| blob | f6854078b6c95c51ba142307ac2d24c07f7a7f89 |
3 <html>
4 <head>
8 </head>
9 <body>
12 LLVM Programmer's Manual
13 </div>
15 <ol>
18 <ul>
20 <!--
21 <li>The <tt>-time-passes</tt> option</li>
22 <li>How to use the LLVM Makefile system</li>
23 <li>How to write a regression test</li>
25 -->
26 </ul>
27 </li>
29 <ul>
34 <ul>
37 </ul>
38 </li>
40 option</a>
41 <ul>
44 </ul>
45 </li>
47 option</a></li>
48 <!--
49 <li>The <tt>InstVisitor</tt> template
50 <li>The general graph API
51 -->
53 </ul>
54 </li>
56 <ul>
58 <ul>
67 </ul></li>
69 <ul>
79 </ul></li>
81 <ul>
90 </ul></li>
92 <!--<ul>
93 todo
96 <ul>
100 </ul></li>
101 </ul>
102 </li>
104 <ul>
106 <ul>
114 class pointer</a> </li>
116 complex example</a> </li>
118 the same way</a> </li>
120 use-def chains</a> </li>
122 successors of blocks</a></li>
123 </ul>
124 </li>
126 <ul>
133 </ul>
134 </li>
136 <!--
137 <li>Working with the Control Flow Graph
138 <ul>
139 <li>Accessing predecessors and successors of a <tt>BasicBlock</tt>
140 <li>
141 <li>
142 </ul>
143 -->
144 </ul>
145 </li>
148 <ul>
150 </a></li>
155 </ul>
156 </li>
159 <ul>
161 <ul>
166 </ul></li>
168 <li><a href="#SymbolTable">The <tt>ValueSymbolTable</tt> and <tt>TypeSymbolTable</tt> classes</a></li>
169 <li><a href="#UserLayout">The <tt>User</tt> and owned <tt>Use</tt> classes' memory layout</a></li>
170 </ul></li>
173 <ul>
177 <ul>
179 <ul>
182 <ul>
184 <ul>
187 </ul>
188 </li>
189 </ul>
190 </li>
191 </ul>
192 </li>
195 </ul>
196 </li>
197 </ul>
198 </li>
199 </ol>
208 </div>
210 <!-- *********************************************************************** -->
213 </div>
214 <!-- *********************************************************************** -->
218 <p>This document is meant to highlight some of the important classes and
219 interfaces available in the LLVM source-base. This manual is not
220 intended to explain what LLVM is, how it works, and what LLVM code looks
221 like. It assumes that you know the basics of LLVM and are interested
222 in writing transformations or otherwise analyzing or manipulating the
223 code.</p>
225 <p>This document should get you oriented so that you can find your
226 way in the continuously growing source code that makes up the LLVM
227 infrastructure. Note that this manual is not intended to serve as a
228 replacement for reading the source code, so if you think there should be
229 a method in one of these classes to do something, but it's not listed,
231 are provided to make this as easy as possible.</p>
233 <p>The first section of this document describes general information that is
234 useful to know when working in the LLVM infrastructure, and the second describes
235 the Core LLVM classes. In the future this manual will be extended with
236 information describing how to use extension libraries, such as dominator
237 information, CFG traversal routines, and useful utilities like the <tt><a
240 </div>
242 <!-- *********************************************************************** -->
245 </div>
246 <!-- *********************************************************************** -->
250 <p>This section contains general information that is useful if you are working
251 in the LLVM source-base, but that isn't specific to any particular API.</p>
253 </div>
255 <!-- ======================================================================= -->
258 </div>
262 <p>LLVM makes heavy use of the C++ Standard Template Library (STL),
263 perhaps much more than you are used to, or have seen before. Because of
264 this, you might want to do a little background reading in the
265 techniques used and capabilities of the library. There are many good
266 pages that discuss the STL, and several books on the subject that you
267 can get, so it will not be discussed in this document.</p>
271 <ol>
274 C++ Library reference</a> - an excellent reference for the STL and other parts
275 of the standard C++ library.</li>
278 O'Reilly book in the making. It has a decent Standard Library
279 Reference that rivals Dinkumware's, and is unfortunately no longer free since the
280 book has been published.</li>
283 Questions</a></li>
286 Contains a useful <a
291 Page</a></li>
295 the book).</a></li>
297 </ol>
301 to write maintainable code more than where to put your curly braces.</p>
303 </div>
305 <!-- ======================================================================= -->
308 </div>
312 <ol>
314 static and shared libraries across platforms</a></li>
315 </ol>
317 </div>
319 <!-- *********************************************************************** -->
322 </div>
323 <!-- *********************************************************************** -->
327 <p>Here we highlight some LLVM APIs that are generally useful and good to
328 know about when writing transformations.</p>
330 </div>
332 <!-- ======================================================================= -->
336 </div>
340 <p>The LLVM source-base makes extensive use of a custom form of RTTI.
342 operator, but they don't have some drawbacks (primarily stemming from
344 have a v-table). Because they are used so often, you must know what they
345 do and how they work. All of these templates are defined in the <a
347 file (note that you very rarely have to include this file directly).</p>
349 <dl>
353 "<tt>instanceof</tt>" operator. It returns true or false depending on whether
354 a reference or pointer points to an instance of the specified class. This can
355 be very useful for constraint checking of various sorts (example below).</p>
356 </dd>
361 converts a pointer or reference from a base class to a derived class, causing
362 an assertion failure if it is not really an instance of the right type. This
363 should be used in cases where you have some information that makes you believe
368 <pre>
370 if (isa<<a href="#Constant">Constant</a>>(V) || isa<<a href="#Argument">Argument</a>>(V) || isa<<a href="#GlobalValue">GlobalValue</a>>(V))
371 return true;
375 }
376 </pre>
377 </div>
381 operator.</p>
383 </dd>
388 It checks to see if the operand is of the specified type, and if so, returns a
389 pointer to it (this operator does not work with references). If the operand is
390 not of the correct type, a null pointer is returned. Thus, this works very
394 statement like this:</p>
397 <pre>
398 if (<a href="#AllocationInst">AllocationInst</a> *AI = dyn_cast<<a href="#AllocationInst">AllocationInst</a>>(Val)) {
400 }
401 </pre>
402 </div>
406 statement, which is very convenient.</p>
411 blocks to check for lots of different variants of classes. If you find
412 yourself wanting to do this, it is much cleaner and more efficient to use the
415 </dd>
421 argument (which it then propagates). This can sometimes be useful, allowing
422 you to combine several null checks into one.</p></dd>
428 as an argument (which it then propagates). This can sometimes be useful,
429 allowing you to combine several null checks into one.</p></dd>
431 </dl>
433 <p>These five templates can be used with any classes, whether they have a
434 v-table or not. To add support for these templates, you simply need to add
436 to. Describing this is currently outside the scope of this document, but there
437 are lots of examples in the LLVM source base.</p>
439 </div>
442 <!-- ======================================================================= -->
446 </div>
450 <p>Although LLVM generally does not do much string manipulation, we do have
451 several important APIs which take strings. Two important examples are the
452 Value class -- which has names for instructions, functions, etc. -- and the
453 StringMap class which is used extensively in LLVM and Clang.</p>
455 <p>These are generic classes, and they need to be able to accept strings which
456 may have embedded null characters. Therefore, they cannot simply take
458 clients to perform a heap allocation which is usually unnecessary. Instead,
460 passing strings efficiently.</p>
462 </div>
464 <!-- _______________________________________________________________________ -->
467 </div>
472 (a character array and a length) and supports the common operations available
475 <p>It can be implicitly constructed using a C style null-terminated string,
480 iterator find(StringRef Key);
481 </pre>
489 </pre>
495 for more information.</p>
498 pointers to external memory it is not generally safe to store an instance of the
499 class (unless you know that the external storage will not be freed). StringRef is
500 small and pervasive enough in LLVM that it should always be passed by value.</p>
502 </div>
504 <!-- _______________________________________________________________________ -->
507 </div>
512 strings. For example, a common LLVM paradigm is to name one instruction based on
513 the name of another instruction with a suffix, for example:</p>
516 <pre>
518 </pre>
519 </div>
523 which points to temporary (stack allocated) objects. Twines can be implicitly
524 constructed as the result of the plus operator applied to strings (i.e., a C
526 actual concatenation of strings until it is actually required, at which point
527 it can be efficiently rendered directly into a character array. This avoids
528 unnecessary heap allocation involved in constructing the temporary results of
529 string concatenation. See
531 for more information.</p>
534 and should almost never be stored or mentioned directly. They are intended
535 solely for use when defining a function which should be able to efficiently
536 accept concatenated strings.</p>
538 </div>
541 <!-- ======================================================================= -->
544 </div>
548 <p>Often when working on your pass you will put a bunch of debugging printouts
549 and other code into your pass. After you get it working, you want to remove
550 it, but you may need it again in the future (to work out new bugs that you run
551 across).</p>
553 <p> Naturally, because of this, you don't want to delete the debug printouts,
554 but you don't want them to always be noisy. A standard compromise is to comment
