| blob | 46fd33f40d552827589b83ebfbc05c70a1539053 |
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>
89 </ul></li>
91 <!--<ul>
92 todo
95 <ul>
99 </ul></li>
100 </ul>
101 </li>
103 <ul>
105 <ul>
113 class pointer</a> </li>
115 complex example</a> </li>
117 the same way</a> </li>
119 use-def chains</a> </li>
121 successors of blocks</a></li>
122 </ul>
123 </li>
125 <ul>
132 </ul>
133 </li>
135 <!--
136 <li>Working with the Control Flow Graph
137 <ul>
138 <li>Accessing predecessors and successors of a <tt>BasicBlock</tt>
139 <li>
140 <li>
141 </ul>
142 -->
143 </ul>
144 </li>
147 <ul>
149 </a></li>
154 </ul>
155 </li>
158 <ul>
160 <ul>
165 </ul></li>
167 <li><a href="#SymbolTable">The <tt>ValueSymbolTable</tt> and <tt>TypeSymbolTable</tt> classes</a></li>
168 <li><a href="#UserLayout">The <tt>User</tt> and owned <tt>Use</tt> classes' memory layout</a></li>
169 </ul></li>
172 <ul>
176 <ul>
178 <ul>
181 <ul>
183 <ul>
186 </ul>
187 </li>
188 </ul>
189 </li>
190 </ul>
191 </li>
194 </ul>
195 </li>
196 </ul>
197 </li>
198 </ol>
207 </div>
209 <!-- *********************************************************************** -->
212 </div>
213 <!-- *********************************************************************** -->
217 <p>This document is meant to highlight some of the important classes and
218 interfaces available in the LLVM source-base. This manual is not
219 intended to explain what LLVM is, how it works, and what LLVM code looks
220 like. It assumes that you know the basics of LLVM and are interested
221 in writing transformations or otherwise analyzing or manipulating the
222 code.</p>
224 <p>This document should get you oriented so that you can find your
225 way in the continuously growing source code that makes up the LLVM
226 infrastructure. Note that this manual is not intended to serve as a
227 replacement for reading the source code, so if you think there should be
228 a method in one of these classes to do something, but it's not listed,
230 are provided to make this as easy as possible.</p>
232 <p>The first section of this document describes general information that is
233 useful to know when working in the LLVM infrastructure, and the second describes
234 the Core LLVM classes. In the future this manual will be extended with
235 information describing how to use extension libraries, such as dominator
236 information, CFG traversal routines, and useful utilities like the <tt><a
239 </div>
241 <!-- *********************************************************************** -->
244 </div>
245 <!-- *********************************************************************** -->
249 <p>This section contains general information that is useful if you are working
250 in the LLVM source-base, but that isn't specific to any particular API.</p>
252 </div>
254 <!-- ======================================================================= -->
257 </div>
261 <p>LLVM makes heavy use of the C++ Standard Template Library (STL),
262 perhaps much more than you are used to, or have seen before. Because of
263 this, you might want to do a little background reading in the
264 techniques used and capabilities of the library. There are many good
265 pages that discuss the STL, and several books on the subject that you
266 can get, so it will not be discussed in this document.</p>
270 <ol>
273 reference</a> - an excellent reference for the STL and other parts of the
274 standard C++ library.</li>
277 O'Reilly book in the making. It has a decent Standard Library
278 Reference that rivals Dinkumware's, and is unfortunately no longer free since the
279 book has been published.</li>
282 Questions</a></li>
285 Contains a useful <a
290 Page</a></li>
294 the book).</a></li>
296 </ol>
300 to write maintainable code more than where to put your curly braces.</p>
302 </div>
304 <!-- ======================================================================= -->
307 </div>
311 <ol>
313 Branch and Tag Primer</a></li>
315 static and shared libraries across platforms</a></li>
316 </ol>
318 </div>
320 <!-- *********************************************************************** -->
323 </div>
324 <!-- *********************************************************************** -->
328 <p>Here we highlight some LLVM APIs that are generally useful and good to
329 know about when writing transformations.</p>
331 </div>
333 <!-- ======================================================================= -->
337 </div>
341 <p>The LLVM source-base makes extensive use of a custom form of RTTI.
343 operator, but they don't have some drawbacks (primarily stemming from
345 have a v-table). Because they are used so often, you must know what they
346 do and how they work. All of these templates are defined in the <a
348 file (note that you very rarely have to include this file directly).</p>
350 <dl>
354 "<tt>instanceof</tt>" operator. It returns true or false depending on whether
355 a reference or pointer points to an instance of the specified class. This can
356 be very useful for constraint checking of various sorts (example below).</p>
357 </dd>
362 converts a pointer or reference from a base class to a derived class, causing
363 an assertion failure if it is not really an instance of the right type. This
364 should be used in cases where you have some information that makes you believe
369 <pre>
371 if (isa<<a href="#Constant">Constant</a>>(V) || isa<<a href="#Argument">Argument</a>>(V) || isa<<a href="#GlobalValue">GlobalValue</a>>(V))
372 return true;
376 }
377 </pre>
378 </div>
382 operator.</p>
384 </dd>
389 It checks to see if the operand is of the specified type, and if so, returns a
390 pointer to it (this operator does not work with references). If the operand is
391 not of the correct type, a null pointer is returned. Thus, this works very
395 statement like this:</p>
398 <pre>
399 if (<a href="#AllocationInst">AllocationInst</a> *AI = dyn_cast<<a href="#AllocationInst">AllocationInst</a>>(Val)) {
401 }
402 </pre>
403 </div>
407 statement, which is very convenient.</p>
412 blocks to check for lots of different variants of classes. If you find
413 yourself wanting to do this, it is much cleaner and more efficient to use the
416 </dd>
422 argument (which it then propagates). This can sometimes be useful, allowing
423 you to combine several null checks into one.</p></dd>
429 as an argument (which it then propagates). This can sometimes be useful,
430 allowing you to combine several null checks into one.</p></dd>
432 </dl>
434 <p>These five templates can be used with any classes, whether they have a
435 v-table or not. To add support for these templates, you simply need to add
437 to. Describing this is currently outside the scope of this document, but there
438 are lots of examples in the LLVM source base.</p>
440 </div>
443 <!-- ======================================================================= -->
447 </div>
451 <p>Although LLVM generally does not do much string manipulation, we do have
452 several important APIs which take strings. Two important examples are the
453 Value class -- which has names for instructions, functions, etc. -- and the
454 StringMap class which is used extensively in LLVM and Clang.</p>
456 <p>These are generic classes, and they need to be able to accept strings which
457 may have embedded null characters. Therefore, they cannot simply take
459 clients to perform a heap allocation which is usually unnecessary. Instead,
463 </div>
465 <!-- _______________________________________________________________________ -->
468 </div>
473 (a character array and a length) and supports the common operations available
476 <p>It can be implicitly constructed using a C style null-terminated string,
481 iterator find(const StringRef &Key);
482 </div>
487 <pre>
491 </pre>
492 </div>
498 for more information.</p>
501 pointers to external memory it is not generally safe to store an instance of the
502 class (unless you know that the external storage will not be freed).</p>
504 </div>
506 <!-- _______________________________________________________________________ -->
509 </div>
514 strings. For example, a common LLVM paradigm is to name one instruction based on
515 the name of another instruction with a suffix, for example:</p>
518 <pre>
520 </pre>
521 </div>
525 which points to temporary (stack allocated) objects. Twines can be implicitly
526 constructed as the result of the plus operator applied to strings (i.e., a C
528 actual concatenation of strings until it is actually required, at which point
529 it can be efficiently rendered directly into a character array. This avoids
530 unnecessary heap allocation involved in constructing the temporary results of
531 string concatenation. See
533 for more information.</p>
536 and should almost never be stored or mentioned directly. They are intended
537 solely for use when defining a function which should be able to efficiently
538 accept concatenated strings.</p>
540 </div>
543 <!-- ======================================================================= -->
546 </div>
550 <p>Often when working on your pass you will put a bunch of debugging printouts
551 and other code into your pass. After you get it working, you want to remove
552 it, but you may need it again in the future (to work out new bugs that you run
553 across).</p>
555 <p> Naturally, because of this, you don't want to delete the debug printouts,
556 but you don't want them to always be noisy. A standard compromise is to comment
