| blob | c1edd2a60a6fd299cb2a2dc5f15096cafbef7320 |
3 <html>
4 <head>
8 </head>
9 <body>
11 <h1>
12 LLVM Programmer's Manual
13 </h1>
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>
69 </ul></li>
71 <ul>
81 </ul></li>
83 <ul>
93 </ul></li>
95 <!--<ul>
96 todo
99 <ul>
103 </ul></li>
104 </ul>
105 </li>
107 <ul>
109 <ul>
117 class pointer</a> </li>
119 complex example</a> </li>
121 the same way</a> </li>
123 use-def chains</a> </li>
125 successors of blocks</a></li>
126 </ul>
127 </li>
129 <ul>
136 </ul>
137 </li>
139 <!--
140 <li>Working with the Control Flow Graph
141 <ul>
142 <li>Accessing predecessors and successors of a <tt>BasicBlock</tt>
143 <li>
144 <li>
145 </ul>
146 -->
147 </ul>
148 </li>
151 <ul>
153 </a></li>
158 </ul>
159 </li>
162 <ul>
165 <li><a href="#UserLayout">The <tt>User</tt> and owned <tt>Use</tt> classes' memory layout</a></li>
166 </ul></li>
169 <ul>
173 <ul>
175 <ul>
178 <ul>
180 <ul>
183 </ul>
184 </li>
185 </ul>
186 </li>
187 </ul>
188 </li>
191 </ul>
192 </li>
193 </ul>
194 </li>
195 </ol>
204 </div>
206 <!-- *********************************************************************** -->
207 <h2>
209 </h2>
210 <!-- *********************************************************************** -->
212 <div>
214 <p>This document is meant to highlight some of the important classes and
215 interfaces available in the LLVM source-base. This manual is not
216 intended to explain what LLVM is, how it works, and what LLVM code looks
217 like. It assumes that you know the basics of LLVM and are interested
218 in writing transformations or otherwise analyzing or manipulating the
219 code.</p>
221 <p>This document should get you oriented so that you can find your
222 way in the continuously growing source code that makes up the LLVM
223 infrastructure. Note that this manual is not intended to serve as a
224 replacement for reading the source code, so if you think there should be
225 a method in one of these classes to do something, but it's not listed,
227 are provided to make this as easy as possible.</p>
229 <p>The first section of this document describes general information that is
230 useful to know when working in the LLVM infrastructure, and the second describes
231 the Core LLVM classes. In the future this manual will be extended with
232 information describing how to use extension libraries, such as dominator
233 information, CFG traversal routines, and useful utilities like the <tt><a
236 </div>
238 <!-- *********************************************************************** -->
239 <h2>
241 </h2>
242 <!-- *********************************************************************** -->
244 <div>
246 <p>This section contains general information that is useful if you are working
247 in the LLVM source-base, but that isn't specific to any particular API.</p>
249 <!-- ======================================================================= -->
250 <h3>
252 </h3>
254 <div>
256 <p>LLVM makes heavy use of the C++ Standard Template Library (STL),
257 perhaps much more than you are used to, or have seen before. Because of
258 this, you might want to do a little background reading in the
259 techniques used and capabilities of the library. There are many good
260 pages that discuss the STL, and several books on the subject that you
261 can get, so it will not be discussed in this document.</p>
265 <ol>
268 C++ Library reference</a> - an excellent reference for the STL and other parts
269 of the standard C++ library.</li>
272 O'Reilly book in the making. It has a decent Standard Library
273 Reference that rivals Dinkumware's, and is unfortunately no longer free since the
274 book has been published.</li>
277 Questions</a></li>
280 Contains a useful <a
285 Page</a></li>
289 the book).</a></li>
291 </ol>
295 to write maintainable code more than where to put your curly braces.</p>
297 </div>
299 <!-- ======================================================================= -->
300 <h3>
302 </h3>
304 <div>
306 <ol>
308 static and shared libraries across platforms</a></li>
309 </ol>
311 </div>
313 </div>
315 <!-- *********************************************************************** -->
316 <h2>
318 </h2>
319 <!-- *********************************************************************** -->
321 <div>
323 <p>Here we highlight some LLVM APIs that are generally useful and good to
324 know about when writing transformations.</p>
326 <!-- ======================================================================= -->
327 <h3>
330 </h3>
332 <div>
334 <p>The LLVM source-base makes extensive use of a custom form of RTTI.