555 them out, allowing you to enable them if you need them in the future.</p>
559 this problem. Basically, you can put arbitrary code into the argument of the
564 <pre>
566 </pre>
567 </div>
572 <pre>
576 I am here!
577 </pre>
578 </div>
581 to not have to create "yet another" command line option for the debug output for
583 so they do not cause a performance impact at all (for the same reason, they
584 should also not contain side-effects!).</p>
587 enable or disable it directly in gdb. Just use "<tt>set DebugFlag=0</tt>" or
588 "<tt>set DebugFlag=1</tt>" from the gdb if the program is running. If the
589 program hasn't been started yet, you can always just run it with
592 </div>
594 <!-- _______________________________________________________________________ -->
598 </div>
604 generator). If you want to enable debug information with more fine-grained
606 option as follows:</p>
609 <pre>
610 #undef DEBUG_TYPE
612 #define DEBUG_TYPE "foo"
614 #undef DEBUG_TYPE
615 #define DEBUG_TYPE "bar"
617 #undef DEBUG_TYPE
618 #define DEBUG_TYPE ""
620 </pre>
621 </div>
626 <pre>
630 No debug type
631 'foo' debug type
632 'bar' debug type
633 No debug type (2)
635 'foo' debug type
637 'bar' debug type
638 </pre>
639 </div>
642 a file, to specify the debug type for the entire module (if you do this before
645 "bar", because there is no system in place to ensure that names do not
646 conflict. If two different modules use the same string, they will all be turned
647 on when the name is specified. This allows, for example, all debug information
649 even if the source lives in multiple files.</p>
653 statement. It takes an additional first parameter, which is the type to use. For
654 example, the preceding example could be written as:</p>
658 <pre>
663 </pre>
664 </div>
666 </div>
668 <!-- ======================================================================= -->
671 option</a>
672 </div>
679 keep track of what the LLVM compiler is doing and how effective various
680 optimizations are. It is useful to see what optimizations are contributing to
681 making a particular program run faster.</p>
683 <p>Often you may run your pass on some big program, and you're interested to see
684 how many times it makes a certain transformation. Although you can do this with
685 hand inspection, or some ad-hoc method, this is a real pain and not very useful
687 keep track of this information, and the calculated information is presented in a
688 uniform manner with the rest of the passes being executed.</p>
691 it are as follows:</p>
693 <ol>
697 <pre>
698 #define <a href="#DEBUG_TYPE">DEBUG_TYPE</a> "mypassname" <i>// This goes before any #includes.</i>
699 STATISTIC(NumXForms, "The # of times I did stuff");
700 </pre>
701 </div>
704 specified by the first argument. The pass name is taken from the DEBUG_TYPE
705 macro, and the description is taken from the second argument. The variable
711 <pre>
713 </pre>
714 </div>
716 </li>
717 </ol>
723 <pre>
726 </pre>
727 </div>
730 suite, it gives a report that looks like this:</p>
733 <pre>
734 7646 bitcodewriter - Number of normal instructions
735 725 bitcodewriter - Number of oversized instructions
736 129996 bitcodewriter - Number of bitcode bytes written
737 2817 raise - Number of insts DCEd or constprop'd
738 3213 raise - Number of cast-of-self removed
739 5046 raise - Number of expression trees converted
740 75 raise - Number of other getelementptr's formed
741 138 raise - Number of load/store peepholes
742 42 deadtypeelim - Number of unused typenames removed from symtab
743 392 funcresolve - Number of varargs functions resolved
744 27 globaldce - Number of global variables removed
745 2 adce - Number of basic blocks removed
746 134 cee - Number of branches revectored
747 49 cee - Number of setcc instruction eliminated
748 532 gcse - Number of loads removed
749 2919 gcse - Number of instructions removed
750 86 indvars - Number of canonical indvars added
751 87 indvars - Number of aux indvars removed
752 25 instcombine - Number of dead inst eliminate
753 434 instcombine - Number of insts combined
754 248 licm - Number of load insts hoisted
755 1298 licm - Number of insts hoisted to a loop pre-header
756 3 licm - Number of insts hoisted to multiple loop preds (bad, no loop pre-header)
757 75 mem2reg - Number of alloca's promoted
758 1444 cfgsimplify - Number of blocks simplified
759 </pre>
760 </div>
762 <p>Obviously, with so many optimizations, having a unified framework for this
763 stuff is very nice. Making your pass fit well into the framework makes it more
764 maintainable and useful.</p>
766 </div>
768 <!-- ======================================================================= -->
771 </div>
775 <p>Several of the important data structures in LLVM are graphs: for example
779 DAGs</a>. In many cases, while debugging various parts of the compiler, it is
780 nice to instantly visualize these graphs.</p>
782 <p>LLVM provides several callbacks that are available in a debug build to do
784 the current LLVM tool will pop up a window containing the CFG for the function
785 where each basic block is a node in the graph, and each node contains the
786 instructions in the block. Similarly, there also exists
791 up a window. Alternatively, you can sprinkle calls to these functions in your
792 code in places you want to debug.</p>
794 <p>Getting this to work requires a small amount of configuration. On Unix
796 toolkit, and make sure 'dot' and 'gv' are in your path. If you are running on
797 Mac OS/X, download and install the Mac OS/X <a
800 it) to your path. Once in your system and path are set up, rerun the LLVM
801 configure script and rebuild LLVM to enable this functionality.</p>
807 specified color (choices of colors can be found at <a
809 complex node attributes can be provided with <tt>call
812 Attributes</a>.) If you want to restart and clear all the current graph
815 </div>
817 <!-- *********************************************************************** -->
820 </div>
821 <!-- *********************************************************************** -->
826 and we commonly use STL data structures. This section describes the trade-offs
827 you should consider when you pick one.</p>
829 <p>
830 The first step is a choose your own adventure: do you want a sequential
831 container, a set-like container, or a map-like container? The most important
832 thing when choosing a container is the algorithmic properties of how you plan to
833 access the container. Based on that, you should use:</p>
835 <ul>
837 of an value based on another value. Map-like containers also support
838 efficient queries for containment (whether a key is in the map). Map-like
839 containers generally do not support efficient reverse mapping (values to
840 keys). If you need that, use two maps. Some map-like containers also
841 support efficient iteration through the keys in sorted order. Map-like
842 containers are the most expensive sort, only use them if you need one of
843 these capabilities.</li>
846 stuff into a container that automatically eliminates duplicates. Some
847 set-like containers support efficient iteration through the elements in
848 sorted order. Set-like containers are more expensive than sequential
849 containers.
850 </li>
853 the most efficient way to add elements and keeps track of the order they are
854 added to the collection. They permit duplicates and support efficient
855 iteration, but do not support efficient look-up based on a key.
856 </li>
859 container or reference structure that is used for character or byte
860 arrays.</li>
863 perform set operations on sets of numeric id's, while automatically
864 eliminating duplicates. Bit containers require a maximum of 1 bit for each
865 identifier you want to store.
866 </li>
867 </ul>
869 <p>
870 Once the proper category of container is determined, you can fine tune the
871 memory use, constant factors, and cache behaviors of access by intelligently
872 picking a member of the category. Note that constant factors and cache behavior
873 can be a big deal. If you have a vector that usually only contains a few
874 elements (but could contain many), for example, it's much better to use
876 . Doing so avoids (relatively) expensive malloc/free calls, which dwarf the
877 cost of adding the elements to the container. </p>
879 </div>
881 <!-- ======================================================================= -->
884 </div>
887 There are a variety of sequential containers available for you, based on your
888 needs. Pick the first in this section that will do what you want.
889 </div>
891 <!-- _______________________________________________________________________ -->
894 </div>
897 <p>Fixed size arrays are very simple and very fast. They are good if you know
898 exactly how many elements you have, or you have a (low) upper bound on how many
899 you have.</p>
900 </div>
902 <!-- _______________________________________________________________________ -->
905 </div>
908 <p>Heap allocated arrays (new[] + delete[]) are also simple. They are good if
909 the number of elements is variable, if you know how many elements you will need
910 before the array is allocated, and if the array is usually large (if not,
912 allocated array is the cost of the new/delete (aka malloc/free). Also note that
913 if you are allocating an array of a type with a constructor, the constructor and
914 destructors will be run for every element in the array (re-sizable vectors only
915 construct those elements actually used).</p>
916 </div>
918 <!-- _______________________________________________________________________ -->
921 </div>
926 it supports efficient iteration, lays out elements in memory order (so you can
927 do pointer arithmetic between elements), supports efficient push_back/pop_back
928 operations, supports efficient random access to its elements, etc.</p>
930 <p>The advantage of SmallVector is that it allocates space for
932 the SmallVector is dynamically smaller than N, no malloc is performed. This can
933 be a big win in cases where the malloc/free call is far more expensive than the
934 code that fiddles around with the elements.</p>
937 predecessors/successors of a block is usually less than 8). On the other hand,
938 this makes the size of the SmallVector itself large, so you don't want to
939 allocate lots of them (doing so will waste a lot of space). As such,
940 SmallVectors are most useful when on the stack.</p>
942 <p>SmallVector also provides a nice portable and efficient replacement for
945 </div>
947 <!-- _______________________________________________________________________ -->
950 </div>
953 <p>
954 std::vector is well loved and respected. It is useful when SmallVector isn't:
955 when the size of the vector is often large (thus the small optimization will
956 rarely be a benefit) or if you will be allocating many instances of the vector
957 itself (which would waste space for elements that aren't in the container).
958 vector is also useful when interfacing with code that expects vectors :).
959 </p>
964 <pre>
965 for ( ... ) {
967 use V;
968 }
969 </pre>
970 </div>
975 <pre>
977 for ( ... ) {
978 use V;
979 V.clear();
980 }
981 </pre>
982 </div>
984 <p>Doing so will save (at least) one heap allocation and free per iteration of
985 the loop.</p>
987 </div>
989 <!-- _______________________________________________________________________ -->
992 </div>
995 <p>std::deque is, in some senses, a generalized version of std::vector. Like
996 std::vector, it provides constant time random access and other similar
997 properties, but it also provides efficient access to the front of the list. It
998 does not guarantee continuity of elements within memory.</p>
1000 <p>In exchange for this extra flexibility, std::deque has significantly higher
1001 constant factor costs than std::vector. If possible, use std::vector or
1002 something cheaper.</p>
1003 </div>
1005 <!-- _______________________________________________________________________ -->
1008 </div>
1011 <p>std::list is an extremely inefficient class that is rarely useful.