557 them out, allowing you to enable them if you need them in the future.</p>
561 this problem. Basically, you can put arbitrary code into the argument of the
566 <pre>
568 </pre>
569 </div>
574 <pre>
578 I am here!
579 </pre>
580 </div>
583 to not have to create "yet another" command line option for the debug output for
585 so they do not cause a performance impact at all (for the same reason, they
586 should also not contain side-effects!).</p>
589 enable or disable it directly in gdb. Just use "<tt>set DebugFlag=0</tt>" or
590 "<tt>set DebugFlag=1</tt>" from the gdb if the program is running. If the
591 program hasn't been started yet, you can always just run it with
594 </div>
596 <!-- _______________________________________________________________________ -->
600 </div>
606 generator). If you want to enable debug information with more fine-grained
608 option as follows:</p>
611 <pre>
612 #undef DEBUG_TYPE
614 #define DEBUG_TYPE "foo"
616 #undef DEBUG_TYPE
617 #define DEBUG_TYPE "bar"
619 #undef DEBUG_TYPE
620 #define DEBUG_TYPE ""
622 </pre>
623 </div>
628 <pre>
632 No debug type
633 'foo' debug type
634 'bar' debug type
635 No debug type (2)
637 'foo' debug type
639 'bar' debug type
640 </pre>
641 </div>
644 a file, to specify the debug type for the entire module (if you do this before
647 "bar", because there is no system in place to ensure that names do not
648 conflict. If two different modules use the same string, they will all be turned
649 on when the name is specified. This allows, for example, all debug information
651 even if the source lives in multiple files.</p>
655 statement. It takes an additional first parameter, which is the type to use. For
656 example, the preceding example could be written as:</p>
660 <pre>
665 </pre>
666 </div>
668 </div>
670 <!-- ======================================================================= -->
673 option</a>
674 </div>
681 keep track of what the LLVM compiler is doing and how effective various
682 optimizations are. It is useful to see what optimizations are contributing to
683 making a particular program run faster.</p>
685 <p>Often you may run your pass on some big program, and you're interested to see
686 how many times it makes a certain transformation. Although you can do this with
687 hand inspection, or some ad-hoc method, this is a real pain and not very useful
689 keep track of this information, and the calculated information is presented in a
690 uniform manner with the rest of the passes being executed.</p>
693 it are as follows:</p>
695 <ol>
699 <pre>
700 #define <a href="#DEBUG_TYPE">DEBUG_TYPE</a> "mypassname" <i>// This goes before any #includes.</i>
701 STATISTIC(NumXForms, "The # of times I did stuff");
702 </pre>
703 </div>
706 specified by the first argument. The pass name is taken from the DEBUG_TYPE
707 macro, and the description is taken from the second argument. The variable
713 <pre>
715 </pre>
716 </div>
718 </li>
719 </ol>
725 <pre>
728 </pre>
729 </div>
732 suite, it gives a report that looks like this:</p>
735 <pre>
736 7646 bitcodewriter - Number of normal instructions
737 725 bitcodewriter - Number of oversized instructions
738 129996 bitcodewriter - Number of bitcode bytes written
739 2817 raise - Number of insts DCEd or constprop'd
740 3213 raise - Number of cast-of-self removed
741 5046 raise - Number of expression trees converted
742 75 raise - Number of other getelementptr's formed
743 138 raise - Number of load/store peepholes
744 42 deadtypeelim - Number of unused typenames removed from symtab
745 392 funcresolve - Number of varargs functions resolved
746 27 globaldce - Number of global variables removed
747 2 adce - Number of basic blocks removed
748 134 cee - Number of branches revectored
749 49 cee - Number of setcc instruction eliminated
750 532 gcse - Number of loads removed
751 2919 gcse - Number of instructions removed
752 86 indvars - Number of canonical indvars added
753 87 indvars - Number of aux indvars removed
754 25 instcombine - Number of dead inst eliminate
755 434 instcombine - Number of insts combined
756 248 licm - Number of load insts hoisted
757 1298 licm - Number of insts hoisted to a loop pre-header
758 3 licm - Number of insts hoisted to multiple loop preds (bad, no loop pre-header)
759 75 mem2reg - Number of alloca's promoted
760 1444 cfgsimplify - Number of blocks simplified
761 </pre>
762 </div>
764 <p>Obviously, with so many optimizations, having a unified framework for this
765 stuff is very nice. Making your pass fit well into the framework makes it more
766 maintainable and useful.</p>
768 </div>
770 <!-- ======================================================================= -->
773 </div>
777 <p>Several of the important data structures in LLVM are graphs: for example
781 DAGs</a>. In many cases, while debugging various parts of the compiler, it is
782 nice to instantly visualize these graphs.</p>
784 <p>LLVM provides several callbacks that are available in a debug build to do
786 the current LLVM tool will pop up a window containing the CFG for the function
787 where each basic block is a node in the graph, and each node contains the
788 instructions in the block. Similarly, there also exists
793 up a window. Alternatively, you can sprinkle calls to these functions in your
794 code in places you want to debug.</p>
796 <p>Getting this to work requires a small amount of configuration. On Unix
798 toolkit, and make sure 'dot' and 'gv' are in your path. If you are running on
799 Mac OS/X, download and install the Mac OS/X <a
802 it) to your path. Once in your system and path are set up, rerun the LLVM
803 configure script and rebuild LLVM to enable this functionality.</p>
809 specified color (choices of colors can be found at <a
811 complex node attributes can be provided with <tt>call
814 Attributes</a>.) If you want to restart and clear all the current graph
817 </div>
819 <!-- *********************************************************************** -->
822 </div>
823 <!-- *********************************************************************** -->
828 and we commonly use STL data structures. This section describes the trade-offs
829 you should consider when you pick one.</p>
831 <p>
832 The first step is a choose your own adventure: do you want a sequential
833 container, a set-like container, or a map-like container? The most important
834 thing when choosing a container is the algorithmic properties of how you plan to
835 access the container. Based on that, you should use:</p>
837 <ul>
839 of an value based on another value. Map-like containers also support
840 efficient queries for containment (whether a key is in the map). Map-like
841 containers generally do not support efficient reverse mapping (values to
842 keys). If you need that, use two maps. Some map-like containers also
843 support efficient iteration through the keys in sorted order. Map-like
844 containers are the most expensive sort, only use them if you need one of
845 these capabilities.</li>
848 stuff into a container that automatically eliminates duplicates. Some
849 set-like containers support efficient iteration through the elements in
850 sorted order. Set-like containers are more expensive than sequential
851 containers.
852 </li>
855 the most efficient way to add elements and keeps track of the order they are
856 added to the collection. They permit duplicates and support efficient
857 iteration, but do not support efficient look-up based on a key.
858 </li>
861 container or reference structure that is used for character or byte
862 arrays.</li>
865 perform set operations on sets of numeric id's, while automatically
866 eliminating duplicates. Bit containers require a maximum of 1 bit for each
867 identifier you want to store.
868 </li>
869 </ul>
871 <p>
872 Once the proper category of container is determined, you can fine tune the
873 memory use, constant factors, and cache behaviors of access by intelligently
874 picking a member of the category. Note that constant factors and cache behavior
875 can be a big deal. If you have a vector that usually only contains a few
876 elements (but could contain many), for example, it's much better to use
878 . Doing so avoids (relatively) expensive malloc/free calls, which dwarf the
879 cost of adding the elements to the container. </p>
881 </div>
883 <!-- ======================================================================= -->
886 </div>
889 There are a variety of sequential containers available for you, based on your
890 needs. Pick the first in this section that will do what you want.
891 </div>
893 <!-- _______________________________________________________________________ -->
896 </div>
899 <p>Fixed size arrays are very simple and very fast. They are good if you know
900 exactly how many elements you have, or you have a (low) upper bound on how many
901 you have.</p>
902 </div>
904 <!-- _______________________________________________________________________ -->
907 </div>
910 <p>Heap allocated arrays (new[] + delete[]) are also simple. They are good if
911 the number of elements is variable, if you know how many elements you will need
912 before the array is allocated, and if the array is usually large (if not,
914 allocated array is the cost of the new/delete (aka malloc/free). Also note that
915 if you are allocating an array of a type with a constructor, the constructor and
916 destructors will be run for every element in the array (re-sizable vectors only
917 construct those elements actually used).</p>
918 </div>
920 <!-- _______________________________________________________________________ -->
923 </div>
928 it supports efficient iteration, lays out elements in memory order (so you can
929 do pointer arithmetic between elements), supports efficient push_back/pop_back
930 operations, supports efficient random access to its elements, etc.</p>
932 <p>The advantage of SmallVector is that it allocates space for
934 the SmallVector is dynamically smaller than N, no malloc is performed. This can
935 be a big win in cases where the malloc/free call is far more expensive than the
936 code that fiddles around with the elements.</p>
939 predecessors/successors of a block is usually less than 8). On the other hand,
940 this makes the size of the SmallVector itself large, so you don't want to
941 allocate lots of them (doing so will waste a lot of space). As such,
942 SmallVectors are most useful when on the stack.</p>
944 <p>SmallVector also provides a nice portable and efficient replacement for
947 </div>
949 <!-- _______________________________________________________________________ -->
952 </div>
955 <p>
956 std::vector is well loved and respected. It is useful when SmallVector isn't:
957 when the size of the vector is often large (thus the small optimization will
958 rarely be a benefit) or if you will be allocating many instances of the vector
959 itself (which would waste space for elements that aren't in the container).
960 vector is also useful when interfacing with code that expects vectors :).
961 </p>
966 <pre>
967 for ( ... ) {
969 use V;
970 }
971 </pre>
972 </div>
977 <pre>
979 for ( ... ) {
980 use V;
981 V.clear();
982 }
983 </pre>
984 </div>
986 <p>Doing so will save (at least) one heap allocation and free per iteration of
987 the loop.</p>
989 </div>
991 <!-- _______________________________________________________________________ -->
994 </div>
997 <p>std::deque is, in some senses, a generalized version of std::vector. Like
998 std::vector, it provides constant time random access and other similar
999 properties, but it also provides efficient access to the front of the list. It
1000 does not guarantee continuity of elements within memory.</p>
1002 <p>In exchange for this extra flexibility, std::deque has significantly higher
1003 constant factor costs than std::vector. If possible, use std::vector or
1004 something cheaper.</p>
1005 </div>
1007 <!-- _______________________________________________________________________ -->
1010 </div>
1013 <p>std::list is an extremely inefficient class that is rarely useful.
