
1973 </pre>
1974 </div>
1976 <p>
1977 This code shows the basic approach used to build recursive types: build a
1978 non-recursive type using 'opaque', then use type unification to close the cycle.
1979 The type unification step is performed by the <tt><a
1981 described next. After that, we describe the <a
1983 </p>
1985 </div>
1987 <!-- ______________________________________________________________________ -->
1990 </div>
1993 <p>
1995 While this method is actually a member of the DerivedType class, it is most
1996 often used on OpaqueType instances. Type unification is actually a recursive
1997 process. After unification, types can become structurally isomorphic to
1998 existing types, and all duplicates are deleted (to preserve pointer equality).
1999 </p>
2001 <p>
2002 In the example above, the OpaqueType object is definitely deleted.
2003 Additionally, if there is an "{ \2*, i32}" type already created in the system,
2005 a type is deleted, any "Type*" pointers in the program are invalidated. As
2008 types can never move or be deleted). To deal with this, the <a
2010 reference to a possibly refined type, and the <a
2012 complex datastructures.
2013 </p>
2015 </div>
2017 <!-- ______________________________________________________________________ -->
2020 </div>
2023 <p>
2024 PATypeHolder is a form of a "smart pointer" for Type objects. When VMCore
2025 happily goes about nuking types that become isomorphic to existing types, it
2026 automatically updates all PATypeHolder objects to point to the new type. In the
2027 example above, this allows the code to maintain a pointer to the resultant
2028 resolved recursive type, even though the Type*'s are potentially invalidated.
2029 </p>
2031 <p>
2032 PATypeHolder is an extremely light-weight object that uses a lazy union-find
2033 implementation to update pointers. For example the pointer from a Value to its
2034 Type is maintained by PATypeHolder objects.
2035 </p>
2037 </div>
2039 <!-- ______________________________________________________________________ -->
2042 </div>
2046 <p>
2047 Some data structures need more to perform more complex updates when types get
2048 resolved. To support this, a class can derive from the AbstractTypeUser class.
2049 This class
2050 allows it to get callbacks when certain types are resolved. To register to get
2051 callbacks for a particular type, the DerivedType::{add/remove}AbstractTypeUser
2052 methods can be called on a type. Note that these methods only work for <i>
2053 abstract</i> types. Concrete types (those that do not include any opaque
2054 objects) can never be refined.
2055 </p>
2056 </div>
2059 <!-- ======================================================================= -->
2063 </div>
2073 names for types.</p>
2076 by most clients. It should only be used when iteration over the symbol table
2077 names themselves are required, which is very special purpose. Note that not
2078 all LLVM
2080 an empty name) do not exist in the symbol table.
2081 </p>
2083 <p>These symbol tables support iteration over the values/types in the symbol
2088 appropriate symbol table. For types, use the Module::addTypeName method to
2089 insert entries into the symbol table.</p>
2091 </div>
2095 <!-- *********************************************************************** -->
2098 </div>
2099 <!-- *********************************************************************** -->
2105 <p>The Core LLVM classes are the primary means of representing the program
2106 being inspected or transformed. The core LLVM classes are defined in
2110 </div>
2112 <!-- ======================================================================= -->
2115 </div>
2123 subclasses. They are hidden because they offer no useful functionality beyond
2127 named, but this is not a requirement. There exists exactly
2128 one instance of a given shape at any one time. This allows type equality to
2129 be performed with address equality of the Type Instance. That is, given two
2131 </p>
2132 </div>
2134 <!-- _______________________________________________________________________ -->
2137 </div>
2141 <ul>
2145 floating point types.</li>
2148 an OpaqueType anywhere in its definition).</li>
2151 that don't have a size are abstract types, labels and void.</li>
2153 </ul>
2154 </div>
2156 <!-- _______________________________________________________________________ -->
2159 </div>
2161 <dl>
2163 <dd>Subclass of DerivedType that represents integer types of any bit width.
2166 <ul>
2168 type of a specific bit width.</li>
2170 type.</li>
2171 </ul>
2172 </dd>
2174 <dd>This is subclassed by ArrayType and PointerType
2175 <ul>
2177 of the elements in the sequential type. </li>
2178 </ul>
2179 </dd>
2181 <dd>This is a subclass of SequentialType and defines the interface for array
2182 types.