1012 It performs a heap allocation for every element inserted into it, thus having an
1013 extremely high constant factor, particularly for small data types. std::list
1014 also only supports bidirectional iteration, not random access iteration.</p>
1016 <p>In exchange for this high cost, std::list supports efficient access to both
1017 ends of the list (like std::deque, but unlike std::vector or SmallVector). In
1018 addition, the iterator invalidation characteristics of std::list are stronger
1019 than that of a vector class: inserting or removing an element into the list does
1020 not invalidate iterator or pointers to other elements in the list.</p>
1021 </div>
1023 <!-- _______________________________________________________________________ -->
1026 </div>
1030 intrusive, because it requires the element to store and provide access to the
1031 prev/next pointers for the list.</p>
1035 provides some novel characteristics. In particular, it can efficiently store
1036 polymorphic objects, the traits class is informed when an element is inserted or
1038 constant-time splice operation.</p>
1040 <p>These properties are exactly what we want for things like
1044 Related classes of interest are explained in the following subsections:
1045 <ul>
1050 </ul>
1051 </div>
1053 <!-- _______________________________________________________________________ -->
1056 </div>
1061 publicly derive from this traits class.</p>
1062 </div>
1064 <!-- _______________________________________________________________________ -->
1067 </div>
1071 supports a slightly narrower interface. Notably, inserters from
1075 used for a wide variety of customizations.</p>
1076 </div>
1078 <!-- _______________________________________________________________________ -->
1081 </div>
1086 in the default manner.</p>
1091 </div>
1093 <!-- _______________________________________________________________________ -->
1096 </div>
1100 citizen in the C++ ecosystem, it needs to support the standard container
1105 <p>The only sensible solution to this problem is to allocate a so-called
1107 iterator, providing the back-link to the last element. However conforming to the
1109 also must not be dereferenced.</p>
1112 how to allocate and store the sentinel. The corresponding policy is dictated
1114 whenever the need for a sentinel arises.</p>
1116 <p>While the default policy is sufficient in most cases, it may break down when
1119 is wasted. To alleviate the situation with numerous and voluminous
1125 arithmetic is used to obtain the sentinel, which is relative to the
1127 extra pointer, which serves as the back-link of the sentinel. This is the only
1128 field in the ghostly sentinel which can be legally accessed.</p>
1129 </div>
1131 <!-- _______________________________________________________________________ -->
1134 </div>
1139 <p>There are also various STL adapter classes such as std::queue,
1140 std::priority_queue, std::stack, etc. These provide simplified access to an
1141 underlying container but don't affect the cost of the container itself.</p>
1143 </div>
1146 <!-- ======================================================================= -->
1149 </div>
1153 <p>Set-like containers are useful when you need to canonicalize multiple values
1154 into a single representation. There are several different choices for how to do
1155 this, providing various trade-offs.</p>
1157 </div>
1160 <!-- _______________________________________________________________________ -->
1163 </div>
1167 <p>If you intend to insert a lot of elements, then do a lot of queries, a
1168 great approach is to use a vector (or other sequential container) with
1169 std::sort+std::unique to remove duplicates. This approach works really well if
1170 your usage pattern has these two distinct phases (insert then query), and can be
1172 </p>
1174 <p>
1175 This combination provides the several nice properties: the result data is
1176 contiguous in memory (good for cache locality), has few allocations, is easy to
1177 address (iterators in the final vector are just indices or pointers), and can be
1178 efficiently queried with a standard binary or radix search.</p>
1180 </div>
1182 <!-- _______________________________________________________________________ -->
1185 </div>
1189 <p>If you have a set-like data structure that is usually small and whose elements
1191 has space for N elements in place (thus, if the set is dynamically smaller than
1192 N, no malloc traffic is required) and accesses them with a simple linear search.
1193 When the set grows beyond 'N' elements, it allocates a more expensive representation that
1194 guarantees efficient access (for most types, it falls back to std::set, but for
1195 pointers it uses something far better, <a
1198 <p>The magic of this class is that it handles small sets extremely efficiently,
1199 but gracefully handles extremely large sets without loss of efficiency. The
1200 drawback is that the interface is quite small: it supports insertion, queries
1201 and erasing, but does not support iteration.</p>
1203 </div>
1205 <!-- _______________________________________________________________________ -->
1208 </div>
1212 <p>SmallPtrSet has all the advantages of <tt>SmallSet</tt> (and a <tt>SmallSet</tt> of pointers is
1214 more than 'N' insertions are performed, a single quadratically
1215 probed hash table is allocated and grows as needed, providing extremely
1216 efficient access (constant time insertion/deleting/queries with low constant
1217 factors) and is very stingy with malloc traffic.</p>
1220 whenever an insertion occurs. Also, the values visited by the iterators are not
1221 visited in sorted order.</p>
1223 </div>
1225 <!-- _______________________________________________________________________ -->
1228 </div>
1232 <p>
1233 DenseSet is a simple quadratically probed hash table. It excels at supporting
1234 small values: it uses a single allocation to hold all of the pairs that
1235 are currently inserted in the set. DenseSet is a great way to unique small
1236 values that are not simple pointers (use <a
1238 the same requirements for the value type that <a
1240 </p>
1242 </div>
1244 <!-- _______________________________________________________________________ -->
1247 </div>
1251 <p>
1252 FoldingSet is an aggregate class that is really good at uniquing
1253 expensive-to-create or polymorphic objects. It is a combination of a chained
1254 hash table with intrusive links (uniqued objects are required to inherit from
1256 its ID process.</p>
1259 a complex object (for example, a node in the code generator). The client has a
1260 description of *what* it wants to generate (it knows the opcode and all the
1261 operands), but we don't want to 'new' a node, then try inserting it into a set
1262 only to find out it already exists, at which point we would have to delete it
1263 and return the node that already exists.
1264 </p>
1266 <p>To support this style of client, FoldingSet perform a query with a
1267 FoldingSetNodeID (which wraps SmallVector) that can be used to describe the
1268 element that we want to query for. The query either returns the element
1269 matching the ID or it returns an opaque ID that indicates where insertion should
1270 take place. Construction of the ID usually does not require heap traffic.</p>
1272 <p>Because FoldingSet uses intrusive links, it can support polymorphic objects
1273 in the set (for example, you can have SDNode instances mixed with LoadSDNodes).
1274 Because the elements are individually allocated, pointers to the elements are
1275 stable: inserting or removing elements does not invalidate any pointers to other
1276 elements.
1277 </p>
1279 </div>
1281 <!-- _______________________________________________________________________ -->
1284 </div>
1289 many things but great at nothing. std::set allocates memory for each element
1290 inserted (thus it is very malloc intensive) and typically stores three pointers
1291 per element in the set (thus adding a large amount of per-element space
1292 overhead). It offers guaranteed log(n) performance, which is not particularly
1293 fast from a complexity standpoint (particularly if the elements of the set are
1294 expensive to compare, like strings), and has extremely high constant factors for
1295 lookup, insertion and removal.</p>
1297 <p>The advantages of std::set are that its iterators are stable (deleting or
1298 inserting an element from the set does not affect iterators or pointers to other
1299 elements) and that iteration over the set is guaranteed to be in sorted order.
1300 If the elements in the set are large, then the relative overhead of the pointers
1301 and malloc traffic is not a big deal, but if the elements of the set are small,
1302 std::set is almost never a good choice.</p>
1304 </div>
1306 <!-- _______________________________________________________________________ -->
1309 </div>
1314 Container</a>. The important property
1315 that this provides is efficient insertion with uniquing (duplicate elements are
1316 ignored) with iteration support. It implements this by inserting elements into
1317 both a set-like container and the sequential container, using the set-like
1318 container for uniquing and the sequential container for iteration.
1319 </p>
1321 <p>The difference between SetVector and other sets is that the order of
1322 iteration is guaranteed to match the order of insertion into the SetVector.
1323 This property is really important for things like sets of pointers. Because
1324 pointer values are non-deterministic (e.g. vary across runs of the program on
1325 different machines), iterating over the pointers in the set will
1326 not be in a well-defined order.</p>
1328 <p>
1329 The drawback of SetVector is that it requires twice as much space as a normal
1330 set and has the sum of constant factors from the set-like container and the
1331 sequential container that it uses. Use it *only* if you need to iterate over
1332 the elements in a deterministic order. SetVector is also expensive to delete
1333 elements out of (linear time), unless you use it's "pop_back" method, which is
1334 faster.
1335 </p>
1337 <p>SetVector is an adapter class that defaults to using std::vector and std::set
1338 for the underlying containers, so it is quite expensive. However,
1340 defaults to using a SmallVector and SmallSet of a specified size. If you use
1341 this, and if your sets are dynamically smaller than N, you will save a lot of
1342 heap traffic.</p>
1344 </div>
1346 <!-- _______________________________________________________________________ -->
1349 </div>
1353 <p>
1355 retains a unique ID for each element inserted into the set. It internally
1356 contains a map and a vector, and it assigns a unique ID for each value inserted
1357 into the set.</p>
1359 <p>UniqueVector is very expensive: its cost is the sum of the cost of
1360 maintaining both the map and vector, it has high complexity, high constant
1361 factors, and produces a lot of malloc traffic. It should be avoided.</p>
1363 </div>
1366 <!-- _______________________________________________________________________ -->
1369 </div>
1373 <p>
1374 The STL provides several other options, such as std::multiset and the various
1375 "hash_set" like containers (whether from C++ TR1 or from the SGI library). We
1376 never use hash_set and unordered_set because they are generally very expensive
1377 (each insertion requires a malloc) and very non-portable.