1014 It performs a heap allocation for every element inserted into it, thus having an
1015 extremely high constant factor, particularly for small data types. std::list
1016 also only supports bidirectional iteration, not random access iteration.</p>
1018 <p>In exchange for this high cost, std::list supports efficient access to both
1019 ends of the list (like std::deque, but unlike std::vector or SmallVector). In
1020 addition, the iterator invalidation characteristics of std::list are stronger
1021 than that of a vector class: inserting or removing an element into the list does
1022 not invalidate iterator or pointers to other elements in the list.</p>
1023 </div>
1025 <!-- _______________________________________________________________________ -->
1028 </div>
1032 intrusive, because it requires the element to store and provide access to the
1033 prev/next pointers for the list.</p>
1037 provides some novel characteristics. In particular, it can efficiently store
1038 polymorphic objects, the traits class is informed when an element is inserted or
1040 constant-time splice operation.</p>
1042 <p>These properties are exactly what we want for things like
1046 Related classes of interest are explained in the following subsections:
1047 <ul>
1052 </ul>
1053 </div>
1055 <!-- _______________________________________________________________________ -->
1058 </div>
1063 publicly derive from this traits class.</p>
1064 </div>
1066 <!-- _______________________________________________________________________ -->
1069 </div>
1073 supports a slightly narrower interface. Notably, inserters from
1077 used for a wide variety of customizations.</p>
1078 </div>
1080 <!-- _______________________________________________________________________ -->
1083 </div>
1088 in the default manner.</p>
1093 </div>
1095 <!-- _______________________________________________________________________ -->
1098 </div>
1102 citizen in the C++ ecosystem, it needs to support the standard container
1107 <p>The only sensible solution to this problem is to allocate a so-called
1109 iterator, providing the back-link to the last element. However conforming to the
1111 also must not be dereferenced.</p>
1114 how to allocate and store the sentinel. The corresponding policy is dictated
1116 whenever the need for a sentinel arises.</p>
1118 <p>While the default policy is sufficient in most cases, it may break down when
1121 is wasted. To alleviate the situation with numerous and voluminous
1127 arithmetic is used to obtain the sentinel, which is relative to the
1129 extra pointer, which serves as the back-link of the sentinel. This is the only
1130 field in the ghostly sentinel which can be legally accessed.</p>
1131 </div>
1133 <!-- _______________________________________________________________________ -->
1136 </div>
1141 <p>There are also various STL adapter classes such as std::queue,
1142 std::priority_queue, std::stack, etc. These provide simplified access to an
1143 underlying container but don't affect the cost of the container itself.</p>
1145 </div>
1148 <!-- ======================================================================= -->
1151 </div>
1155 <p>Set-like containers are useful when you need to canonicalize multiple values
1156 into a single representation. There are several different choices for how to do
1157 this, providing various trade-offs.</p>
1159 </div>
1162 <!-- _______________________________________________________________________ -->
1165 </div>
1169 <p>If you intend to insert a lot of elements, then do a lot of queries, a
1170 great approach is to use a vector (or other sequential container) with
1171 std::sort+std::unique to remove duplicates. This approach works really well if
1172 your usage pattern has these two distinct phases (insert then query), and can be
1174 </p>
1176 <p>
1177 This combination provides the several nice properties: the result data is
1178 contiguous in memory (good for cache locality), has few allocations, is easy to
1179 address (iterators in the final vector are just indices or pointers), and can be
1180 efficiently queried with a standard binary or radix search.</p>
1182 </div>
1184 <!-- _______________________________________________________________________ -->
1187 </div>
1191 <p>If you have a set-like data structure that is usually small and whose elements
1193 has space for N elements in place (thus, if the set is dynamically smaller than
1194 N, no malloc traffic is required) and accesses them with a simple linear search.
1195 When the set grows beyond 'N' elements, it allocates a more expensive representation that
1196 guarantees efficient access (for most types, it falls back to std::set, but for
1197 pointers it uses something far better, <a
1200 <p>The magic of this class is that it handles small sets extremely efficiently,
1201 but gracefully handles extremely large sets without loss of efficiency. The
1202 drawback is that the interface is quite small: it supports insertion, queries
1203 and erasing, but does not support iteration.</p>
1205 </div>
1207 <!-- _______________________________________________________________________ -->
1210 </div>
1214 <p>SmallPtrSet has all the advantages of <tt>SmallSet</tt> (and a <tt>SmallSet</tt> of pointers is
1216 more than 'N' insertions are performed, a single quadratically
1217 probed hash table is allocated and grows as needed, providing extremely
1218 efficient access (constant time insertion/deleting/queries with low constant
1219 factors) and is very stingy with malloc traffic.</p>
1222 whenever an insertion occurs. Also, the values visited by the iterators are not
1223 visited in sorted order.</p>
1225 </div>
1227 <!-- _______________________________________________________________________ -->
1230 </div>
1234 <p>
1235 DenseSet is a simple quadratically probed hash table. It excels at supporting
1236 small values: it uses a single allocation to hold all of the pairs that
1237 are currently inserted in the set. DenseSet is a great way to unique small
1238 values that are not simple pointers (use <a
1240 the same requirements for the value type that <a
1242 </p>
1244 </div>
1246 <!-- _______________________________________________________________________ -->
1249 </div>
1253 <p>
1254 FoldingSet is an aggregate class that is really good at uniquing
1255 expensive-to-create or polymorphic objects. It is a combination of a chained
1256 hash table with intrusive links (uniqued objects are required to inherit from
1258 its ID process.</p>
1261 a complex object (for example, a node in the code generator). The client has a
1262 description of *what* it wants to generate (it knows the opcode and all the
1263 operands), but we don't want to 'new' a node, then try inserting it into a set
1264 only to find out it already exists, at which point we would have to delete it
1265 and return the node that already exists.
1266 </p>
1268 <p>To support this style of client, FoldingSet perform a query with a
1269 FoldingSetNodeID (which wraps SmallVector) that can be used to describe the
1270 element that we want to query for. The query either returns the element
1271 matching the ID or it returns an opaque ID that indicates where insertion should
1272 take place. Construction of the ID usually does not require heap traffic.</p>
1274 <p>Because FoldingSet uses intrusive links, it can support polymorphic objects
1275 in the set (for example, you can have SDNode instances mixed with LoadSDNodes).
1276 Because the elements are individually allocated, pointers to the elements are
1277 stable: inserting or removing elements does not invalidate any pointers to other
1278 elements.
1279 </p>
1281 </div>
1283 <!-- _______________________________________________________________________ -->
1286 </div>
1291 many things but great at nothing. std::set allocates memory for each element
1292 inserted (thus it is very malloc intensive) and typically stores three pointers
1293 per element in the set (thus adding a large amount of per-element space
1294 overhead). It offers guaranteed log(n) performance, which is not particularly
1295 fast from a complexity standpoint (particularly if the elements of the set are
1296 expensive to compare, like strings), and has extremely high constant factors for
1297 lookup, insertion and removal.</p>
1299 <p>The advantages of std::set are that its iterators are stable (deleting or
1300 inserting an element from the set does not affect iterators or pointers to other
1301 elements) and that iteration over the set is guaranteed to be in sorted order.
1302 If the elements in the set are large, then the relative overhead of the pointers
1303 and malloc traffic is not a big deal, but if the elements of the set are small,
1304 std::set is almost never a good choice.</p>
1306 </div>
1308 <!-- _______________________________________________________________________ -->
1311 </div>
1316 Container</a>. The important property
1317 that this provides is efficient insertion with uniquing (duplicate elements are
1318 ignored) with iteration support. It implements this by inserting elements into
1319 both a set-like container and the sequential container, using the set-like
1320 container for uniquing and the sequential container for iteration.
1321 </p>
1323 <p>The difference between SetVector and other sets is that the order of
1324 iteration is guaranteed to match the order of insertion into the SetVector.
1325 This property is really important for things like sets of pointers. Because
1326 pointer values are non-deterministic (e.g. vary across runs of the program on
1327 different machines), iterating over the pointers in the set will
1328 not be in a well-defined order.</p>
1330 <p>
1331 The drawback of SetVector is that it requires twice as much space as a normal
1332 set and has the sum of constant factors from the set-like container and the
1333 sequential container that it uses. Use it *only* if you need to iterate over
1334 the elements in a deterministic order. SetVector is also expensive to delete
1335 elements out of (linear time), unless you use it's "pop_back" method, which is
1336 faster.
1337 </p>
1339 <p>SetVector is an adapter class that defaults to using std::vector and std::set
1340 for the underlying containers, so it is quite expensive. However,
1342 defaults to using a SmallVector and SmallSet of a specified size. If you use
1343 this, and if your sets are dynamically smaller than N, you will save a lot of
1344 heap traffic.</p>
1346 </div>
1348 <!-- _______________________________________________________________________ -->
1351 </div>
1355 <p>
1357 retains a unique ID for each element inserted into the set. It internally
1358 contains a map and a vector, and it assigns a unique ID for each value inserted
1359 into the set.</p>
1361 <p>UniqueVector is very expensive: its cost is the sum of the cost of
1362 maintaining both the map and vector, it has high complexity, high constant
1363 factors, and produces a lot of malloc traffic. It should be avoided.</p>
1365 </div>
1368 <!-- _______________________________________________________________________ -->
1371 </div>
1375 <p>
1376 The STL provides several other options, such as std::multiset and the various
1377 "hash_set" like containers (whether from C++ TR1 or from the SGI library). We
1378 never use hash_set and unordered_set because they are generally very expensive
1379 (each insertion requires a malloc) and very non-portable.