336 operator, but they don't have some drawbacks (primarily stemming from
338 have a v-table). Because they are used so often, you must know what they
339 do and how they work. All of these templates are defined in the <a
341 file (note that you very rarely have to include this file directly).</p>
343 <dl>
347 "<tt>instanceof</tt>" operator. It returns true or false depending on whether
348 a reference or pointer points to an instance of the specified class. This can
349 be very useful for constraint checking of various sorts (example below).</p>
350 </dd>
355 converts a pointer or reference from a base class to a derived class, causing
356 an assertion failure if it is not really an instance of the right type. This
357 should be used in cases where you have some information that makes you believe
362 <pre>
364 if (isa<<a href="#Constant">Constant</a>>(V) || isa<<a href="#Argument">Argument</a>>(V) || isa<<a href="#GlobalValue">GlobalValue</a>>(V))
365 return true;
369 }
370 </pre>
371 </div>
375 operator.</p>
377 </dd>
382 It checks to see if the operand is of the specified type, and if so, returns a
383 pointer to it (this operator does not work with references). If the operand is
384 not of the correct type, a null pointer is returned. Thus, this works very
388 statement like this:</p>
391 <pre>
392 if (<a href="#AllocationInst">AllocationInst</a> *AI = dyn_cast<<a href="#AllocationInst">AllocationInst</a>>(Val)) {
394 }
395 </pre>
396 </div>
400 statement, which is very convenient.</p>
405 blocks to check for lots of different variants of classes. If you find
406 yourself wanting to do this, it is much cleaner and more efficient to use the
409 </dd>
415 argument (which it then propagates). This can sometimes be useful, allowing
416 you to combine several null checks into one.</p></dd>
422 as an argument (which it then propagates). This can sometimes be useful,
423 allowing you to combine several null checks into one.</p></dd>
425 </dl>
427 <p>These five templates can be used with any classes, whether they have a
428 v-table or not. To add support for these templates, you simply need to add
430 to. Describing this is currently outside the scope of this document, but there
431 are lots of examples in the LLVM source base.</p>
433 </div>
436 <!-- ======================================================================= -->
437 <h3>
440 </h3>
442 <div>
444 <p>Although LLVM generally does not do much string manipulation, we do have
445 several important APIs which take strings. Two important examples are the
446 Value class -- which has names for instructions, functions, etc. -- and the
447 StringMap class which is used extensively in LLVM and Clang.</p>
449 <p>These are generic classes, and they need to be able to accept strings which
450 may have embedded null characters. Therefore, they cannot simply take
452 clients to perform a heap allocation which is usually unnecessary. Instead,
454 passing strings efficiently.</p>
456 <!-- _______________________________________________________________________ -->
457 <h4>
459 </h4>
461 <div>
464 (a character array and a length) and supports the common operations available
467 <p>It can be implicitly constructed using a C style null-terminated string,
472 iterator find(StringRef Key);
473 </pre>
481 </pre>
487 for more information.</p>
490 pointers to external memory it is not generally safe to store an instance of the
491 class (unless you know that the external storage will not be freed). StringRef is
492 small and pervasive enough in LLVM that it should always be passed by value.</p>
494 </div>
496 <!-- _______________________________________________________________________ -->
497 <h4>
499 </h4>
501 <div>
504 strings. For example, a common LLVM paradigm is to name one instruction based on
505 the name of another instruction with a suffix, for example:</p>
508 <pre>
510 </pre>
511 </div>
515 which points to temporary (stack allocated) objects. Twines can be implicitly
516 constructed as the result of the plus operator applied to strings (i.e., a C
518 actual concatenation of strings until it is actually required, at which point
519 it can be efficiently rendered directly into a character array. This avoids
520 unnecessary heap allocation involved in constructing the temporary results of
521 string concatenation. See
523 for more information.</p>
526 and should almost never be stored or mentioned directly. They are intended
527 solely for use when defining a function which should be able to efficiently
528 accept concatenated strings.</p>
530 </div>
532 </div>
534 <!-- ======================================================================= -->
535 <h3>
537 </h3>
539 <div>
541 <p>Often when working on your pass you will put a bunch of debugging printouts
542 and other code into your pass. After you get it working, you want to remove
543 it, but you may need it again in the future (to work out new bugs that you run
544 across).</p>
546 <p> Naturally, because of this, you don't want to delete the debug printouts,
547 but you don't want them to always be noisy. A standard compromise is to comment