2183 <ul>
2185 elements in the array. </li>
2186 </ul>
2187 </dd>
2191 <dd>Subclass of SequentialType for vector types. A
2192 vector type is similar to an ArrayType but is distinguished because it is
2193 a first class type wherease ArrayType is not. Vector types are used for
2194 vector operations and are usually small vectors of of an integer or floating
2195 point type.</dd>
2199 <dd>Subclass of DerivedTypes for function types.
2200 <ul>
2202 function</li>
2204 return type of the function.</li>
2206 the type of the ith parameter.</li>
2208 number of formal parameters.</li>
2209 </ul>
2210 </dd>
2212 <dd>Sublcass of DerivedType for abstract types. This class
2213 defines no content and is used as a placeholder for some other type. Note
2214 that OpaqueType is used (temporarily) during type resolution for forward
2215 references of types. Once the referenced type is resolved, the OpaqueType
2216 is replaced with the actual type. OpaqueType can also be used for data
2217 abstraction. At link time opaque types can be resolved to actual types
2218 of the same name.</dd>
2219 </dl>
2220 </div>
2224 <!-- ======================================================================= -->
2227 </div>
2236 programs. An LLVM module is effectively either a translation unit of the
2237 original program or a combination of several translation units merged by the
2242 helpful member functions that try to make common operations easy.</p>
2244 </div>
2246 <!-- _______________________________________________________________________ -->
2249 </div>
2253 <ul>
2255 </ul>
2258 provide a name for it (probably based on the name of the translation unit).</p>
2260 <ul>
2267 <p>These are forwarding methods that make it easy to access the contents of
2269 list.</p></li>
2274 necessary to use when you need to update the list or perform a complex
2275 action that doesn't have a forwarding method.</p>
2278 </ul>
2280 <hr>
2282 <ul>
2290 <p> These are forwarding methods that make it easy to access the contents of
2298 use when you need to update the list or perform a complex action that
2299 doesn't have a forwarding method.</p>
2302 </ul>
2304 <hr>
2306 <ul>
2313 </ul>
2315 <hr>
2317 <ul>
2330 external declaration for the function and return it.</p></li>
2337 string.</p></li>
2344 name, true is returned and the <a
2346 </ul>
2348 </div>
2351 <!-- ======================================================================= -->
2354 </div>
2359 <br>
2363 base. It represents a typed value that may be used (among other things) as an
2371 for a program. For example, an incoming argument to a function (represented
2373 every instruction in the function that references the argument. To keep track
2379 methods, shown below.</p>
2383 method. In addition, all LLVM values can be named. The "name" of the
2387 <pre>
2389 </pre>
2390 </div>
2393 that the name of any value may be missing (an empty string), so names should
2395 debugging printouts), they should not be used to keep track of values or map
2399 <p>One important aspect of LLVM is that there is no distinction between an SSA
2400 variable and the operation that produces it. Because of this, any reference to
2401 the value produced by an instruction (or the value available as an incoming
2402 argument, for example) is represented as a direct pointer to the instance of
2403 the class that
2404 represents this value. Although this may take some getting used to, it
2405 simplifies the representation and makes it easier to manipulate.</p>
2407 </div>
2409 <!-- _______________________________________________________________________ -->
2412 </div>
2416 <ul>
2418 use-list<br>
2420 the use-list<br>
2422 value.<br>
2425 the use-list.<br>
2427 use-list.<br>
2429 element in the list.
2430 <p> These methods are the interface to access the def-use
2431 information in LLVM. As with all other iterators in LLVM, the naming
2433 </li>
2436 </li>
2442 </li>
2447 "<tt>V</tt>" instead. For example, if you detect that an instruction always
2448 produces a constant value (for example through constant folding), you can
2449 replace all uses of the instruction with the constant like this:</p>
2452 <pre>
2453 Inst->replaceAllUsesWith(ConstVal);
2454 </pre>
2455 </div>
2457 </ul>
2459 </div>
2461 <!-- ======================================================================= -->
2464 </div>
2468 <p>
2481 Single Assignment (SSA) form, there can only be one definition referred to,
2482 allowing this direct connection. This connection provides the use-def
2483 information in LLVM.</p>
2485 </div>
2487 <!-- _______________________________________________________________________ -->
2490 </div>
2495 an index access interface and through an iterator based interface.</p>
2497 <ul>
2501 convenient form for direct access.</p></li>
2504 list<br>
2506 the operand list.<br>
2508 operand list.