1378 </p>
1380 <p>std::multiset is useful if you're not interested in elimination of
1381 duplicates, but has all the drawbacks of std::set. A sorted vector (where you
1382 don't delete duplicate entries) or some other approach is almost always
1383 better.</p>
1385 </div>
1387 <!-- ======================================================================= -->
1390 </div>
1393 Map-like containers are useful when you want to associate data to a key. As
1394 usual, there are a lot of different ways to do this. :)
1395 </div>
1397 <!-- _______________________________________________________________________ -->
1400 </div>
1404 <p>
1405 If your usage pattern follows a strict insert-then-query approach, you can
1407 for set-like containers</a>. The only difference is that your query function
1408 (which uses std::lower_bound to get efficient log(n) lookup) should only compare
1409 the key, not both the key and value. This yields the same advantages as sorted
1410 vectors for sets.
1411 </p>
1412 </div>
1414 <!-- _______________________________________________________________________ -->
1417 </div>
1421 <p>
1422 Strings are commonly used as keys in maps, and they are difficult to support
1423 efficiently: they are variable length, inefficient to hash and compare when
1424 long, expensive to copy, etc. StringMap is a specialized container designed to
1425 cope with these issues. It supports mapping an arbitrary range of bytes to an
1426 arbitrary other object.</p>
1428 <p>The StringMap implementation uses a quadratically-probed hash table, where
1429 the buckets store a pointer to the heap allocated entries (and some other
1430 stuff). The entries in the map must be heap allocated because the strings are
1431 variable length. The string data (key) and the element object (value) are
1432 stored in the same allocation with the string data immediately after the element
1433 object. This container guarantees the "<tt>(char*)(&Value+1)</tt>" points
1434 to the key string for a value.</p>
1436 <p>The StringMap is very fast for several reasons: quadratic probing is very
1437 cache efficient for lookups, the hash value of strings in buckets is not
1438 recomputed when looking up an element, StringMap rarely has to touch the
1439 memory for unrelated objects when looking up a value (even when hash collisions
1440 happen), hash table growth does not recompute the hash values for strings
1441 already in the table, and each pair in the map is store in a single allocation
1442 (the string data is stored in the same allocation as the Value of a pair).</p>
1444 <p>StringMap also provides query methods that take byte ranges, so it only ever
1445 copies a string if a value is inserted into the table.</p>
1446 </div>
1448 <!-- _______________________________________________________________________ -->
1451 </div>
1454 <p>
1455 IndexedMap is a specialized container for mapping small dense integers (or
1456 values that can be mapped to small dense integers) to some other type. It is
1457 internally implemented as a vector with a mapping function that maps the keys to
1458 the dense integer range.
1459 </p>
1461 <p>
1462 This is useful for cases like virtual registers in the LLVM code generator: they
1463 have a dense mapping that is offset by a compile-time constant (the first
1464 virtual register ID).</p>
1466 </div>
1468 <!-- _______________________________________________________________________ -->
1471 </div>
1475 <p>
1476 DenseMap is a simple quadratically probed hash table. It excels at supporting
1477 small keys and values: it uses a single allocation to hold all of the pairs that
1478 are currently inserted in the map. DenseMap is a great way to map pointers to
1479 pointers, or map other small types to each other.
1480 </p>
1482 <p>
1483 There are several aspects of DenseMap that you should be aware of, however. The
1484 iterators in a densemap are invalidated whenever an insertion occurs, unlike
1485 map. Also, because DenseMap allocates space for a large number of key/value
1486 pairs (it starts with 64 by default), it will waste a lot of space if your keys
1487 or values are large. Finally, you must implement a partial specialization of
1488 DenseMapInfo for the key that you want, if it isn't already supported. This
1489 is required to tell DenseMap about two special marker values (which can never be
1490 inserted into the map) that it needs internally.</p>
1492 </div>
1494 <!-- _______________________________________________________________________ -->
1497 </div>
1501 <p>
1503 Value*s (or subclasses) to another type. When a Value is deleted or RAUW'ed,
1504 ValueMap will update itself so the new version of the key is mapped to the same
1505 value, just as if the key were a WeakVH. You can configure exactly how this
1506 happens, and what else happens on these two events, by passing
1509 </div>
1511 <!-- _______________________________________________________________________ -->
1514 </div>
1518 <p> IntervalMap is a compact map for small keys and values. It maps key
1519 intervals instead of single keys, and it will automatically coalesce adjacent
1520 intervals. When then map only contains a few intervals, they are stored in the
1521 map object itself to avoid allocations.</p>
1523 <p> The IntervalMap iterators are quite big, so they should not be passed around
1524 as STL iterators. The heavyweight iterators allow a smaller data structure.</p>
1526 </div>
1528 <!-- _______________________________________________________________________ -->
1531 </div>
1535 <p>
1537 a single allocation per pair inserted into the map, it offers log(n) lookup with
1538 an extremely large constant factor, imposes a space penalty of 3 pointers per
1539 pair in the map, etc.</p>
1541 <p>std::map is most useful when your keys or values are very large, if you need
1542 to iterate over the collection in sorted order, or if you need stable iterators
1543 into the map (i.e. they don't get invalidated if an insertion or deletion of
1544 another element takes place).</p>
1546 </div>
1548 <!-- _______________________________________________________________________ -->
1551 </div>
1555 <p>
1556 The STL provides several other options, such as std::multimap and the various
1557 "hash_map" like containers (whether from C++ TR1 or from the SGI library). We
1558 never use hash_set and unordered_set because they are generally very expensive
1559 (each insertion requires a malloc) and very non-portable.</p>
1561 <p>std::multimap is useful if you want to map a key to multiple values, but has
1562 all the drawbacks of std::map. A sorted vector or some other approach is almost
1563 always better.</p>
1565 </div>
1567 <!-- ======================================================================= -->
1570 </div>
1574 <p>
1575 TODO: const char* vs stringref vs smallstring vs std::string. Describe twine,
1576 xref to #string_apis.
1577 </p>
1579 </div>
1581 <!-- ======================================================================= -->
1584 </div>
1587 <p>Unlike the other containers, there are only two bit storage containers, and
1588 choosing when to use each is relatively straightforward.</p>
1590 <p>One additional option is
1592 implementation in many common compilers (e.g. commonly available versions of
1593 GCC) is extremely inefficient and 2) the C++ standards committee is likely to
1594 deprecate this container and/or change it significantly somehow. In any case,
1595 please don't use it.</p>
1596 </div>
1598 <!-- _______________________________________________________________________ -->
1601 </div>
1604 <p> The BitVector container provides a dynamic size set of bits for manipulation.
1605 It supports individual bit setting/testing, as well as set operations. The set
1606 operations take time O(size of bitvector), but operations are performed one word
1607 at a time, instead of one bit at a time. This makes the BitVector very fast for
1608 set operations compared to other containers. Use the BitVector when you expect
1609 the number of set bits to be high (IE a dense set).
1610 </p>
1611 </div>
1613 <!-- _______________________________________________________________________ -->
1616 </div>
1619 <p> The SmallBitVector container provides the same interface as BitVector, but
1620 it is optimized for the case where only a small number of bits, less than
1621 25 or so, are needed. It also transparently supports larger bit counts, but
1622 slightly less efficiently than a plain BitVector, so SmallBitVector should
1623 only be used when larger counts are rare.
1624 </p>
1626 <p>
1627 At this time, SmallBitVector does not support set operations (and, or, xor),
1628 and its operator[] does not provide an assignable lvalue.
1629 </p>
1630 </div>
1632 <!-- _______________________________________________________________________ -->
1635 </div>
1638 <p> The SparseBitVector container is much like BitVector, with one major
1639 difference: Only the bits that are set, are stored. This makes the
1640 SparseBitVector much more space efficient than BitVector when the set is sparse,
1641 as well as making set operations O(number of set bits) instead of O(size of
1642 universe). The downside to the SparseBitVector is that setting and testing of random bits is O(N), and on large SparseBitVectors, this can be slower than BitVector. In our implementation, setting or testing bits in sorted order
1643 (either forwards or reverse) is O(1) worst case. Testing and setting bits within 128 bits (depends on size) of the current bit is also O(1). As a general statement, testing/setting bits in a SparseBitVector is O(distance away from last set bit).
1644 </p>
1645 </div>
1647 <!-- *********************************************************************** -->
1650 </div>
1651 <!-- *********************************************************************** -->
1655 <p>This section describes how to perform some very simple transformations of
1656 LLVM code. This is meant to give examples of common idioms used, showing the
1658 you should also read about the main classes that you will be working with. The
1660 and descriptions of the main classes that you should know about.</p>
1662 </div>
1664 <!-- NOTE: this section should be heavy on example code -->
1665 <!-- ======================================================================= -->
1668 </div>
1672 <p>The LLVM compiler infrastructure have many different data structures that may
1673 be traversed. Following the example of the C++ standard template library, the
1674 techniques used to traverse these various data structures are all basically the
1677 function returns an iterator pointing to one past the last valid element of the
1679 between the two operations.</p>
1681 <p>Because the pattern for iteration is common across many different aspects of
1682 the program representation, the standard template library algorithms may be used
1683 on them, and it is easier to remember how to iterate. First we show a few common
1684 examples of the data structures that need to be traversed. Other data
1685 structures are traversed in very similar ways.</p>
1687 </div>
1689 <!-- _______________________________________________________________________ -->
1694 </div>
1699 transform in some way; in particular, you'd like to manipulate its
1706 <pre>
1713 </pre>
1714 </div>
1716 <p>Note that i can be used as if it were a pointer for the purposes of
1718 because the indirection operator is overloaded for the iterator
1722 </div>
1724 <!-- _______________________________________________________________________ -->
1729 </div>
1734 easy to iterate over the individual instructions that make up
1739 <pre>
1745 </pre>
1746 </div>
1748 <p>However, this isn't really the best way to print out the contents of a
1750 anything you'll care about, you could have just invoked the print routine on the
1753 </div>
1755 <!-- _______________________________________________________________________ -->
1760 </div>
1769 small example that shows how to dump all instructions in a function to the standard error stream:<p>
1772 <pre>
1776 for (inst_iterator I = inst_begin(F), E = inst_end(F); I != E; ++I)
1778 </pre>
1779 </div>
1782 work list with its initial contents. For example, if you wanted to
1784 F, all you would need to do is something like:</p>
1787 <pre>
1791 for (inst_iterator I = inst_begin(F), E = inst_end(F); I != E; ++I)
1792 worklist.insert(&*I);
1793 </pre>
1794 </div>
1799 </div>
1801 <!-- _______________________________________________________________________ -->
1804 vice-versa)</a>
1805 </div>
1809 <p>Sometimes, it'll be useful to grab a reference (or pointer) to a class
1810 instance when all you've got at hand is an iterator. Well, extracting
1811 a reference or a pointer from an iterator is very straight-forward.