548 them out, allowing you to enable them if you need them in the future.</p>
552 this problem. Basically, you can put arbitrary code into the argument of the
557 <pre>
559 </pre>
560 </div>
565 <pre>
569 I am here!
570 </pre>
571 </div>
574 to not have to create "yet another" command line option for the debug output for
576 so they do not cause a performance impact at all (for the same reason, they
577 should also not contain side-effects!).</p>
580 enable or disable it directly in gdb. Just use "<tt>set DebugFlag=0</tt>" or
581 "<tt>set DebugFlag=1</tt>" from the gdb if the program is running. If the
582 program hasn't been started yet, you can always just run it with
585 <!-- _______________________________________________________________________ -->
586 <h4>
589 </h4>
591 <div>
595 generator). If you want to enable debug information with more fine-grained
597 option as follows:</p>
600 <pre>
601 #undef DEBUG_TYPE
603 #define DEBUG_TYPE "foo"
605 #undef DEBUG_TYPE
606 #define DEBUG_TYPE "bar"
608 #undef DEBUG_TYPE
609 #define DEBUG_TYPE ""
611 </pre>
612 </div>
617 <pre>
621 No debug type
622 'foo' debug type
623 'bar' debug type
624 No debug type (2)
626 'foo' debug type
628 'bar' debug type
629 </pre>
630 </div>
633 a file, to specify the debug type for the entire module (if you do this before
636 "bar", because there is no system in place to ensure that names do not
637 conflict. If two different modules use the same string, they will all be turned
638 on when the name is specified. This allows, for example, all debug information
640 even if the source lives in multiple files.</p>
644 statement. It takes an additional first parameter, which is the type to use. For
645 example, the preceding example could be written as:</p>
649 <pre>
654 </pre>
655 </div>
657 </div>
659 </div>
661 <!-- ======================================================================= -->
662 <h3>
664 option</a>
665 </h3>
667 <div>
672 keep track of what the LLVM compiler is doing and how effective various
673 optimizations are. It is useful to see what optimizations are contributing to
674 making a particular program run faster.</p>
676 <p>Often you may run your pass on some big program, and you're interested to see
677 how many times it makes a certain transformation. Although you can do this with
678 hand inspection, or some ad-hoc method, this is a real pain and not very useful
680 keep track of this information, and the calculated information is presented in a
681 uniform manner with the rest of the passes being executed.</p>
684 it are as follows:</p>
686 <ol>
690 <pre>
691 #define <a href="#DEBUG_TYPE">DEBUG_TYPE</a> "mypassname" <i>// This goes before any #includes.</i>
692 STATISTIC(NumXForms, "The # of times I did stuff");
693 </pre>
694 </div>
697 specified by the first argument. The pass name is taken from the DEBUG_TYPE
698 macro, and the description is taken from the second argument. The variable
704 <pre>
706 </pre>
707 </div>
709 </li>
710 </ol>
716 <pre>
719 </pre>
720 </div>
723 suite, it gives a report that looks like this:</p>
726 <pre>
727 7646 bitcodewriter - Number of normal instructions
728 725 bitcodewriter - Number of oversized instructions
729 129996 bitcodewriter - Number of bitcode bytes written
730 2817 raise - Number of insts DCEd or constprop'd
731 3213 raise - Number of cast-of-self removed
732 5046 raise - Number of expression trees converted
733 75 raise - Number of other getelementptr's formed
734 138 raise - Number of load/store peepholes
735 42 deadtypeelim - Number of unused typenames removed from symtab
736 392 funcresolve - Number of varargs functions resolved
737 27 globaldce - Number of global variables removed
738 2 adce - Number of basic blocks removed
739 134 cee - Number of branches revectored
740 49 cee - Number of setcc instruction eliminated
741 532 gcse - Number of loads removed
742 2919 gcse - Number of instructions removed
743 86 indvars - Number of canonical indvars added
744 87 indvars - Number of aux indvars removed
745 25 instcombine - Number of dead inst eliminate
746 434 instcombine - Number of insts combined
747 248 licm - Number of load insts hoisted
748 1298 licm - Number of insts hoisted to a loop pre-header
749 3 licm - Number of insts hoisted to multiple loop preds (bad, no loop pre-header)
750 75 mem2reg - Number of alloca's promoted
751 1444 cfgsimplify - Number of blocks simplified
752 </pre>
753 </div>
755 <p>Obviously, with so many optimizations, having a unified framework for this
756 stuff is very nice. Making your pass fit well into the framework makes it more
757 maintainable and useful.</p>
759 </div>