1380 </p>
1382 <p>std::multiset is useful if you're not interested in elimination of
1383 duplicates, but has all the drawbacks of std::set. A sorted vector (where you
1384 don't delete duplicate entries) or some other approach is almost always
1385 better.</p>
1387 </div>
1389 <!-- ======================================================================= -->
1392 </div>
1395 Map-like containers are useful when you want to associate data to a key. As
1396 usual, there are a lot of different ways to do this. :)
1397 </div>
1399 <!-- _______________________________________________________________________ -->
1402 </div>
1406 <p>
1407 If your usage pattern follows a strict insert-then-query approach, you can
1409 for set-like containers</a>. The only difference is that your query function
1410 (which uses std::lower_bound to get efficient log(n) lookup) should only compare
1411 the key, not both the key and value. This yields the same advantages as sorted
1412 vectors for sets.
1413 </p>
1414 </div>
1416 <!-- _______________________________________________________________________ -->
1419 </div>
1423 <p>
1424 Strings are commonly used as keys in maps, and they are difficult to support
1425 efficiently: they are variable length, inefficient to hash and compare when
1426 long, expensive to copy, etc. StringMap is a specialized container designed to
1427 cope with these issues. It supports mapping an arbitrary range of bytes to an
1428 arbitrary other object.</p>
1430 <p>The StringMap implementation uses a quadratically-probed hash table, where
1431 the buckets store a pointer to the heap allocated entries (and some other
1432 stuff). The entries in the map must be heap allocated because the strings are
1433 variable length. The string data (key) and the element object (value) are
1434 stored in the same allocation with the string data immediately after the element
1435 object. This container guarantees the "<tt>(char*)(&Value+1)</tt>" points
1436 to the key string for a value.</p>
1438 <p>The StringMap is very fast for several reasons: quadratic probing is very
1439 cache efficient for lookups, the hash value of strings in buckets is not
1440 recomputed when lookup up an element, StringMap rarely has to touch the
1441 memory for unrelated objects when looking up a value (even when hash collisions
1442 happen), hash table growth does not recompute the hash values for strings
1443 already in the table, and each pair in the map is store in a single allocation
1444 (the string data is stored in the same allocation as the Value of a pair).</p>
1446 <p>StringMap also provides query methods that take byte ranges, so it only ever
1447 copies a string if a value is inserted into the table.</p>
1448 </div>
1450 <!-- _______________________________________________________________________ -->
1453 </div>
1456 <p>
1457 IndexedMap is a specialized container for mapping small dense integers (or
1458 values that can be mapped to small dense integers) to some other type. It is
1459 internally implemented as a vector with a mapping function that maps the keys to
1460 the dense integer range.
1461 </p>
1463 <p>
1464 This is useful for cases like virtual registers in the LLVM code generator: they
1465 have a dense mapping that is offset by a compile-time constant (the first
1466 virtual register ID).</p>
1468 </div>
1470 <!-- _______________________________________________________________________ -->
1473 </div>
1477 <p>
1478 DenseMap is a simple quadratically probed hash table. It excels at supporting
1479 small keys and values: it uses a single allocation to hold all of the pairs that
1480 are currently inserted in the map. DenseMap is a great way to map pointers to
1481 pointers, or map other small types to each other.
1482 </p>
1484 <p>
1485 There are several aspects of DenseMap that you should be aware of, however. The
1486 iterators in a densemap are invalidated whenever an insertion occurs, unlike
1487 map. Also, because DenseMap allocates space for a large number of key/value
1488 pairs (it starts with 64 by default), it will waste a lot of space if your keys
1489 or values are large. Finally, you must implement a partial specialization of
1490 DenseMapInfo for the key that you want, if it isn't already supported. This
1491 is required to tell DenseMap about two special marker values (which can never be
1492 inserted into the map) that it needs internally.</p>
1494 </div>
1496 <!-- _______________________________________________________________________ -->
1499 </div>
1503 <p>
1505 Value*s (or subclasses) to another type. When a Value is deleted or RAUW'ed,
1506 ValueMap will update itself so the new version of the key is mapped to the same
1507 value, just as if the key were a WeakVH. You can configure exactly how this
1508 happens, and what else happens on these two events, by passing
1511 </div>
1513 <!-- _______________________________________________________________________ -->
1516 </div>
1520 <p>
1522 a single allocation per pair inserted into the map, it offers log(n) lookup with
1523 an extremely large constant factor, imposes a space penalty of 3 pointers per
1524 pair in the map, etc.</p>
1526 <p>std::map is most useful when your keys or values are very large, if you need
1527 to iterate over the collection in sorted order, or if you need stable iterators
1528 into the map (i.e. they don't get invalidated if an insertion or deletion of
1529 another element takes place).</p>
1531 </div>
1533 <!-- _______________________________________________________________________ -->
1536 </div>
1540 <p>
1541 The STL provides several other options, such as std::multimap and the various
1542 "hash_map" like containers (whether from C++ TR1 or from the SGI library). We
1543 never use hash_set and unordered_set because they are generally very expensive
1544 (each insertion requires a malloc) and very non-portable.</p>
1546 <p>std::multimap is useful if you want to map a key to multiple values, but has
1547 all the drawbacks of std::map. A sorted vector or some other approach is almost
1548 always better.</p>
1550 </div>
1552 <!-- ======================================================================= -->
1555 </div>
1559 <p>
1560 TODO: const char* vs stringref vs smallstring vs std::string. Describe twine,
1561 xref to #string_apis.
1562 </p>
1564 </div>
1566 <!-- ======================================================================= -->
1569 </div>
1572 <p>Unlike the other containers, there are only two bit storage containers, and
1573 choosing when to use each is relatively straightforward.</p>
1575 <p>One additional option is
1577 implementation in many common compilers (e.g. commonly available versions of
1578 GCC) is extremely inefficient and 2) the C++ standards committee is likely to
1579 deprecate this container and/or change it significantly somehow. In any case,
1580 please don't use it.</p>
1581 </div>
1583 <!-- _______________________________________________________________________ -->
1586 </div>
1589 <p> The BitVector container provides a dynamic size set of bits for manipulation.
1590 It supports individual bit setting/testing, as well as set operations. The set
1591 operations take time O(size of bitvector), but operations are performed one word
1592 at a time, instead of one bit at a time. This makes the BitVector very fast for
1593 set operations compared to other containers. Use the BitVector when you expect
1594 the number of set bits to be high (IE a dense set).
1595 </p>
1596 </div>
1598 <!-- _______________________________________________________________________ -->
1601 </div>
1604 <p> The SmallBitVector container provides the same interface as BitVector, but
1605 it is optimized for the case where only a small number of bits, less than
1606 25 or so, are needed. It also transparently supports larger bit counts, but
1607 slightly less efficiently than a plain BitVector, so SmallBitVector should
1608 only be used when larger counts are rare.
1609 </p>
1611 <p>
1612 At this time, SmallBitVector does not support set operations (and, or, xor),
1613 and its operator[] does not provide an assignable lvalue.
1614 </p>
1615 </div>
1617 <!-- _______________________________________________________________________ -->
1620 </div>
1623 <p> The SparseBitVector container is much like BitVector, with one major
1624 difference: Only the bits that are set, are stored. This makes the
1625 SparseBitVector much more space efficient than BitVector when the set is sparse,
1626 as well as making set operations O(number of set bits) instead of O(size of
1627 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
1628 (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).
1629 </p>
1630 </div>
1632 <!-- *********************************************************************** -->
1635 </div>
1636 <!-- *********************************************************************** -->
1640 <p>This section describes how to perform some very simple transformations of
1641 LLVM code. This is meant to give examples of common idioms used, showing the
1643 you should also read about the main classes that you will be working with. The
1645 and descriptions of the main classes that you should know about.</p>
1647 </div>
1649 <!-- NOTE: this section should be heavy on example code -->
1650 <!-- ======================================================================= -->
1653 </div>
1657 <p>The LLVM compiler infrastructure have many different data structures that may
1658 be traversed. Following the example of the C++ standard template library, the
1659 techniques used to traverse these various data structures are all basically the
1662 function returns an iterator pointing to one past the last valid element of the
1664 between the two operations.</p>
1666 <p>Because the pattern for iteration is common across many different aspects of
1667 the program representation, the standard template library algorithms may be used
1668 on them, and it is easier to remember how to iterate. First we show a few common
1669 examples of the data structures that need to be traversed. Other data
1670 structures are traversed in very similar ways.</p>
1672 </div>
1674 <!-- _______________________________________________________________________ -->
1679 </div>
1684 transform in some way; in particular, you'd like to manipulate its
1691 <pre>
1698 </pre>
1699 </div>
1701 <p>Note that i can be used as if it were a pointer for the purposes of
1703 because the indirection operator is overloaded for the iterator
1707 </div>
1709 <!-- _______________________________________________________________________ -->
1714 </div>
1719 easy to iterate over the individual instructions that make up
1724 <pre>
1730 </pre>
1731 </div>
1733 <p>However, this isn't really the best way to print out the contents of a
1735 anything you'll care about, you could have just invoked the print routine on the
1738 </div>
1740 <!-- _______________________________________________________________________ -->
1745 </div>
1754 small example that shows how to dump all instructions in a function to the standard error stream:<p>
1757 <pre>
1761 for (inst_iterator I = inst_begin(F), E = inst_end(F); I != E; ++I)
1763 </pre>
1764 </div>
1767 work list with its initial contents. For example, if you wanted to
1769 F, all you would need to do is something like:</p>
1772 <pre>
1776 for (inst_iterator I = inst_begin(F), E = inst_end(F); I != E; ++I)
1777 worklist.insert(&*I);
1778 </pre>
1779 </div>
1784 </div>
1786 <!-- _______________________________________________________________________ -->
1789 vice-versa)</a>
1790 </div>
1794 <p>Sometimes, it'll be useful to grab a reference (or pointer) to a class
1795 instance when all you've got at hand is an iterator. Well, extracting
1796 a reference or a pointer from an iterator is very straight-forward.