1816 <pre>
1819 const Instruction& inst = *j;
1820 </pre>
1821 </div>
1823 <p>However, the iterators you'll be working with in the LLVM framework are
1824 special: they will automatically convert to a ptr-to-instance type whenever they
1825 need to. Instead of dereferencing the iterator and then taking the address of
1826 the result, you can simply assign the iterator to the proper pointer type and
1827 you get the dereference and address-of operation as a result of the assignment
1828 (behind the scenes, this is a result of overloading casting mechanisms). Thus
1829 the last line of the last example,</p>
1832 <pre>
1833 Instruction *pinst = &*i;
1834 </pre>
1835 </div>
1840 <pre>
1841 Instruction *pinst = i;
1842 </pre>
1843 </div>
1845 <p>It's also possible to turn a class pointer into the corresponding iterator,
1846 and this is a constant time operation (very efficient). The following code
1847 snippet illustrates use of the conversion constructors provided by LLVM
1848 iterators. By using these, you can explicitly grab the iterator of something
1849 without actually obtaining it via iteration over some structure:</p>
1852 <pre>
1853 void printNextInstruction(Instruction* inst) {
1854 BasicBlock::iterator it(inst);
1857 }
1858 </pre>
1859 </div>
1861 <p>Unfortunately, these implicit conversions come at a cost; they prevent
1862 these iterators from conforming to standard iterator conventions, and thus
1863 from being usable with standard algorithms and containers. For example, they
1865 from compiling:</p>
1868 <pre>
1870 </pre>
1871 </div>
1873 <p>Because of this, these implicit conversions may be removed some day,
1876 </div>
1878 <!--_______________________________________________________________________-->
1881 example</a>
1882 </div>
1886 <p>Say that you're writing a FunctionPass and would like to count all the
1890 much more straight-forward manner, but this example will allow us to explore how
1892 is what we want to do:</p>
1895 <pre>
1896 initialize callCounter to zero
1897 for each Function f in the Module
1898 for each BasicBlock b in f
1899 for each Instruction i in b
1900 if (i is a CallInst and calls the given function)
1901 increment callCounter
1902 </pre>
1903 </div>
1905 <p>And the actual code is (remember, because we're writing a
1910 <pre>
1911 Function* targetFunc = ...;
1913 class OurFunctionPass : public FunctionPass {
1914 public:
1915 OurFunctionPass(): callCounter(0) { }
1917 virtual runOnFunction(Function& F) {
1918 for (Function::iterator b = F.begin(), be = F.end(); b != be; ++b) {
1925 if (callInst->getCalledFunction() == targetFunc)
1926 ++callCounter;
1927 }
1928 }
1929 }
1930 }
1932 private:
1933 unsigned callCounter;
1934 };
1935 </pre>
1936 </div>
1938 </div>
1940 <!--_______________________________________________________________________-->
1943 </div>
1947 <p>You may have noticed that the previous example was a bit oversimplified in
1948 that it did not deal with call sites generated by 'invoke' instructions. In
1949 this, and in other situations, you may find that you want to treat
1952 less closely-related things. For these cases, LLVM provides a handy wrapper
1953 class called <a
1960 reference and it should not be dynamically allocated or deallocated using
1962 assignable and constructable, with costs equivalents to that of a bare pointer.
1963 If you look at its definition, it has only a single pointer member.</p>
1965 </div>
1967 <!--_______________________________________________________________________-->
1970 </div>
1984 <pre>
1985 Function *F = ...;
1991 }
1992 </pre>
1993 </div>
1997 doing it only once in the loop body and reusing its result.</p>
2004 all of the values that a particular instruction uses (that is, the operands of
2008 <pre>
2009 Instruction *pi = ...;
2012 Value *v = *i;
2014 }
2015 </pre>
2016 </div>
2019 mutation free algorithms (such as analyses, etc.). For this purpose above
2026 </div>
2028 <!--_______________________________________________________________________-->
2031 successors of blocks</a>
2032 </div>
2036 <p>Iterating over the predecessors and successors of a block is quite easy
2038 this to iterate over all predecessors of BB:</p>
2041 <pre>
2042 #include "llvm/Support/CFG.h"
2043 BasicBlock *BB = ...;
2045 for (pred_iterator PI = pred_begin(BB), E = pred_end(BB); PI != E; ++PI) {
2046 BasicBlock *Pred = *PI;
2048 }
2049 </pre>
2050 </div>
2052 <p>Similarly, to iterate over successors use
2053 succ_iterator/succ_begin/succ_end.</p>
2055 </div>
2058 <!-- ======================================================================= -->
2061 </div>
2065 <p>There are some primitive transformation operations present in the LLVM
2066 infrastructure that are worth knowing about. When performing
2067 transformations, it's fairly common to manipulate the contents of basic
2068 blocks. This section describes some of the common methods for doing so
2069 and gives example code.</p>
2071 </div>
2073 <!--_______________________________________________________________________-->
2077 </div>
2084 constructor for the kind of instruction to instantiate and provide the necessary
2089 <pre>
2090 AllocaInst* ai = new AllocaInst(Type::Int32Ty);
2091 </pre>
2092 </div>
2096 subclass is likely to have varying default parameters which change the semantics
2097 of the instruction, so refer to the <a
2103 <p>It is very useful to name the values of instructions when you're able to, as
2104 this facilitates the debugging of your transformations. If you end up looking
2105 at generated LLVM machine code, you definitely want to have logical names
2106 associated with the results of instructions! By supplying a value for the
2108 associate a logical name with the result of the instruction's execution at
2109 run time. For example, say that I'm writing a transformation that dynamically
2110 allocates space for an integer on the stack, and that integer is going to be
2111 used as some kind of index by some other code. To accomplish this, I place an
2117 <pre>
2119 </pre>
2120 </div>
2123 execution value, which is a pointer to an integer on the run time stack.</p>
2130 <ul>
2131 <li>Insertion into an explicit instruction list
2138 <pre>
2139 BasicBlock *pb = ...;
2140 Instruction *pi = ...;
2141 Instruction *newInst = new Instruction(...);
2144 </pre>
2145 </div>
2149 classes provide constructors which take a pointer to a
2151 looked like: </p>
2154 <pre>
2155 BasicBlock *pb = ...;
2156 Instruction *newInst = new Instruction(...);
2159 </pre>
2160 </div>
2165 <pre>
2166 BasicBlock *pb = ...;
2167 Instruction *newInst = new Instruction(..., pb);
2168 </pre>
2169 </div>
2171 <p>which is much cleaner, especially if you are creating
2172 long instruction streams.</p></li>
2174 <li>Insertion into an implicit instruction list
2177 are implicitly associated with an existing instruction list: the instruction
2178 list of the enclosing basic block. Thus, we could have accomplished the same
2180 </p>
2183 <pre>
2184 Instruction *pi = ...;
2185 Instruction *newInst = new Instruction(...);
2188 </pre>
2189 </div>
2191 <p>In fact, this sequence of steps occurs so frequently that the
2193 constructors which take (as a default parameter) a pointer to an
2197 provided instruction, immediately before that instruction. Using an
2199 parameter, the above code becomes:</p>
2202 <pre>
2203 Instruction* pi = ...;
2204 Instruction* newInst = new Instruction(..., pi);
2205 </pre>
2206 </div>
2208 <p>which is much cleaner, especially if you're creating a lot of
2210 </ul>
2212 </div>
2214 <!--_______________________________________________________________________-->
2217 </div>
2221 <p>Deleting an instruction from an existing sequence of instructions that form a
2223 you must have a pointer to the instruction that you wish to delete. Second, you
2224 need to obtain the pointer to that instruction's basic block. You use the
2225 pointer to the basic block to get its list of instructions and then use the
2226 erase function to remove your instruction. For example:</p>
2229 <pre>
2231 I->eraseFromParent();
2232 </pre>
2233 </div>
2235 </div>
2237 <!--_______________________________________________________________________-->
2241 </div>
2247 <p>Including "<a href="/doxygen/BasicBlockUtils_8h-source.html">llvm/Transforms/Utils/BasicBlockUtils.h</a>"
2253 <ul>
2256 <p>This function replaces all uses of a given instruction with a value,
2257 and then removes the original instruction. The following example
2258 illustrates the replacement of the result of a particular
2260 pointer to an integer.</p>
2263 <pre>
2264 AllocaInst* instToReplace = ...;
2265 BasicBlock::iterator ii(instToReplace);
2268 Constant::getNullValue(PointerType::getUnqual(Type::Int32Ty)));
2269 </pre></div></li>
2273 <p>This function replaces a particular instruction with another
2274 instruction, inserting the new instruction into the basic block at the
2275 location where the old instruction was, and replacing any uses of the old
2276 instruction with the new instruction. The following example illustrates
2280 <pre>
2281 AllocaInst* instToReplace = ...;
2282 BasicBlock::iterator ii(instToReplace);
2286 </pre></div></li>
2287 </ul>
2295 information.</p>
2297 <!-- Value::replaceAllUsesWith User::replaceUsesOfWith Point out:
2298 include/llvm/Transforms/Utils/ especially BasicBlockUtils.h with:
2299 ReplaceInstWithValue, ReplaceInstWithInst -->
2301 </div>
2303 <!--_______________________________________________________________________-->
2306 </div>
2310 <p>Deleting a global variable from a module is just as easy as deleting an
2311 Instruction. First, you must have a pointer to the global variable that you wish
2312 to delete. You use this pointer to erase it from its parent, the module.