761 <!-- ======================================================================= -->
762 <h3>
764 </h3>
766 <div>
768 <p>Several of the important data structures in LLVM are graphs: for example
772 DAGs</a>. In many cases, while debugging various parts of the compiler, it is
773 nice to instantly visualize these graphs.</p>
775 <p>LLVM provides several callbacks that are available in a debug build to do
777 the current LLVM tool will pop up a window containing the CFG for the function
778 where each basic block is a node in the graph, and each node contains the
779 instructions in the block. Similarly, there also exists
784 up a window. Alternatively, you can sprinkle calls to these functions in your
785 code in places you want to debug.</p>
787 <p>Getting this to work requires a small amount of configuration. On Unix
789 toolkit, and make sure 'dot' and 'gv' are in your path. If you are running on
790 Mac OS/X, download and install the Mac OS/X <a
793 it) to your path. Once in your system and path are set up, rerun the LLVM
794 configure script and rebuild LLVM to enable this functionality.</p>
800 specified color (choices of colors can be found at <a
802 complex node attributes can be provided with <tt>call
805 Attributes</a>.) If you want to restart and clear all the current graph
808 <p>Note that graph visualization features are compiled out of Release builds
809 to reduce file size. This means that you need a Debug+Asserts or
810 Release+Asserts build to use these features.</p>
812 </div>
814 </div>
816 <!-- *********************************************************************** -->
817 <h2>
819 </h2>
820 <!-- *********************************************************************** -->
822 <div>
825 and we commonly use STL data structures. This section describes the trade-offs
826 you should consider when you pick one.</p>
828 <p>
829 The first step is a choose your own adventure: do you want a sequential
830 container, a set-like container, or a map-like container? The most important
831 thing when choosing a container is the algorithmic properties of how you plan to
832 access the container. Based on that, you should use:</p>
834 <ul>
836 of an value based on another value. Map-like containers also support
837 efficient queries for containment (whether a key is in the map). Map-like
838 containers generally do not support efficient reverse mapping (values to
839 keys). If you need that, use two maps. Some map-like containers also
840 support efficient iteration through the keys in sorted order. Map-like
841 containers are the most expensive sort, only use them if you need one of
842 these capabilities.</li>
845 stuff into a container that automatically eliminates duplicates. Some
846 set-like containers support efficient iteration through the elements in
847 sorted order. Set-like containers are more expensive than sequential
848 containers.
849 </li>
852 the most efficient way to add elements and keeps track of the order they are
853 added to the collection. They permit duplicates and support efficient
854 iteration, but do not support efficient look-up based on a key.
855 </li>
858 container or reference structure that is used for character or byte
859 arrays.</li>
862 perform set operations on sets of numeric id's, while automatically
863 eliminating duplicates. Bit containers require a maximum of 1 bit for each
864 identifier you want to store.
865 </li>
866 </ul>
868 <p>
869 Once the proper category of container is determined, you can fine tune the
870 memory use, constant factors, and cache behaviors of access by intelligently
871 picking a member of the category. Note that constant factors and cache behavior
872 can be a big deal. If you have a vector that usually only contains a few
873 elements (but could contain many), for example, it's much better to use
875 . Doing so avoids (relatively) expensive malloc/free calls, which dwarf the
876 cost of adding the elements to the container. </p>
878 <!-- ======================================================================= -->
879 <h3>
881 </h3>
883 <div>
884 There are a variety of sequential containers available for you, based on your
885 needs. Pick the first in this section that will do what you want.
887 <!-- _______________________________________________________________________ -->
888 <h4>
890 </h4>
892 <div>
893 <p>The llvm::ArrayRef class is the preferred class to use in an interface that
894 accepts a sequential list of elements in memory and just reads from them. By
895 taking an ArrayRef, the API can be passed a fixed size array, an std::vector,
896 an llvm::SmallVector and anything else that is contiguous in memory.