1801 <pre>
1804 const Instruction& inst = *j;
1805 </pre>
1806 </div>
1808 <p>However, the iterators you'll be working with in the LLVM framework are
1809 special: they will automatically convert to a ptr-to-instance type whenever they
1810 need to. Instead of dereferencing the iterator and then taking the address of
1811 the result, you can simply assign the iterator to the proper pointer type and
1812 you get the dereference and address-of operation as a result of the assignment
1813 (behind the scenes, this is a result of overloading casting mechanisms). Thus
1814 the last line of the last example,</p>
1817 <pre>
1818 Instruction *pinst = &*i;
1819 </pre>
1820 </div>
1825 <pre>
1826 Instruction *pinst = i;
1827 </pre>
1828 </div>
1830 <p>It's also possible to turn a class pointer into the corresponding iterator,
1831 and this is a constant time operation (very efficient). The following code
1832 snippet illustrates use of the conversion constructors provided by LLVM
1833 iterators. By using these, you can explicitly grab the iterator of something
1834 without actually obtaining it via iteration over some structure:</p>
1837 <pre>
1838 void printNextInstruction(Instruction* inst) {
1839 BasicBlock::iterator it(inst);
1842 }
1843 </pre>
1844 </div>
1846 <p>Unfortunately, these implicit conversions come at a cost; they prevent
1847 these iterators from conforming to standard iterator conventions, and thus
1848 from being usable with standard algorithms and containers. For example, they
1850 from compiling:</p>
1853 <pre>
1855 </pre>
1856 </div>
1858 <p>Because of this, these implicit conversions may be removed some day,
1861 </div>
1863 <!--_______________________________________________________________________-->
1866 example</a>
1867 </div>
1871 <p>Say that you're writing a FunctionPass and would like to count all the
1875 much more straight-forward manner, but this example will allow us to explore how
1877 is what we want to do:</p>
1880 <pre>
1881 initialize callCounter to zero
1882 for each Function f in the Module
1883 for each BasicBlock b in f
1884 for each Instruction i in b
1885 if (i is a CallInst and calls the given function)
1886 increment callCounter
1887 </pre>
1888 </div>
1890 <p>And the actual code is (remember, because we're writing a
1895 <pre>
1896 Function* targetFunc = ...;
1898 class OurFunctionPass : public FunctionPass {
1899 public:
1900 OurFunctionPass(): callCounter(0) { }
1902 virtual runOnFunction(Function& F) {
1903 for (Function::iterator b = F.begin(), be = F.end(); b != be; ++b) {
1910 if (callInst->getCalledFunction() == targetFunc)
1911 ++callCounter;
1912 }
1913 }
1914 }
1915 }
1917 private:
1918 unsigned callCounter;
1919 };
1920 </pre>
1921 </div>
1923 </div>
1925 <!--_______________________________________________________________________-->
1928 </div>
1932 <p>You may have noticed that the previous example was a bit oversimplified in
1933 that it did not deal with call sites generated by 'invoke' instructions. In
1934 this, and in other situations, you may find that you want to treat
1937 less closely-related things. For these cases, LLVM provides a handy wrapper
1938 class called <a
1945 reference and it should not be dynamically allocated or deallocated using
1947 assignable and constructable, with costs equivalents to that of a bare pointer.
1948 If you look at its definition, it has only a single pointer member.</p>
1950 </div>
1952 <!--_______________________________________________________________________-->
1955 </div>
1969 <pre>
1970 Function *F = ...;
1976 }
1977 </pre>
1978 </div>
1982 doing it only once in the loop body and reusing its result.</p>
1989 all of the values that a particular instruction uses (that is, the operands of
1993 <pre>
1994 Instruction *pi = ...;
1997 Value *v = *i;
1999 }
2000 </pre>
2001 </div>
2004 mutation free algorithms (such as analyses, etc.). For this purpose above
2011 </div>
2013 <!--_______________________________________________________________________-->
2016 successors of blocks</a>
2017 </div>
2021 <p>Iterating over the predecessors and successors of a block is quite easy
2023 this to iterate over all predecessors of BB:</p>
2026 <pre>
2027 #include "llvm/Support/CFG.h"
2028 BasicBlock *BB = ...;
2030 for (pred_iterator PI = pred_begin(BB), E = pred_end(BB); PI != E; ++PI) {
2031 BasicBlock *Pred = *PI;
2033 }
2034 </pre>
2035 </div>
2037 <p>Similarly, to iterate over successors use
2038 succ_iterator/succ_begin/succ_end.</p>
2040 </div>
2043 <!-- ======================================================================= -->
2046 </div>
2050 <p>There are some primitive transformation operations present in the LLVM
2051 infrastructure that are worth knowing about. When performing
2052 transformations, it's fairly common to manipulate the contents of basic
2053 blocks. This section describes some of the common methods for doing so
2054 and gives example code.</p>
2056 </div>
2058 <!--_______________________________________________________________________-->
2062 </div>
2069 constructor for the kind of instruction to instantiate and provide the necessary
2074 <pre>
2075 AllocaInst* ai = new AllocaInst(Type::Int32Ty);
2076 </pre>
2077 </div>
2081 subclass is likely to have varying default parameters which change the semantics
2082 of the instruction, so refer to the <a
2088 <p>It is very useful to name the values of instructions when you're able to, as
2089 this facilitates the debugging of your transformations. If you end up looking
2090 at generated LLVM machine code, you definitely want to have logical names
2091 associated with the results of instructions! By supplying a value for the
2093 associate a logical name with the result of the instruction's execution at
2094 run time. For example, say that I'm writing a transformation that dynamically
2095 allocates space for an integer on the stack, and that integer is going to be
2096 used as some kind of index by some other code. To accomplish this, I place an
2102 <pre>
2104 </pre>
2105 </div>
2108 execution value, which is a pointer to an integer on the run time stack.</p>
2115 <ul>
2116 <li>Insertion into an explicit instruction list
2123 <pre>
2124 BasicBlock *pb = ...;
2125 Instruction *pi = ...;
2126 Instruction *newInst = new Instruction(...);
2129 </pre>
2130 </div>
2134 classes provide constructors which take a pointer to a
2136 looked like: </p>
2139 <pre>
2140 BasicBlock *pb = ...;
2141 Instruction *newInst = new Instruction(...);
2144 </pre>
2145 </div>
2150 <pre>
2151 BasicBlock *pb = ...;
2152 Instruction *newInst = new Instruction(..., pb);
2153 </pre>
2154 </div>
2156 <p>which is much cleaner, especially if you are creating
2157 long instruction streams.</p></li>
2159 <li>Insertion into an implicit instruction list
2162 are implicitly associated with an existing instruction list: the instruction
2163 list of the enclosing basic block. Thus, we could have accomplished the same
2165 </p>
2168 <pre>
2169 Instruction *pi = ...;
2170 Instruction *newInst = new Instruction(...);
2173 </pre>
2174 </div>
2176 <p>In fact, this sequence of steps occurs so frequently that the
2178 constructors which take (as a default parameter) a pointer to an
2182 provided instruction, immediately before that instruction. Using an
2184 parameter, the above code becomes:</p>
2187 <pre>
2188 Instruction* pi = ...;
2189 Instruction* newInst = new Instruction(..., pi);
2190 </pre>
2191 </div>
2193 <p>which is much cleaner, especially if you're creating a lot of
2195 </ul>
2197 </div>
2199 <!--_______________________________________________________________________-->
2202 </div>
2206 <p>Deleting an instruction from an existing sequence of instructions that form a
2208 you must have a pointer to the instruction that you wish to delete. Second, you
2209 need to obtain the pointer to that instruction's basic block. You use the
2210 pointer to the basic block to get its list of instructions and then use the
2211 erase function to remove your instruction. For example:</p>
2214 <pre>
2216 I->eraseFromParent();
2217 </pre>
2218 </div>
2220 </div>
2222 <!--_______________________________________________________________________-->
2226 </div>
2232 <p>Including "<a href="/doxygen/BasicBlockUtils_8h-source.html">llvm/Transforms/Utils/BasicBlockUtils.h</a>"
2238 <ul>
2241 <p>This function replaces all uses of a given instruction with a value,
2242 and then removes the original instruction. The following example
2243 illustrates the replacement of the result of a particular
2245 pointer to an integer.</p>
2248 <pre>
2249 AllocaInst* instToReplace = ...;
2250 BasicBlock::iterator ii(instToReplace);
2253 Constant::getNullValue(PointerType::getUnqual(Type::Int32Ty)));
2254 </pre></div></li>
2258 <p>This function replaces a particular instruction with another
2259 instruction, inserting the new instruction into the basic block at the
2260 location where the old instruction was, and replacing any uses of the old
2261 instruction with the new instruction. The following example illustrates
2265 <pre>
2266 AllocaInst* instToReplace = ...;
2267 BasicBlock::iterator ii(instToReplace);
2271 </pre></div></li>
2272 </ul>
2280 information.</p>
2282 <!-- Value::replaceAllUsesWith User::replaceUsesOfWith Point out:
2283 include/llvm/Transforms/Utils/ especially BasicBlockUtils.h with:
2284 ReplaceInstWithValue, ReplaceInstWithInst -->
2286 </div>
2288 <!--_______________________________________________________________________-->
2291 </div>
2295 <p>Deleting a global variable from a module is just as easy as deleting an
2296 Instruction. First, you must have a pointer to the global variable that you wish
2297 to delete. You use this pointer to erase it from its parent, the module.