2313 For example:</p>
2316 <pre>
2319 GV->eraseFromParent();
2320 </pre>
2321 </div>
2323 </div>
2325 <!-- ======================================================================= -->
2328 </div>
2332 <p>In generating IR, you may need some complex types. If you know these types
2335 has two forms depending on whether you're building types for cross-compilation
2338 out of types from
2340 namespace and pointers, functions, arrays, etc. built of
2342 whose size may depend on the host compiler. For example,</p>
2345 <pre>
2347 </pre>
2348 </div>
2353 <pre>
2355 params.push_back(PointerType::getUnqual(Type::Int32Ty));
2356 FunctionType *ft = FunctionType::get(Type::Int8Ty, params, false);
2357 </pre>
2358 </div>
2363 </div>
2365 <!-- *********************************************************************** -->
2368 </div>
2369 <!-- *********************************************************************** -->
2372 <p>
2373 This section describes the interaction of the LLVM APIs with multithreading,
2374 both on the part of client applications, and in the JIT, in the hosted
2375 application.
2376 </p>
2378 <p>
2379 Note that LLVM's support for multithreading is still relatively young. Up
2380 through version 2.5, the execution of threaded hosted applications was
2381 supported, but not threaded client access to the APIs. While this use case is
2383 ensure proper operation in multithreaded mode.
2384 </p>
2386 <p>
2387 Note that, on Unix-like platforms, LLVM requires the presence of GCC's atomic
2388 intrinsics in order to support threaded operation. If you need a
2389 multhreading-capable LLVM on a platform without a suitably modern system
2390 compiler, consider compiling LLVM and LLVM-GCC in single-threaded mode, and
2391 using the resultant compiler to build a copy of LLVM with multithreading
2392 support.
2393 </p>
2394 </div>
2396 <!-- ======================================================================= -->
2399 </div>
2403 <p>
2404 In order to properly protect its internal data structures while avoiding
2405 excessive locking overhead in the single-threaded case, the LLVM must intialize
2406 certain data structures necessary to provide guards around its internals. To do
2408 making any concurrent LLVM API calls. To subsequently tear down these
2411 mode.
2412 </p>
2414 <p>
2416 say that no other LLVM API calls may be executing at any time during the
2418 </tt>. It's is the client's responsibility to enforce this isolation.
2419 </p>
2421 <p>
2423 failure of the initialization. Failure typically indicates that your copy of
2424 LLVM was built without multithreading support, typically because GCC atomic
2425 intrinsics were not found in your system compiler. In this case, the LLVM API
2428 must be taken</a> to ensure that side exits and the like do not accidentally
2429 result in concurrent LLVM API calls.
2430 </p>
2431 </div>
2433 <!-- ======================================================================= -->
2436 </div>
2439 <p>
2441 to deallocate memory used for internal structures. This will also invoke
2445 </p>
2447 <p>
2448 Note that, if you use scope-based shutdown, you can use the
2450 destructor.
2451 </div>
2453 <!-- ======================================================================= -->
2456 </div>
2459 <p>
2461 initialization of static resources, such as the global type tables. Before the
2464 however, it uses double-checked locking to implement thread-safe lazy
2465 initialization.
2466 </p>
2468 <p>
2469 Note that, because no other threads are allowed to issue LLVM API calls before
2472 </p>
2474 <p>
2476 APIs provide access to the global lock used to implement the double-checked
2477 locking for lazy initialization. These should only be used internally to LLVM,
2478 and only if you know what you're doing!
2479 </p>
2480 </div>
2482 <!-- ======================================================================= -->
2485 </div>
2488 <p>
2490 to operate multiple, isolated instances of LLVM concurrently within the same
2491 address space. For instance, in a hypothetical compile-server, the compilation
2492 of an individual translation unit is conceptually independent from all the
2493 others, and it would be desirable to be able to compile incoming translation
2494 units concurrently on independent server threads. Fortunately,
2496 </p>
2498 <p>
2505 safe to compile on multiple threads simultaneously, as long as no two threads
2506 operate on entities within the same context.
2507 </p>
2509 <p>
2510 In practice, very few places in the API require the explicit specification of a
2513 other entities can determine what context they belong to by looking at their
2515 maintain this interface design.
2516 </p>
2518 <p>
2522 isolation is not a concern.
2523 </p>
2524 </div>
2526 <!-- ======================================================================= -->
2529 </div>
2532 <p>
2533 LLVM's "eager" JIT compiler is safe to use in threaded programs. Multiple
2536 run code output by the JIT concurrently. The user must still ensure that only
2538 might be modifying it. One way to do that is to always hold the JIT lock while
2542 </p>
2544 <p>When the JIT is configured to compile lazily (using
2547 updating call sites after a function is lazily-jitted. It's still possible to
2548 use the lazy JIT in a threaded program if you ensure that only one thread at a
2549 time can call any particular lazy stub and that the JIT lock guards any IR
2550 access, but we suggest using only the eager JIT in threaded programs.
2551 </p>
2552 </div>
2554 <!-- *********************************************************************** -->
2557 </div>
2558 <!-- *********************************************************************** -->
2561 <p>
2562 This section describes some of the advanced or obscure API's that most clients
2563 do not need to be aware of. These API's tend manage the inner workings of the
2564 LLVM system, and only need to be accessed in unusual circumstances.
2565 </p>
2566 </div>
2568 <!-- ======================================================================= -->
2571 </div>
2575 <p>
2576 The LLVM type system has a very simple goal: allow clients to compare types for
2577 structural equality with a simple pointer comparison (aka a shallow compare).
2578 This goal makes clients much simpler and faster, and is used throughout the LLVM
2579 system.
2580 </p>
2582 <p>
2583 Unfortunately achieving this goal is not a simple matter. In particular,
2584 recursive types and late resolution of opaque types makes the situation very
2585 difficult to handle. Fortunately, for the most part, our implementation makes
2586 most clients able to be completely unaware of the nasty internal details. The
2587 primary case where clients are exposed to the inner workings of it are when
2588 building a recursive type. In addition to this case, the LLVM bitcode reader,
2589 assembly parser, and linker also have to be aware of the inner workings of this
2590 system.
2591 </p>
2593 <p>
2594 For our purposes below, we need three concepts. First, an "Opaque Type" is
2597 opaque type as part of its type graph (for example "<tt>{ opaque, i32 }</tt>").
2598 Third, a concrete type is a type that is not an abstract type (e.g. "<tt>{ i32,
2599 float }</tt>").
2600 </p>
2602 </div>
2604 <!-- ______________________________________________________________________ -->
2607 </div>
2611 <p>
2612 Because the most common question is "how do I build a recursive type with LLVM",
2613 we answer it now and explain it as we go. Here we include enough to cause this
2614 to be emitted to an output .ll file:
2615 </p>
2618 <pre>
2619 %mylist = type { %mylist*, i32 }
2620 </pre>
2621 </div>
2623 <p>
2624 To build this, use the following LLVM APIs:
2625 </p>
2628 <pre>
2632 Elts.push_back(PointerType::getUnqual(StructTy));
2633 Elts.push_back(Type::Int32Ty);
2634 StructType *NewSTy = StructType::get(Elts);
2638 cast<OpaqueType>(StructTy.get())-><a href="#refineAbstractTypeTo">refineAbstractTypeTo</a>(NewSTy);
2640 // <i>NewSTy is potentially invalidated, but StructTy (a <a href="#PATypeHolder">PATypeHolder</a>) is</i>
2646 </pre>
2647 </div>
2649 <p>
2650 This code shows the basic approach used to build recursive types: build a
2651 non-recursive type using 'opaque', then use type unification to close the cycle.
2652 The type unification step is performed by the <tt><a
2654 described next. After that, we describe the <a
2656 </p>
2658 </div>
2660 <!-- ______________________________________________________________________ -->
2663 </div>
2666 <p>
2668 While this method is actually a member of the DerivedType class, it is most
2669 often used on OpaqueType instances. Type unification is actually a recursive
2670 process. After unification, types can become structurally isomorphic to
2671 existing types, and all duplicates are deleted (to preserve pointer equality).
2672 </p>
2674 <p>
2675 In the example above, the OpaqueType object is definitely deleted.
2676 Additionally, if there is an "{ \2*, i32}" type already created in the system,
2678 a type is deleted, any "Type*" pointers in the program are invalidated. As
2681 types can never move or be deleted). To deal with this, the <a
2683 reference to a possibly refined type, and the <a
2685 complex datastructures.
2686 </p>
2688 </div>
2690 <!-- ______________________________________________________________________ -->
2693 </div>
2696 <p>
2697 PATypeHolder is a form of a "smart pointer" for Type objects. When VMCore
2698 happily goes about nuking types that become isomorphic to existing types, it
2699 automatically updates all PATypeHolder objects to point to the new type. In the
2700 example above, this allows the code to maintain a pointer to the resultant
2701 resolved recursive type, even though the Type*'s are potentially invalidated.