897 </p>
898 </div>
902 <!-- _______________________________________________________________________ -->
903 <h4>
905 </h4>
907 <div>
908 <p>Fixed size arrays are very simple and very fast. They are good if you know
909 exactly how many elements you have, or you have a (low) upper bound on how many
910 you have.</p>
911 </div>
913 <!-- _______________________________________________________________________ -->
914 <h4>
916 </h4>
918 <div>
919 <p>Heap allocated arrays (new[] + delete[]) are also simple. They are good if
920 the number of elements is variable, if you know how many elements you will need
921 before the array is allocated, and if the array is usually large (if not,
923 allocated array is the cost of the new/delete (aka malloc/free). Also note that
924 if you are allocating an array of a type with a constructor, the constructor and
925 destructors will be run for every element in the array (re-sizable vectors only
926 construct those elements actually used).</p>
927 </div>
929 <!-- _______________________________________________________________________ -->
930 <h4>
932 </h4>
934 <div>
937 it supports efficient iteration, lays out elements in memory order (so you can
938 do pointer arithmetic between elements), supports efficient push_back/pop_back
939 operations, supports efficient random access to its elements, etc.</p>
941 <p>The advantage of SmallVector is that it allocates space for
943 the SmallVector is dynamically smaller than N, no malloc is performed. This can
944 be a big win in cases where the malloc/free call is far more expensive than the
945 code that fiddles around with the elements.</p>
948 predecessors/successors of a block is usually less than 8). On the other hand,
949 this makes the size of the SmallVector itself large, so you don't want to
950 allocate lots of them (doing so will waste a lot of space). As such,
951 SmallVectors are most useful when on the stack.</p>
953 <p>SmallVector also provides a nice portable and efficient replacement for
956 </div>
958 <!-- _______________________________________________________________________ -->
959 <h4>
961 </h4>
963 <div>
964 <p>
965 std::vector is well loved and respected. It is useful when SmallVector isn't:
966 when the size of the vector is often large (thus the small optimization will
967 rarely be a benefit) or if you will be allocating many instances of the vector
968 itself (which would waste space for elements that aren't in the container).
969 vector is also useful when interfacing with code that expects vectors :).
970 </p>
975 <pre>
976 for ( ... ) {
978 use V;
979 }
980 </pre>
981 </div>
986 <pre>
988 for ( ... ) {
989 use V;
990 V.clear();
991 }
992 </pre>
993 </div>
995 <p>Doing so will save (at least) one heap allocation and free per iteration of
996 the loop.</p>
998 </div>
1000 <!-- _______________________________________________________________________ -->
1001 <h4>
1003 </h4>
1005 <div>
1006 <p>std::deque is, in some senses, a generalized version of std::vector. Like
1007 std::vector, it provides constant time random access and other similar
1008 properties, but it also provides efficient access to the front of the list. It
1009 does not guarantee continuity of elements within memory.</p>
1011 <p>In exchange for this extra flexibility, std::deque has significantly higher
1012 constant factor costs than std::vector. If possible, use std::vector or
1013 something cheaper.</p>
1014 </div>
1016 <!-- _______________________________________________________________________ -->
1017 <h4>
1019 </h4>
1021 <div>
1022 <p>std::list is an extremely inefficient class that is rarely useful.
1023 It performs a heap allocation for every element inserted into it, thus having an
1024 extremely high constant factor, particularly for small data types. std::list
1025 also only supports bidirectional iteration, not random access iteration.</p>
1027 <p>In exchange for this high cost, std::list supports efficient access to both
1028 ends of the list (like std::deque, but unlike std::vector or SmallVector). In
1029 addition, the iterator invalidation characteristics of std::list are stronger
1030 than that of a vector class: inserting or removing an element into the list does
1031 not invalidate iterator or pointers to other elements in the list.</p>
1032 </div>
1034 <!-- _______________________________________________________________________ -->
1035 <h4>
1037 </h4>
1039 <div>
1041 intrusive, because it requires the element to store and provide access to the
1042 prev/next pointers for the list.</p>
1046 provides some novel characteristics. In particular, it can efficiently store
1047 polymorphic objects, the traits class is informed when an element is inserted or
1049 constant-time splice operation.</p>
1051 <p>These properties are exactly what we want for things like
1055 Related classes of interest are explained in the following subsections:
1056 <ul>
1061 </ul>
1062 </div>
1064 <!-- _______________________________________________________________________ -->
1065 <h4>
1067 </h4>
1069 <div>
1070 <p>
1071 Useful for storing a vector of values using only a few number of bits for each
1072 value. Apart from the standard operations of a vector-like container, it can
1073 also perform an 'or' set operation.