2298 For example:</p>
2301 <pre>
2304 GV->eraseFromParent();
2305 </pre>
2306 </div>
2308 </div>
2310 <!-- ======================================================================= -->
2313 </div>
2317 <p>In generating IR, you may need some complex types. If you know these types
2320 has two forms depending on whether you're building types for cross-compilation
2323 out of types from
2325 namespace and pointers, functions, arrays, etc. built of
2327 whose size may depend on the host compiler. For example,</p>
2330 <pre>
2332 </pre>
2333 </div>
2338 <pre>
2340 params.push_back(PointerType::getUnqual(Type::Int32Ty));
2341 FunctionType *ft = FunctionType::get(Type::Int8Ty, params, false);
2342 </pre>
2343 </div>
2348 </div>
2350 <!-- *********************************************************************** -->
2353 </div>
2354 <!-- *********************************************************************** -->
2357 <p>
2358 This section describes the interaction of the LLVM APIs with multithreading,
2359 both on the part of client applications, and in the JIT, in the hosted
2360 application.
2361 </p>
2363 <p>
2364 Note that LLVM's support for multithreading is still relatively young. Up
2365 through version 2.5, the execution of threaded hosted applications was
2366 supported, but not threaded client access to the APIs. While this use case is
2368 ensure proper operation in multithreaded mode.
2369 </p>
2371 <p>
2372 Note that, on Unix-like platforms, LLVM requires the presence of GCC's atomic
2373 intrinsics in order to support threaded operation. If you need a
2374 multhreading-capable LLVM on a platform without a suitably modern system
2375 compiler, consider compiling LLVM and LLVM-GCC in single-threaded mode, and
2376 using the resultant compiler to build a copy of LLVM with multithreading
2377 support.
2378 </p>
2379 </div>
2381 <!-- ======================================================================= -->
2384 </div>
2388 <p>
2389 In order to properly protect its internal data structures while avoiding
2390 excessive locking overhead in the single-threaded case, the LLVM must intialize
2391 certain data structures necessary to provide guards around its internals. To do
2393 making any concurrent LLVM API calls. To subsequently tear down these
2396 mode.
2397 </p>
2399 <p>
2401 say that no other LLVM API calls may be executing at any time during the
2403 </tt>. It's is the client's responsibility to enforce this isolation.
2404 </p>
2406 <p>
2408 failure of the initialization. Failure typically indicates that your copy of
2409 LLVM was built without multithreading support, typically because GCC atomic
2410 intrinsics were not found in your system compiler. In this case, the LLVM API
2413 must be taken</a> to ensure that side exits and the like do not accidentally
2414 result in concurrent LLVM API calls.
2415 </p>
2416 </div>
2418 <!-- ======================================================================= -->
2421 </div>
2424 <p>
2426 to deallocate memory used for internal structures. This will also invoke
2430 </p>
2432 <p>
2433 Note that, if you use scope-based shutdown, you can use the
2435 destructor.
2436 </div>
2438 <!-- ======================================================================= -->
2441 </div>
2444 <p>
2446 initialization of static resources, such as the global type tables. Before the
2449 however, it uses double-checked locking to implement thread-safe lazy
2450 initialization.
2451 </p>
2453 <p>
2454 Note that, because no other threads are allowed to issue LLVM API calls before
2457 </p>
2459 <p>
2461 APIs provide access to the global lock used to implement the double-checked
2462 locking for lazy initialization. These should only be used internally to LLVM,
2463 and only if you know what you're doing!
2464 </p>
2465 </div>
2467 <!-- ======================================================================= -->
2470 </div>
2473 <p>
2475 to operate multiple, isolated instances of LLVM concurrently within the same
2476 address space. For instance, in a hypothetical compile-server, the compilation
2477 of an individual translation unit is conceptually independent from all the
2478 others, and it would be desirable to be able to compile incoming translation
2479 units concurrently on independent server threads. Fortunately,
2481 </p>
2483 <p>
2490 safe to compile on multiple threads simultaneously, as long as no two threads
2491 operate on entities within the same context.
2492 </p>
2494 <p>
2495 In practice, very few places in the API require the explicit specification of a
2498 other entities can determine what context they belong to by looking at their
2500 maintain this interface design.
2501 </p>
2503 <p>
2507 isolation is not a concern.
2508 </p>
2509 </div>
2511 <!-- ======================================================================= -->
2514 </div>
2517 <p>
2518 LLVM's "eager" JIT compiler is safe to use in threaded programs. Multiple
2521 run code output by the JIT concurrently. The user must still ensure that only
2523 might be modifying it. One way to do that is to always hold the JIT lock while
2527 </p>
2529 <p>When the JIT is configured to compile lazily (using
2532 updating call sites after a function is lazily-jitted. It's still possible to
2533 use the lazy JIT in a threaded program if you ensure that only one thread at a
2534 time can call any particular lazy stub and that the JIT lock guards any IR
2535 access, but we suggest using only the eager JIT in threaded programs.
2536 </p>
2537 </div>
2539 <!-- *********************************************************************** -->
2542 </div>
2543 <!-- *********************************************************************** -->
2546 <p>
2547 This section describes some of the advanced or obscure API's that most clients
2548 do not need to be aware of. These API's tend manage the inner workings of the
2549 LLVM system, and only need to be accessed in unusual circumstances.
2550 </p>
2551 </div>
2553 <!-- ======================================================================= -->
2556 </div>
2560 <p>
2561 The LLVM type system has a very simple goal: allow clients to compare types for
2562 structural equality with a simple pointer comparison (aka a shallow compare).
2563 This goal makes clients much simpler and faster, and is used throughout the LLVM
2564 system.
2565 </p>
2567 <p>
2568 Unfortunately achieving this goal is not a simple matter. In particular,
2569 recursive types and late resolution of opaque types makes the situation very
2570 difficult to handle. Fortunately, for the most part, our implementation makes
2571 most clients able to be completely unaware of the nasty internal details. The
2572 primary case where clients are exposed to the inner workings of it are when
2573 building a recursive type. In addition to this case, the LLVM bitcode reader,
2574 assembly parser, and linker also have to be aware of the inner workings of this
2575 system.
2576 </p>
2578 <p>
2579 For our purposes below, we need three concepts. First, an "Opaque Type" is
2582 opaque type as part of its type graph (for example "<tt>{ opaque, i32 }</tt>").
2583 Third, a concrete type is a type that is not an abstract type (e.g. "<tt>{ i32,
2584 float }</tt>").
2585 </p>
2587 </div>
2589 <!-- ______________________________________________________________________ -->
2592 </div>
2596 <p>
2597 Because the most common question is "how do I build a recursive type with LLVM",
2598 we answer it now and explain it as we go. Here we include enough to cause this
2599 to be emitted to an output .ll file:
2600 </p>
2603 <pre>
2604 %mylist = type { %mylist*, i32 }
2605 </pre>
2606 </div>
2608 <p>
2609 To build this, use the following LLVM APIs:
2610 </p>
2613 <pre>
2617 Elts.push_back(PointerType::getUnqual(StructTy));
2618 Elts.push_back(Type::Int32Ty);
2619 StructType *NewSTy = StructType::get(Elts);
2623 cast<OpaqueType>(StructTy.get())-><a href="#refineAbstractTypeTo">refineAbstractTypeTo</a>(NewSTy);
2625 // <i>NewSTy is potentially invalidated, but StructTy (a <a href="#PATypeHolder">PATypeHolder</a>) is</i>
2631 </pre>
2632 </div>
2634 <p>
2635 This code shows the basic approach used to build recursive types: build a
2636 non-recursive type using 'opaque', then use type unification to close the cycle.
2637 The type unification step is performed by the <tt><a
2639 described next. After that, we describe the <a
2641 </p>
2643 </div>
2645 <!-- ______________________________________________________________________ -->
2648 </div>
2651 <p>
2653 While this method is actually a member of the DerivedType class, it is most
2654 often used on OpaqueType instances. Type unification is actually a recursive
2655 process. After unification, types can become structurally isomorphic to
2656 existing types, and all duplicates are deleted (to preserve pointer equality).
2657 </p>
2659 <p>
2660 In the example above, the OpaqueType object is definitely deleted.
2661 Additionally, if there is an "{ \2*, i32}" type already created in the system,
2663 a type is deleted, any "Type*" pointers in the program are invalidated. As
2666 types can never move or be deleted). To deal with this, the <a
2668 reference to a possibly refined type, and the <a
2670 complex datastructures.
2671 </p>
2673 </div>
2675 <!-- ______________________________________________________________________ -->
2678 </div>
2681 <p>
2682 PATypeHolder is a form of a "smart pointer" for Type objects. When VMCore
2683 happily goes about nuking types that become isomorphic to existing types, it
2684 automatically updates all PATypeHolder objects to point to the new type. In the
2685 example above, this allows the code to maintain a pointer to the resultant
2686 resolved recursive type, even though the Type*'s are potentially invalidated.
2687 </p>
2689 <p>
2690 PATypeHolder is an extremely light-weight object that uses a lazy union-find
2691 implementation to update pointers. For example the pointer from a Value to its
2692 Type is maintained by PATypeHolder objects.
2693 </p>
2695 </div>
2697 <!-- ______________________________________________________________________ -->
2700 </div>
2704 <p>
2705 Some data structures need more to perform more complex updates when types get
2706 resolved. To support this, a class can derive from the AbstractTypeUser class.