2702 </p>
2704 <p>
2705 PATypeHolder is an extremely light-weight object that uses a lazy union-find
2706 implementation to update pointers. For example the pointer from a Value to its
2707 Type is maintained by PATypeHolder objects.
2708 </p>
2710 </div>
2712 <!-- ______________________________________________________________________ -->
2715 </div>
2719 <p>
2720 Some data structures need more to perform more complex updates when types get
2721 resolved. To support this, a class can derive from the AbstractTypeUser class.
2722 This class
2723 allows it to get callbacks when certain types are resolved. To register to get
2724 callbacks for a particular type, the DerivedType::{add/remove}AbstractTypeUser
2725 methods can be called on a type. Note that these methods only work for <i>
2726 abstract</i> types. Concrete types (those that do not include any opaque
2727 objects) can never be refined.
2728 </p>
2729 </div>
2732 <!-- ======================================================================= -->
2736 </div>
2746 names for types.</p>
2749 by most clients. It should only be used when iteration over the symbol table
2750 names themselves are required, which is very special purpose. Note that not
2751 all LLVM
2753 an empty name) do not exist in the symbol table.
2754 </p>
2756 <p>These symbol tables support iteration over the values/types in the symbol
2761 appropriate symbol table. For types, use the Module::addTypeName method to
2762 insert entries into the symbol table.</p>
2764 </div>
2768 <!-- ======================================================================= -->
2771 </div>
2779 addition and removal.</p>
2781 <!-- ______________________________________________________________________ -->
2783 <a name="Use2User">Interaction and relationship between <tt>User</tt> and <tt>Use</tt> objects</a>
2784 </div>
2787 <p>
2789 or refer to them out-of-line by means of a pointer. A mixed variant
2792 </p>
2793 </div>
2795 <p>
2797 <ul>
2798 <li><p>Layout a)
2800 object and there are a fixed number of them.</p>
2802 <li><p>Layout b)
2805 number of them.</p>
2806 </ul>
2807 <p>
2808 As of v2.4 each layout still possesses a direct pointer to the
2810 we stick to this redundancy for the sake of simplicity.
2812 has. (Theoretically this information can also be calculated
2813 given the scheme presented below.)</p>
2814 <p>
2816 enforce the following memory layouts:</p>
2818 <ul>
2819 <li><p>Layout a) is modelled by prepending the <tt>User</tt> object by the <tt>Use[]</tt> array.</p>
2821 <pre>
2822 ...---.---.---.---.-------...
2823 | P | P | P | P | User
2824 '''---'---'---'---'-------'''
2825 </pre>
2828 <pre>
2829 .-------...
2830 | User
2831 '-------'''
2832 |
2833 v
2834 .---.---.---.---...
2835 | P | P | P | P |
2836 '---'---'---'---'''
2837 </pre>
2838 </ul>
2842 <!-- ______________________________________________________________________ -->
2845 </div>
2848 <p>
2851 recover it. This is accomplished by the following scheme:</p>
2852 </div>
2854 A bit-encoding in the 2 LSBits (least significant bits) of the <tt>Use::Prev</tt> allows to find the
2856 <ul>
2861 </ul>
2862 <p>
2865 we have to walk to the next stop picking up digits
2866 and calculating the offset:</p>
2867 <pre>
2868 .---.---.---.---.---.---.---.---.---.---.---.---.---.---.---.---.----------------
2870 '---'---'---'---'---'---'---'---'---'---'---'---'---'---'---'---'----------------
2872 | | | | |__>
2873 | | | |__________>
2874 | | |______________________>
2875 | |______________________________________>
2876 |__________________________________________________________>
2877 </pre>
2878 <p>
2879 Only the significant number of bits need to be stored between the
2883 <!-- ______________________________________________________________________ -->
2886 </div>
2889 <p>
2890 The following literate Haskell fragment demonstrates the concept:</p>
2891 </div>
2894 <pre>
2895 > import Test.QuickCheck
2896 >
2901 >
2907 >
2908 > takeLast n ss = reverse $ take n $ reverse ss
2909 >
2911 >
2912 </pre>
2913 </div>
2914 <p>
2916 <p>
2917 The reverse algorithm computes the length of the string just by examining
2918 a certain prefix:</p>
2921 <pre>
2926 >
2929 > decode walk acc _ = walk + acc
2930 >
2931 </pre>
2932 </div>
2933 <p>
2935 <p>
2939 <pre>
2941 > testcaseLength = length testcase
2942 >
2944 > where arr = takeLast n testcase
2945 >
2946 </pre>
2947 </div>
2948 <p>
2951 <pre>
2952 *Main> quickCheck identityProp
2953 OK, passed 100 tests.
2954 </pre>
2955 <p>
2956 Let's be a bit more exhaustive:</p>
2959 <pre>
2960 >
2962 >
2963 </pre>
2964 </div>
2965 <p>
2968 <pre>
2969 *Main> deepCheck identityProp
2970 OK, passed 500 tests.
2971 </pre>
2973 <!-- ______________________________________________________________________ -->
2976 </div>
2978 <p>
2982 tag bits.</p>
2983 <p>
2987 the LSBit set. (Portability is relying on the fact that all known compilers place the
2990 </div>
2992 <!-- *********************************************************************** -->
2995 </div>
2996 <!-- *********************************************************************** -->
3002 <p>The Core LLVM classes are the primary means of representing the program
3003 being inspected or transformed. The core LLVM classes are defined in
3007 </div>
3009 <!-- ======================================================================= -->
3012 </div>
3020 subclasses. They are hidden because they offer no useful functionality beyond
3024 named, but this is not a requirement. There exists exactly
3025 one instance of a given shape at any one time. This allows type equality to
3026 be performed with address equality of the Type Instance. That is, given two
3028 </p>
3029 </div>
3031 <!-- _______________________________________________________________________ -->
3034 </div>
3038 <ul>
3042 floating point types.</li>
3045 an OpaqueType anywhere in its definition).</li>
3048 that don't have a size are abstract types, labels and void.</li>
3050 </ul>
3051 </div>
3053 <!-- _______________________________________________________________________ -->
3056 </div>
3058 <dl>
3060 <dd>Subclass of DerivedType that represents integer types of any bit width.
3063 <ul>
3065 type of a specific bit width.</li>
3067 type.</li>
3068 </ul>
3069 </dd>
3071 <dd>This is subclassed by ArrayType and PointerType
3072 <ul>
3074 of the elements in the sequential type. </li>
3075 </ul>
3076 </dd>
3078 <dd>This is a subclass of SequentialType and defines the interface for array
3079 types.
3080 <ul>
3082 elements in the array. </li>
3083 </ul>
3084 </dd>
3088 <dd>Subclass of SequentialType for vector types. A
3089 vector type is similar to an ArrayType but is distinguished because it is
3090 a first class type whereas ArrayType is not. Vector types are used for
3091 vector operations and are usually small vectors of of an integer or floating
3092 point type.</dd>
3096 <dd>Subclass of DerivedTypes for function types.
3097 <ul>
3099 function</li>
3101 return type of the function.</li>
3103 the type of the ith parameter.</li>
3105 number of formal parameters.</li>
3106 </ul>
3107 </dd>
3109 <dd>Sublcass of DerivedType for abstract types. This class
3110 defines no content and is used as a placeholder for some other type. Note
3111 that OpaqueType is used (temporarily) during type resolution for forward
3112 references of types. Once the referenced type is resolved, the OpaqueType
3113 is replaced with the actual type. OpaqueType can also be used for data
3114 abstraction. At link time opaque types can be resolved to actual types
3115 of the same name.</dd>
3116 </dl>
3117 </div>
3121 <!-- ======================================================================= -->
3124 </div>
3133 programs. An LLVM module is effectively either a translation unit of the
3134 original program or a combination of several translation units merged by the
3139 helpful member functions that try to make common operations easy.</p>
3141 </div>
3143 <!-- _______________________________________________________________________ -->
3146 </div>
3150 <ul>
3152 </ul>
3155 provide a name for it (probably based on the name of the translation unit).</p>
3157 <ul>
3164 <p>These are forwarding methods that make it easy to access the contents of
3166 list.</p></li>
3171 necessary to use when you need to update the list or perform a complex
3172 action that doesn't have a forwarding method.</p>
3175 </ul>
3177 <hr>
3179 <ul>
3187 <p> These are forwarding methods that make it easy to access the contents of
3195 use when you need to update the list or perform a complex action that
3196 doesn't have a forwarding method.</p>
3199 </ul>
3201 <hr>
3203 <ul>
3210 </ul>
3212 <hr>
3214 <ul>
3227 external declaration for the function and return it.</p></li>
3234 string.</p></li>
3241 name, true is returned and the <a
3243 </ul>
3245 </div>
3248 <!-- ======================================================================= -->
3251 </div>
3256 <br>
3260 base. It represents a typed value that may be used (among other things) as an
3268 for a program. For example, an incoming argument to a function (represented
3270 every instruction in the function that references the argument. To keep track
3276 methods, shown below.</p>
3280 method. In addition, all LLVM values can be named. The "name" of the
3284 <pre>
3286 </pre>
3287 </div>
3290 that the name of any value may be missing (an empty string), so names should
3292 debugging printouts), they should not be used to keep track of values or map
3296 <p>One important aspect of LLVM is that there is no distinction between an SSA
3297 variable and the operation that produces it. Because of this, any reference to
3298 the value produced by an instruction (or the value available as an incoming
3299 argument, for example) is represented as a direct pointer to the instance of
3300 the class that
3301 represents this value. Although this may take some getting used to, it
3302 simplifies the representation and makes it easier to manipulate.</p>
3304 </div>
3306 <!-- _______________________________________________________________________ -->
3309 </div>
3313 <ul>
3315 use-list<br>
3317 the use-list<br>
3319 value.<br>
3322 the use-list.<br>
3324 use-list.<br>
3326 element in the list.