1074 </p>
1079 <pre>
1080 enum State {
1081 None = 0x0,
1082 FirstCondition = 0x1,
1083 SecondCondition = 0x2,
1084 Both = 0x3
1085 };
1087 State get() {
1089 Vec1.push_back(FirstCondition);
1092 Vec2.push_back(SecondCondition);
1094 Vec1 |= Vec2;
1095 return Vec1[0]; // returns 'Both'.
1096 }
1097 </pre>
1098 </div>
1100 </div>
1102 <!-- _______________________________________________________________________ -->
1103 <h4>
1105 </h4>
1107 <div>
1110 publicly derive from this traits class.</p>
1111 </div>
1113 <!-- _______________________________________________________________________ -->
1114 <h4>
1116 </h4>
1118 <div>
1120 supports a slightly narrower interface. Notably, inserters from
1124 used for a wide variety of customizations.</p>
1125 </div>
1127 <!-- _______________________________________________________________________ -->
1128 <h4>
1130 </h4>
1132 <div>
1135 in the default manner.</p>
1140 </div>
1142 <!-- _______________________________________________________________________ -->
1143 <h4>
1145 </h4>
1147 <div>
1149 citizen in the C++ ecosystem, it needs to support the standard container
1154 <p>The only sensible solution to this problem is to allocate a so-called
1156 iterator, providing the back-link to the last element. However conforming to the
1158 also must not be dereferenced.</p>
1161 how to allocate and store the sentinel. The corresponding policy is dictated
1163 whenever the need for a sentinel arises.</p>
1165 <p>While the default policy is sufficient in most cases, it may break down when
1168 is wasted. To alleviate the situation with numerous and voluminous
1174 arithmetic is used to obtain the sentinel, which is relative to the
1176 extra pointer, which serves as the back-link of the sentinel. This is the only
1177 field in the ghostly sentinel which can be legally accessed.</p>
1178 </div>
1180 <!-- _______________________________________________________________________ -->
1181 <h4>
1183 </h4>
1185 <div>
1188 <p>There are also various STL adapter classes such as std::queue,
1189 std::priority_queue, std::stack, etc. These provide simplified access to an
1190 underlying container but don't affect the cost of the container itself.</p>
1192 </div>
1194 </div>
1196 <!-- ======================================================================= -->
1197 <h3>
1199 </h3>
1201 <div>
1203 <p>Set-like containers are useful when you need to canonicalize multiple values
1204 into a single representation. There are several different choices for how to do
1205 this, providing various trade-offs.</p>
1207 <!-- _______________________________________________________________________ -->
1208 <h4>
1210 </h4>
1212 <div>
1214 <p>If you intend to insert a lot of elements, then do a lot of queries, a
1215 great approach is to use a vector (or other sequential container) with
1216 std::sort+std::unique to remove duplicates. This approach works really well if
1217 your usage pattern has these two distinct phases (insert then query), and can be
1219 </p>
1221 <p>
1222 This combination provides the several nice properties: the result data is
1223 contiguous in memory (good for cache locality), has few allocations, is easy to
1224 address (iterators in the final vector are just indices or pointers), and can be
1225 efficiently queried with a standard binary or radix search.</p>
1227 </div>
1229 <!-- _______________________________________________________________________ -->
1230 <h4>
1232 </h4>
1234 <div>
1236 <p>If you have a set-like data structure that is usually small and whose elements
1238 has space for N elements in place (thus, if the set is dynamically smaller than