2509 <p> Together, these methods make up the iterator based interface to
2511 </ul>
2513 </div>
2515 <!-- ======================================================================= -->
2518 </div>
2529 instructions. It provides only a few methods, but is a very commonly used
2531 opcode (instruction type) and the parent <a
2533 into. To represent a specific type of instruction, one of many subclasses of
2542 file contains some meta-data about the various different types of instructions
2543 in LLVM. It describes the enum values that are used as opcodes (for example
2548 this file confuses doxygen, so these enum values don't show up correctly in the
2551 </div>
2553 <!-- _______________________________________________________________________ -->
2556 class</a>
2557 </div>
2559 <ul>
2561 <p>This subclasses represents all two operand instructions whose operands
2562 must be the same type, except for the comparison instructions.</p></li>
2565 common operations on cast instructions.</p>
2567 <p>This subclass respresents the two comparison instructions,
2571 <p>This subclass is the parent of all terminator instructions (those which
2572 can terminate a block).</p>
2573 </ul>
2574 </div>
2576 <!-- _______________________________________________________________________ -->
2579 class</a>
2580 </div>
2584 <ul>
2589 <p>Returns true if the instruction writes to memory, i.e. it is a
2594 <p>Returns another instance of the specified instruction, identical
2595 in all ways to the original except that the instruction has no parent
2597 and it has no name</p></li>
2598 </ul>
2600 </div>
2602 <!-- ======================================================================= -->
2605 </div>
2609 <p>Constant represents a base class for different types of constants. It
2610 is subclassed by ConstantInt, ConstantArray, etc. for representing
2612 a subclass, which represents the address of a global variable or function.
2613 </p>
2615 </div>
2617 <!-- _______________________________________________________________________ -->
2620 <ul>
2621 <li>ConstantInt : This subclass of Constant represents an integer constant of
2622 any width.
2623 <ul>
2625 value of this constant, an APInt value.</li>
2627 value to an int64_t via sign extension. If the value (not the bit width)
2628 of the APInt is too large to fit in an int64_t, an assertion will result.
2629 For this reason, use of this method is discouraged.</li>
2631 value to a uint64_t via zero extension. IF the value (not the bit width)
2632 of the APInt is too large to fit in a uint64_t, an assertion will result.
2633 For this reason, use of this method is discouraged.</li>
2636 The type is implied as the IntegerType that corresponds to the bit width
2639 Returns the ConstantInt object that represents the value provided by
2641 </ul>
2642 </li>
2643 <li>ConstantFP : This class represents a floating point constant.
2644 <ul>
2646 this constant. </li>
2647 </ul>
2648 </li>
2649 <li>ConstantArray : This represents a constant array.
2650 <ul>
2652 a vector of component constants that makeup this array. </li>
2653 </ul>
2654 </li>
2655 <li>ConstantStruct : This represents a constant struct.
2656 <ul>
2658 a vector of component constants that makeup this array. </li>
2659 </ul>
2660 </li>
2661 <li>GlobalValue : This represents either a global variable or a function. In
2662 either case, the value is a constant fixed address (after linking).