2707 This class
2708 allows it to get callbacks when certain types are resolved. To register to get
2709 callbacks for a particular type, the DerivedType::{add/remove}AbstractTypeUser
2710 methods can be called on a type. Note that these methods only work for <i>
2711 abstract</i> types. Concrete types (those that do not include any opaque
2712 objects) can never be refined.
2713 </p>
2714 </div>
2717 <!-- ======================================================================= -->
2721 </div>
2731 names for types.</p>
2734 by most clients. It should only be used when iteration over the symbol table
2735 names themselves are required, which is very special purpose. Note that not
2736 all LLVM
2738 an empty name) do not exist in the symbol table.
2739 </p>
2741 <p>These symbol tables support iteration over the values/types in the symbol
2746 appropriate symbol table. For types, use the Module::addTypeName method to
2747 insert entries into the symbol table.</p>
2749 </div>
2753 <!-- ======================================================================= -->
2756 </div>
2764 addition and removal.</p>
2766 <!-- ______________________________________________________________________ -->
2768 <a name="Use2User">Interaction and relationship between <tt>User</tt> and <tt>Use</tt> objects</a>
2769 </div>
2772 <p>
2774 or refer to them out-of-line by means of a pointer. A mixed variant
2777 </p>
2778 </div>
2780 <p>
2782 <ul>
2783 <li><p>Layout a)
2785 object and there are a fixed number of them.</p>
2787 <li><p>Layout b)
2790 number of them.</p>
2791 </ul>
2792 <p>
2793 As of v2.4 each layout still possesses a direct pointer to the
2795 we stick to this redundancy for the sake of simplicity.
2797 has. (Theoretically this information can also be calculated
2798 given the scheme presented below.)</p>
2799 <p>
2801 enforce the following memory layouts:</p>
2803 <ul>
2804 <li><p>Layout a) is modelled by prepending the <tt>User</tt> object by the <tt>Use[]</tt> array.</p>
2806 <pre>
2807 ...---.---.---.---.-------...
2808 | P | P | P | P | User
2809 '''---'---'---'---'-------'''
2810 </pre>
2813 <pre>
2814 .-------...
2815 | User
2816 '-------'''
2817 |
2818 v
2819 .---.---.---.---...
2820 | P | P | P | P |
2821 '---'---'---'---'''
2822 </pre>
2823 </ul>
2827 <!-- ______________________________________________________________________ -->
2830 </div>
2833 <p>
2836 recover it. This is accomplished by the following scheme:</p>
2837 </div>
2839 A bit-encoding in the 2 LSBits (least significant bits) of the <tt>Use::Prev</tt> allows to find the
2841 <ul>
2846 </ul>
2847 <p>
2850 we have to walk to the next stop picking up digits
2851 and calculating the offset:</p>
2852 <pre>
2853 .---.---.---.---.---.---.---.---.---.---.---.---.---.---.---.---.----------------
2855 '---'---'---'---'---'---'---'---'---'---'---'---'---'---'---'---'----------------
2857 | | | | |__>
2858 | | | |__________>
2859 | | |______________________>
2860 | |______________________________________>
2861 |__________________________________________________________>
2862 </pre>
2863 <p>
2864 Only the significant number of bits need to be stored between the
2868 <!-- ______________________________________________________________________ -->
2871 </div>
2874 <p>
2875 The following literate Haskell fragment demonstrates the concept:</p>
2876 </div>
2879 <pre>
2880 > import Test.QuickCheck
2881 >
2886 >
2892 >
2893 > takeLast n ss = reverse $ take n $ reverse ss
2894 >
2896 >
2897 </pre>
2898 </div>
2899 <p>
2901 <p>
2902 The reverse algorithm computes the length of the string just by examining
2903 a certain prefix:</p>
2906 <pre>
2911 >
2914 > decode walk acc _ = walk + acc
2915 >
2916 </pre>
2917 </div>
2918 <p>
2920 <p>
2924 <pre>
2926 > testcaseLength = length testcase
2927 >
2929 > where arr = takeLast n testcase
2930 >
2931 </pre>
2932 </div>
2933 <p>
2936 <pre>
2937 *Main> quickCheck identityProp
2938 OK, passed 100 tests.
2939 </pre>
2940 <p>
2941 Let's be a bit more exhaustive:</p>
2944 <pre>
2945 >
2947 >
2948 </pre>
2949 </div>
2950 <p>
2953 <pre>
2954 *Main> deepCheck identityProp
2955 OK, passed 500 tests.
2956 </pre>
2958 <!-- ______________________________________________________________________ -->
2961 </div>
2963 <p>
2967 tag bits.</p>
2968 <p>
2972 the LSBit set. (Portability is relying on the fact that all known compilers place the
2975 </div>
2977 <!-- *********************************************************************** -->
2980 </div>
2981 <!-- *********************************************************************** -->
2987 <p>The Core LLVM classes are the primary means of representing the program
2988 being inspected or transformed. The core LLVM classes are defined in
2992 </div>
2994 <!-- ======================================================================= -->
2997 </div>
3005 subclasses. They are hidden because they offer no useful functionality beyond
3009 named, but this is not a requirement. There exists exactly
3010 one instance of a given shape at any one time. This allows type equality to
3011 be performed with address equality of the Type Instance. That is, given two
3013 </p>
3014 </div>
3016 <!-- _______________________________________________________________________ -->
3019 </div>
3023 <ul>
3027 floating point types.</li>
3030 an OpaqueType anywhere in its definition).</li>
3033 that don't have a size are abstract types, labels and void.</li>
3035 </ul>
3036 </div>
3038 <!-- _______________________________________________________________________ -->
3041 </div>
3043 <dl>
3045 <dd>Subclass of DerivedType that represents integer types of any bit width.
3048 <ul>
3050 type of a specific bit width.</li>
3052 type.</li>
3053 </ul>
3054 </dd>
3056 <dd>This is subclassed by ArrayType and PointerType
3057 <ul>
3059 of the elements in the sequential type. </li>
3060 </ul>
3061 </dd>
3063 <dd>This is a subclass of SequentialType and defines the interface for array
3064 types.
3065 <ul>
3067 elements in the array. </li>
3068 </ul>
3069 </dd>
3073 <dd>Subclass of SequentialType for vector types. A
3074 vector type is similar to an ArrayType but is distinguished because it is
3075 a first class type whereas ArrayType is not. Vector types are used for
3076 vector operations and are usually small vectors of of an integer or floating
3077 point type.</dd>
3081 <dd>Subclass of DerivedTypes for function types.
3082 <ul>
3084 function</li>
3086 return type of the function.</li>
3088 the type of the ith parameter.</li>
3090 number of formal parameters.</li>
3091 </ul>
3092 </dd>
3094 <dd>Sublcass of DerivedType for abstract types. This class
3095 defines no content and is used as a placeholder for some other type. Note
3096 that OpaqueType is used (temporarily) during type resolution for forward
3097 references of types. Once the referenced type is resolved, the OpaqueType
3098 is replaced with the actual type. OpaqueType can also be used for data
3099 abstraction. At link time opaque types can be resolved to actual types
3100 of the same name.</dd>
3101 </dl>
3102 </div>
3106 <!-- ======================================================================= -->
3109 </div>
3118 programs. An LLVM module is effectively either a translation unit of the
3119 original program or a combination of several translation units merged by the
3124 helpful member functions that try to make common operations easy.</p>
3126 </div>
3128 <!-- _______________________________________________________________________ -->
3131 </div>
3135 <ul>
3137 </ul>
3140 provide a name for it (probably based on the name of the translation unit).</p>
3142 <ul>
3149 <p>These are forwarding methods that make it easy to access the contents of
3151 list.</p></li>
3156 necessary to use when you need to update the list or perform a complex
3157 action that doesn't have a forwarding method.</p>
3160 </ul>
3162 <hr>
3164 <ul>
3172 <p> These are forwarding methods that make it easy to access the contents of
3180 use when you need to update the list or perform a complex action that
3181 doesn't have a forwarding method.</p>
3184 </ul>
3186 <hr>
3188 <ul>
3195 </ul>
3197 <hr>
3199 <ul>
3212 external declaration for the function and return it.</p></li>
3219 string.</p></li>
3226 name, true is returned and the <a
3228 </ul>
3230 </div>
3233 <!-- ======================================================================= -->
3236 </div>
3241 <br>
3245 base. It represents a typed value that may be used (among other things) as an
3253 for a program. For example, an incoming argument to a function (represented
3255 every instruction in the function that references the argument. To keep track
3261 methods, shown below.</p>
3265 method. In addition, all LLVM values can be named. The "name" of the
3269 <pre>
3271 </pre>
3272 </div>
3275 that the name of any value may be missing (an empty string), so names should
3277 debugging printouts), they should not be used to keep track of values or map
3281 <p>One important aspect of LLVM is that there is no distinction between an SSA
3282 variable and the operation that produces it. Because of this, any reference to
3283 the value produced by an instruction (or the value available as an incoming
3284 argument, for example) is represented as a direct pointer to the instance of
3285 the class that
3286 represents this value. Although this may take some getting used to, it
3287 simplifies the representation and makes it easier to manipulate.</p>
3289 </div>
3291 <!-- _______________________________________________________________________ -->
3294 </div>
3298 <ul>
3300 use-list<br>
3302 the use-list<br>
3304 value.<br>
3307 the use-list.<br>
3309 use-list.<br>
3311 element in the list.