3327 <p> These methods are the interface to access the def-use
3328 information in LLVM. As with all other iterators in LLVM, the naming
3330 </li>
3333 </li>
3339 </li>
3344 "<tt>V</tt>" instead. For example, if you detect that an instruction always
3345 produces a constant value (for example through constant folding), you can
3346 replace all uses of the instruction with the constant like this:</p>
3349 <pre>
3350 Inst->replaceAllUsesWith(ConstVal);
3351 </pre>
3352 </div>
3354 </ul>
3356 </div>
3358 <!-- ======================================================================= -->
3361 </div>
3365 <p>
3378 Single Assignment (SSA) form, there can only be one definition referred to,
3379 allowing this direct connection. This connection provides the use-def
3380 information in LLVM.</p>
3382 </div>
3384 <!-- _______________________________________________________________________ -->
3387 </div>
3392 an index access interface and through an iterator based interface.</p>
3394 <ul>
3398 convenient form for direct access.</p></li>
3401 list<br>
3403 the operand list.<br>
3405 operand list.
3406 <p> Together, these methods make up the iterator based interface to
3408 </ul>
3410 </div>
3412 <!-- ======================================================================= -->
3415 </div>
3426 instructions. It provides only a few methods, but is a very commonly used
3428 opcode (instruction type) and the parent <a
3430 into. To represent a specific type of instruction, one of many subclasses of
3439 file contains some meta-data about the various different types of instructions
3440 in LLVM. It describes the enum values that are used as opcodes (for example
3445 this file confuses doxygen, so these enum values don't show up correctly in the
3448 </div>
3450 <!-- _______________________________________________________________________ -->
3453 class</a>
3454 </div>
3456 <ul>
3458 <p>This subclasses represents all two operand instructions whose operands
3459 must be the same type, except for the comparison instructions.</p></li>
3462 common operations on cast instructions.</p>
3464 <p>This subclass respresents the two comparison instructions,
3468 <p>This subclass is the parent of all terminator instructions (those which
3469 can terminate a block).</p>
3470 </ul>
3471 </div>
3473 <!-- _______________________________________________________________________ -->
3476 class</a>
3477 </div>
3481 <ul>
3486 <p>Returns true if the instruction writes to memory, i.e. it is a
3491 <p>Returns another instance of the specified instruction, identical
3492 in all ways to the original except that the instruction has no parent
3494 and it has no name</p></li>
3495 </ul>
3497 </div>
3499 <!-- ======================================================================= -->
3502 </div>
3506 <p>Constant represents a base class for different types of constants. It
3507 is subclassed by ConstantInt, ConstantArray, etc. for representing
3509 a subclass, which represents the address of a global variable or function.
3510 </p>
3512 </div>
3514 <!-- _______________________________________________________________________ -->
3517 <ul>
3518 <li>ConstantInt : This subclass of Constant represents an integer constant of
3519 any width.
3520 <ul>
3522 value of this constant, an APInt value.</li>
3524 value to an int64_t via sign extension. If the value (not the bit width)
3525 of the APInt is too large to fit in an int64_t, an assertion will result.
3526 For this reason, use of this method is discouraged.</li>
3528 value to a uint64_t via zero extension. IF the value (not the bit width)
3529 of the APInt is too large to fit in a uint64_t, an assertion will result.
3530 For this reason, use of this method is discouraged.</li>
3533 The type is implied as the IntegerType that corresponds to the bit width
3536 Returns the ConstantInt object that represents the value provided by
3538 </ul>
3539 </li>
3540 <li>ConstantFP : This class represents a floating point constant.
3541 <ul>
3543 this constant. </li>
3544 </ul>
3545 </li>
3546 <li>ConstantArray : This represents a constant array.
3547 <ul>
3549 a vector of component constants that makeup this array. </li>
3550 </ul>
3551 </li>
3552 <li>ConstantStruct : This represents a constant struct.
3553 <ul>
3555 a vector of component constants that makeup this array. </li>
3556 </ul>
3557 </li>
3558 <li>GlobalValue : This represents either a global variable or a function. In
3559 either case, the value is a constant fixed address (after linking).
3560 </li>
3561 </ul>
3562 </div>
3565 <!-- ======================================================================= -->
3568 </div>
3575 Class</a><br>
3582 Because they are visible at global scope, they are also subject to linking with
3583 other globals defined in different translation units. To control the linking
3590 unit, and does not participate in linking. If it has external linkage, it is
3591 visible to external code, and does participate in linking. In addition to
3597 global is always a pointer to its contents. It is important to remember this
3602 the address of the first element of this array and the value of the
3610 </div>
3612 <!-- _______________________________________________________________________ -->
3615 class</a>
3616 </div>
3620 <ul>
3625 <p> </p>
3626 </li>
3629 GlobalValue is currently embedded into.</p></li>
3630 </ul>
3632 </div>
3634 <!-- ======================================================================= -->
3637 </div>
3650 actually one of the more complex classes in the LLVM hierarchy because it must
3658 ordering of the blocks in the function, which indicate how the code will be
3659 laid out by the backend. Additionally, the first <a
3662 block. There are no implicit exit nodes, and in fact there may be multiple exit
3666 function hasn't been linked in yet.</p>
3676 LLVM feature that is only used when you have to look up a value by name. Aside
3678 internally to make sure that there are not conflicts between the names of <a
3685 is its address (after linking) which is guaranteed to be constant.</p>
3686 </div>
3688 <!-- _______________________________________________________________________ -->
3691 class</a>
3692 </div>
3696 <ul>
3701 the the program. The constructor must specify the type of the function to
3702 create and what type of linkage the function should have. The <a
3704 specifies the formal arguments and return value for the function. The same
3707 in which the function is defined. If this argument is provided, the function
3708 will automatically be inserted into that module's list of
3709 functions.</p></li>
3714 function is "external", it does not have a body, and thus must be resolved
3715 by linking with a function defined in a different translation unit.</p></li>
3723 <p>These are forwarding methods that make it easy to access the contents of
3725 list.</p></li>
3730 is necessary to use when you need to update the list or perform a complex
3731 action that doesn't have a forwarding method.</p></li>
3734 iterator<br>
3740 <p>These are forwarding methods that make it easy to access the contents of
3742 list.</p></li>
3747 necessary to use when you need to update the list or perform a complex
3748 action that doesn't have a forwarding method.</p></li>
3753 function. Because the entry block for the function is always the first
3762 function.</p></li>
3768 </ul>
3770 </div>
3772 <!-- ======================================================================= -->
3775 </div>
3781 <br>
3783 Class</a><br>
3789 <p>Global variables are represented with the (surprise surprise)
3792 always referenced by their address (global values must live in memory, so their
3793 "name" refers to their constant address). See
3795 variables may have an initial value (which must be a
3797 they may be marked as "constant" themselves (indicating that their contents
3798 never change at runtime).</p>
3799 </div>
3801 <!-- _______________________________________________________________________ -->
3805 </div>
3809 <ul>
3814 <p>Create a new global variable of the specified type. If
3816 unchanging for the program. The Linkage parameter specifies the type of
3817 linkage (internal, external, weak, linkonce, appending) for the variable.
3818 If the linkage is InternalLinkage, WeakAnyLinkage, WeakODRLinkage,
3819 LinkOnceAnyLinkage or LinkOnceODRLinkage, then the resultant
3820 global variable will have internal linkage. AppendingLinkage concatenates
3821 together all instances (in different translation units) of the variable
3822 into a single variable but is only applicable to arrays. See
3824 further details on linkage types. Optionally an initializer, a name, and the
3825 module to put the variable into may be specified for the global variable as
3826 well.</p></li>
3830 <p>Returns true if this is a global variable that is known not to
3831 be modified at runtime.</p></li>
3840 to call this method if there is no initializer.</p></li>
3841 </ul>
3843 </div>
3846 <!-- ======================================================================= -->
3849 </div>
3856 Class</a><br>
3859 <p>This class represents a single entry multiple exit section of the code,
3860 commonly known as a basic block by the compiler community. The
3863 Matching the language definition, the last element of this list of instructions
3864 is always a terminator instruction (a subclass of the <a
3867 <p>In addition to tracking the list of instructions that make up the block, the
3876 </div>
3878 <!-- _______________________________________________________________________ -->
3881 class</a>
3882 </div>
3885 <ul>
3891 insertion into a function. The constructor optionally takes a name for the new
3894 automatically inserted at the end of the specified <a
3902 STL-style functions for accessing the instruction list.
3904 <p>These methods and typedefs are forwarding functions that have the same
3905 semantics as the standard library methods of the same names. These methods
3906 expose the underlying instruction list of a basic block in a way that is easy to
3907 manipulate. To get the full complement of container operations (including
3909 method.</p></li>
3913 <p>This method is used to get access to the underlying container that actually
3914 holds the Instructions. This method must be used when there isn't a forwarding
3916 to perform. Because there are no forwarding functions for "updating"
3917 operations, you need to use this if you want to update the contents of a
3923 embedded into, or a null pointer if it is homeless.</p></li>
3927 <p> Returns a pointer to the terminator instruction that appears at the end of
3929 instruction in the block is not a terminator, then a null pointer is
3930 returned.</p></li>
3932 </ul>
3934 </div>
3937 <!-- ======================================================================= -->
3940 </div>
3944 <p>This subclass of Value defines the interface for incoming formal
3945 arguments to a function. A Function maintains a list of its formal
3946 arguments. An argument has a pointer to the parent Function.</p>
3948 </div>
3950 <!-- *********************************************************************** -->
3951 <hr>
3952 <address>
3961 Last modified: $Date$
3962 </address>
3964 </body>
3965 </html>