2663 </li>
2664 </ul>
2665 </div>
2668 <!-- ======================================================================= -->
2671 </div>
2678 Class</a><br>
2685 Because they are visible at global scope, they are also subject to linking with
2686 other globals defined in different translation units. To control the linking
2693 unit, and does not participate in linking. If it has external linkage, it is
2694 visible to external code, and does participate in linking. In addition to
2700 global is always a pointer to its contents. It is important to remember this
2705 the address of the first element of this array and the value of the
2713 </div>
2715 <!-- _______________________________________________________________________ -->
2718 class</a>
2719 </div>
2723 <ul>
2728 <p> </p>
2729 </li>
2732 GlobalValue is currently embedded into.</p></li>
2733 </ul>
2735 </div>
2737 <!-- ======================================================================= -->
2740 </div>
2753 actually one of the more complex classes in the LLVM heirarchy because it must
2761 ordering of the blocks in the function, which indicate how the code will be
2762 layed out by the backend. Additionally, the first <a
2765 block. There are no implicit exit nodes, and in fact there may be multiple exit
2769 function hasn't been linked in yet.</p>
2779 LLVM feature that is only used when you have to look up a value by name. Aside
2781 internally to make sure that there are not conflicts between the names of <a
2788 is its address (after linking) which is guaranteed to be constant.</p>
2789 </div>
2791 <!-- _______________________________________________________________________ -->
2794 class</a>
2795 </div>
2799 <ul>
2804 the the program. The constructor must specify the type of the function to
2805 create and what type of linkage the function should have. The <a
2807 specifies the formal arguments and return value for the function. The same
2810 in which the function is defined. If this argument is provided, the function
2811 will automatically be inserted into that module's list of
2812 functions.</p></li>
2817 function is "external", it does not have a body, and thus must be resolved
2818 by linking with a function defined in a different translation unit.</p></li>
2826 <p>These are forwarding methods that make it easy to access the contents of
2828 list.</p></li>
2833 is necessary to use when you need to update the list or perform a complex
2834 action that doesn't have a forwarding method.</p></li>
2837 iterator<br>
2843 <p>These are forwarding methods that make it easy to access the contents of
2845 list.</p></li>
2850 necessary to use when you need to update the list or perform a complex
2851 action that doesn't have a forwarding method.</p></li>
2856 function. Because the entry block for the function is always the first
2865 function.</p></li>
2871 </ul>
2873 </div>
2875 <!-- ======================================================================= -->
2878 </div>
2884 <br>
2886 Class</a><br>
2892 <p>Global variables are represented with the (suprise suprise)
2895 always referenced by their address (global values must live in memory, so their
2896 "name" refers to their constant address). See
2898 variables may have an initial value (which must be a
2900 they may be marked as "constant" themselves (indicating that their contents
2901 never change at runtime).</p>
2902 </div>
2904 <!-- _______________________________________________________________________ -->
2908 </div>
2912 <ul>
2917 <p>Create a new global variable of the specified type. If
2919 unchanging for the program. The Linkage parameter specifies the type of
2920 linkage (internal, external, weak, linkonce, appending) for the variable. If
2921 the linkage is InternalLinkage, WeakLinkage, or LinkOnceLinkage, then
2922 the resultant global variable will have internal linkage. AppendingLinkage
2923 concatenates together all instances (in different translation units) of the
2924 variable into a single variable but is only applicable to arrays. See
2926 further details on linkage types. Optionally an initializer, a name, and the
2927 module to put the variable into may be specified for the global variable as
2928 well.</p></li>
2932 <p>Returns true if this is a global variable that is known not to
2933 be modified at runtime.</p></li>
2942 to call this method if there is no initializer.</p></li>
2943 </ul>
2945 </div>
2948 <!-- ======================================================================= -->
2951 </div>
2958 Class</a><br>
2961 <p>This class represents a single entry multiple exit section of the code,
2962 commonly known as a basic block by the compiler community. The
2965 Matching the language definition, the last element of this list of instructions
2966 is always a terminator instruction (a subclass of the <a
2969 <p>In addition to tracking the list of instructions that make up the block, the
2978 </div>
2980 <!-- _______________________________________________________________________ -->
2983 class</a>
2984 </div>
2987 <ul>
2993 insertion into a function. The constructor optionally takes a name for the new
2996 automatically inserted at the end of the specified <a
3004 STL-style functions for accessing the instruction list.
3006 <p>These methods and typedefs are forwarding functions that have the same
3007 semantics as the standard library methods of the same names. These methods
3008 expose the underlying instruction list of a basic block in a way that is easy to
3009 manipulate. To get the full complement of container operations (including
3011 method.</p></li>
3015 <p>This method is used to get access to the underlying container that actually
3016 holds the Instructions. This method must be used when there isn't a forwarding
3018 to perform. Because there are no forwarding functions for "updating"
3019 operations, you need to use this if you want to update the contents of a
3025 embedded into, or a null pointer if it is homeless.</p></li>
3029 <p> Returns a pointer to the terminator instruction that appears at the end of
3031 instruction in the block is not a terminator, then a null pointer is
3032 returned.</p></li>
3034 </ul>
3036 </div>
3039 <!-- ======================================================================= -->
3042 </div>
3046 <p>This subclass of Value defines the interface for incoming formal
3047 arguments to a function. A Function maintains a list of its formal
3048 arguments. An argument has a pointer to the parent Function.</p>
3050 </div>
3052 <!-- *********************************************************************** -->
3053 <hr>
3054 <address>
3063 Last modified: $Date$
3064 </address>
3066 </body>
3067 </html>