3312 <p> These methods are the interface to access the def-use
3313 information in LLVM. As with all other iterators in LLVM, the naming
3315 </li>
3318 </li>
3324 </li>
3329 "<tt>V</tt>" instead. For example, if you detect that an instruction always
3330 produces a constant value (for example through constant folding), you can
3331 replace all uses of the instruction with the constant like this:</p>
3334 <pre>
3335 Inst->replaceAllUsesWith(ConstVal);
3336 </pre>
3337 </div>
3339 </ul>
3341 </div>
3343 <!-- ======================================================================= -->
3346 </div>
3350 <p>
3363 Single Assignment (SSA) form, there can only be one definition referred to,
3364 allowing this direct connection. This connection provides the use-def
3365 information in LLVM.</p>
3367 </div>
3369 <!-- _______________________________________________________________________ -->
3372 </div>
3377 an index access interface and through an iterator based interface.</p>
3379 <ul>
3383 convenient form for direct access.</p></li>
3386 list<br>
3388 the operand list.<br>
3390 operand list.
3391 <p> Together, these methods make up the iterator based interface to
3393 </ul>
3395 </div>
3397 <!-- ======================================================================= -->
3400 </div>
3411 instructions. It provides only a few methods, but is a very commonly used
3413 opcode (instruction type) and the parent <a
3415 into. To represent a specific type of instruction, one of many subclasses of
3424 file contains some meta-data about the various different types of instructions
3425 in LLVM. It describes the enum values that are used as opcodes (for example
3430 this file confuses doxygen, so these enum values don't show up correctly in the
3433 </div>
3435 <!-- _______________________________________________________________________ -->
3438 class</a>
3439 </div>
3441 <ul>
3443 <p>This subclasses represents all two operand instructions whose operands
3444 must be the same type, except for the comparison instructions.</p></li>
3447 common operations on cast instructions.</p>
3449 <p>This subclass respresents the two comparison instructions,
3453 <p>This subclass is the parent of all terminator instructions (those which
3454 can terminate a block).</p>
3455 </ul>
3456 </div>
3458 <!-- _______________________________________________________________________ -->
3461 class</a>
3462 </div>
3466 <ul>
3471 <p>Returns true if the instruction writes to memory, i.e. it is a
3476 <p>Returns another instance of the specified instruction, identical
3477 in all ways to the original except that the instruction has no parent
3479 and it has no name</p></li>
3480 </ul>
3482 </div>
3484 <!-- ======================================================================= -->
3487 </div>
3491 <p>Constant represents a base class for different types of constants. It
3492 is subclassed by ConstantInt, ConstantArray, etc. for representing
3494 a subclass, which represents the address of a global variable or function.
3495 </p>
3497 </div>
3499 <!-- _______________________________________________________________________ -->
3502 <ul>
3503 <li>ConstantInt : This subclass of Constant represents an integer constant of
3504 any width.
3505 <ul>
3507 value of this constant, an APInt value.</li>
3509 value to an int64_t via sign extension. If the value (not the bit width)
3510 of the APInt is too large to fit in an int64_t, an assertion will result.
3511 For this reason, use of this method is discouraged.</li>
3513 value to a uint64_t via zero extension. IF the value (not the bit width)
3514 of the APInt is too large to fit in a uint64_t, an assertion will result.
3515 For this reason, use of this method is discouraged.</li>
3518 The type is implied as the IntegerType that corresponds to the bit width
3521 Returns the ConstantInt object that represents the value provided by
3523 </ul>
3524 </li>
3525 <li>ConstantFP : This class represents a floating point constant.
3526 <ul>
3528 this constant. </li>
3529 </ul>
3530 </li>
3531 <li>ConstantArray : This represents a constant array.
3532 <ul>
3534 a vector of component constants that makeup this array. </li>
3535 </ul>
3536 </li>
3537 <li>ConstantStruct : This represents a constant struct.
3538 <ul>
3540 a vector of component constants that makeup this array. </li>
3541 </ul>
3542 </li>
3543 <li>GlobalValue : This represents either a global variable or a function. In
3544 either case, the value is a constant fixed address (after linking).
3545 </li>
3546 </ul>
3547 </div>
3550 <!-- ======================================================================= -->
3553 </div>
3560 Class</a><br>
3567 Because they are visible at global scope, they are also subject to linking with
3568 other globals defined in different translation units. To control the linking
3575 unit, and does not participate in linking. If it has external linkage, it is
3576 visible to external code, and does participate in linking. In addition to
3582 global is always a pointer to its contents. It is important to remember this
3587 the address of the first element of this array and the value of the
3595 </div>
3597 <!-- _______________________________________________________________________ -->
3600 class</a>
3601 </div>
3605 <ul>
3610 <p> </p>
3611 </li>
3614 GlobalValue is currently embedded into.</p></li>
3615 </ul>
3617 </div>
3619 <!-- ======================================================================= -->
3622 </div>
3635 actually one of the more complex classes in the LLVM hierarchy because it must
3643 ordering of the blocks in the function, which indicate how the code will be
3644 laid out by the backend. Additionally, the first <a
3647 block. There are no implicit exit nodes, and in fact there may be multiple exit
3651 function hasn't been linked in yet.</p>
3661 LLVM feature that is only used when you have to look up a value by name. Aside
3663 internally to make sure that there are not conflicts between the names of <a
3670 is its address (after linking) which is guaranteed to be constant.</p>
3671 </div>
3673 <!-- _______________________________________________________________________ -->
3676 class</a>
3677 </div>
3681 <ul>
3686 the the program. The constructor must specify the type of the function to
3687 create and what type of linkage the function should have. The <a
3689 specifies the formal arguments and return value for the function. The same
3692 in which the function is defined. If this argument is provided, the function
3693 will automatically be inserted into that module's list of
3694 functions.</p></li>
3699 function is "external", it does not have a body, and thus must be resolved
3700 by linking with a function defined in a different translation unit.</p></li>
3708 <p>These are forwarding methods that make it easy to access the contents of
3710 list.</p></li>
3715 is necessary to use when you need to update the list or perform a complex
3716 action that doesn't have a forwarding method.</p></li>
3719 iterator<br>
3725 <p>These are forwarding methods that make it easy to access the contents of
3727 list.</p></li>
3732 necessary to use when you need to update the list or perform a complex
3733 action that doesn't have a forwarding method.</p></li>
3738 function. Because the entry block for the function is always the first
3747 function.</p></li>
3753 </ul>
3755 </div>
3757 <!-- ======================================================================= -->
3760 </div>
3766 <br>
3768 Class</a><br>
3774 <p>Global variables are represented with the (surprise surprise)
3777 always referenced by their address (global values must live in memory, so their
3778 "name" refers to their constant address). See
3780 variables may have an initial value (which must be a
3782 they may be marked as "constant" themselves (indicating that their contents
3783 never change at runtime).</p>
3784 </div>
3786 <!-- _______________________________________________________________________ -->
3790 </div>
3794 <ul>
3799 <p>Create a new global variable of the specified type. If
3801 unchanging for the program. The Linkage parameter specifies the type of
3802 linkage (internal, external, weak, linkonce, appending) for the variable.
3803 If the linkage is InternalLinkage, WeakAnyLinkage, WeakODRLinkage,
3804 LinkOnceAnyLinkage or LinkOnceODRLinkage, then the resultant
3805 global variable will have internal linkage. AppendingLinkage concatenates
3806 together all instances (in different translation units) of the variable
3807 into a single variable but is only applicable to arrays. See
3809 further details on linkage types. Optionally an initializer, a name, and the
3810 module to put the variable into may be specified for the global variable as
3811 well.</p></li>
3815 <p>Returns true if this is a global variable that is known not to
3816 be modified at runtime.</p></li>
3825 to call this method if there is no initializer.</p></li>
3826 </ul>
3828 </div>
3831 <!-- ======================================================================= -->
3834 </div>
3841 Class</a><br>
3844 <p>This class represents a single entry multiple exit section of the code,
3845 commonly known as a basic block by the compiler community. The
3848 Matching the language definition, the last element of this list of instructions
3849 is always a terminator instruction (a subclass of the <a
3852 <p>In addition to tracking the list of instructions that make up the block, the
3861 </div>
3863 <!-- _______________________________________________________________________ -->
3866 class</a>
3867 </div>
3870 <ul>
3876 insertion into a function. The constructor optionally takes a name for the new
3879 automatically inserted at the end of the specified <a
3887 STL-style functions for accessing the instruction list.
3889 <p>These methods and typedefs are forwarding functions that have the same
3890 semantics as the standard library methods of the same names. These methods
3891 expose the underlying instruction list of a basic block in a way that is easy to
3892 manipulate. To get the full complement of container operations (including
3894 method.</p></li>
3898 <p>This method is used to get access to the underlying container that actually
3899 holds the Instructions. This method must be used when there isn't a forwarding
3901 to perform. Because there are no forwarding functions for "updating"
3902 operations, you need to use this if you want to update the contents of a
3908 embedded into, or a null pointer if it is homeless.</p></li>
3912 <p> Returns a pointer to the terminator instruction that appears at the end of
3914 instruction in the block is not a terminator, then a null pointer is
3915 returned.</p></li>
3917 </ul>
3919 </div>
3922 <!-- ======================================================================= -->
3925 </div>
3929 <p>This subclass of Value defines the interface for incoming formal
3930 arguments to a function. A Function maintains a list of its formal
3931 arguments. An argument has a pointer to the parent Function.</p>
3933 </div>
3935 <!-- *********************************************************************** -->
3936 <hr>
3937 <address>
3946 Last modified: $Date$
3947 </address>
3949 </body>
3950 </html>