1239 N, no malloc traffic is required) and accesses them with a simple linear search.
1240 When the set grows beyond 'N' elements, it allocates a more expensive representation that
1241 guarantees efficient access (for most types, it falls back to std::set, but for
1242 pointers it uses something far better, <a
1245 <p>The magic of this class is that it handles small sets extremely efficiently,
1246 but gracefully handles extremely large sets without loss of efficiency. The
1247 drawback is that the interface is quite small: it supports insertion, queries
1248 and erasing, but does not support iteration.</p>
1250 </div>
1252 <!-- _______________________________________________________________________ -->
1253 <h4>
1255 </h4>
1257 <div>
1259 <p>SmallPtrSet has all the advantages of <tt>SmallSet</tt> (and a <tt>SmallSet</tt> of pointers is
1261 more than 'N' insertions are performed, a single quadratically
1262 probed hash table is allocated and grows as needed, providing extremely
1263 efficient access (constant time insertion/deleting/queries with low constant
1264 factors) and is very stingy with malloc traffic.</p>
1267 whenever an insertion occurs. Also, the values visited by the iterators are not
1268 visited in sorted order.</p>
1270 </div>
1272 <!-- _______________________________________________________________________ -->
1273 <h4>
1275 </h4>
1277 <div>
1279 <p>
1280 DenseSet is a simple quadratically probed hash table. It excels at supporting
1281 small values: it uses a single allocation to hold all of the pairs that
1282 are currently inserted in the set. DenseSet is a great way to unique small
1283 values that are not simple pointers (use <a
1285 the same requirements for the value type that <a
1287 </p>
1289 </div>
1291 <!-- _______________________________________________________________________ -->
1292 <h4>
1294 </h4>
1296 <div>
1298 <p>
1299 FoldingSet is an aggregate class that is really good at uniquing
1300 expensive-to-create or polymorphic objects. It is a combination of a chained
1301 hash table with intrusive links (uniqued objects are required to inherit from
1303 its ID process.</p>
1306 a complex object (for example, a node in the code generator). The client has a
1307 description of *what* it wants to generate (it knows the opcode and all the
1308 operands), but we don't want to 'new' a node, then try inserting it into a set
1309 only to find out it already exists, at which point we would have to delete it
1310 and return the node that already exists.
1311 </p>
1313 <p>To support this style of client, FoldingSet perform a query with a
1314 FoldingSetNodeID (which wraps SmallVector) that can be used to describe the
1315 element that we want to query for. The query either returns the element
1316 matching the ID or it returns an opaque ID that indicates where insertion should
1317 take place. Construction of the ID usually does not require heap traffic.</p>
1319 <p>Because FoldingSet uses intrusive links, it can support polymorphic objects
1320 in the set (for example, you can have SDNode instances mixed with LoadSDNodes).
1321 Because the elements are individually allocated, pointers to the elements are
1322 stable: inserting or removing elements does not invalidate any pointers to other
1323 elements.
1324 </p>
1326 </div>
1328 <!-- _______________________________________________________________________ -->
1329 <h4>
1331 </h4>
1333 <div>
1336 many things but great at nothing. std::set allocates memory for each element
1337 inserted (thus it is very malloc intensive) and typically stores three pointers
1338 per element in the set (thus adding a large amount of per-element space
1339 overhead). It offers guaranteed log(n) performance, which is not particularly
1340 fast from a complexity standpoint (particularly if the elements of the set are
1341 expensive to compare, like strings), and has extremely high constant factors for
1342 lookup, insertion and removal.</p>
1344 <p>The advantages of std::set are that its iterators are stable (deleting or
1345 inserting an element from the set does not affect iterators or pointers to other
1346 elements) and that iteration over the set is guaranteed to be in sorted order.
1347 If the elements in the set are large, then the relative overhead of the pointers
1348 and malloc traffic is not a big deal, but if the elements of the set are small,
1349 std::set is almost never a good choice.</p>
1351 </div>
1353 <!-- _______________________________________________________________________ -->
1354 <h4>
1356 </h4>
1358 <div>
1361 Container</a>. The important property
1362 that this provides is efficient insertion with uniquing (duplicate elements are
1363 ignored) with iteration support. It implements this by inserting elements into
1364 both a set-like container and the sequential container, using the set-like
1365 container for uniquing and the sequential container for iteration.
1366 </p>
1368 <p>The difference between SetVector and other sets is that the order of
1369 iteration is guaranteed to match the order of insertion into the SetVector.
1370 This property is really important for things like sets of pointers. Because
1371 pointer values are non-deterministic (e.g. vary across runs of the program on
1372 different machines), iterating over the pointers in the set will
1373 not be in a well-defined order.</p>
1375 <p>
1376 The drawback of SetVector is that it requires twice as much space as a normal
1377 set and has the sum of constant factors from the set-like container and the
1378 sequential container that it uses. Use it *only* if you need to iterate over
1379 the elements in a deterministic order. SetVector is also expensive to delete
1380 elements out of (linear time), unless you use it's "pop_back" method, which is
1381 faster.
1382 </p>
1384 <p>SetVector is an adapter class that defaults to using std::vector and std::set
1385 for the underlying containers, so it is quite expensive. However,
1387 defaults to using a SmallVector and SmallSet of a specified size. If you use
1388 this, and if your sets are dynamically smaller than N, you will save a lot of
1389 heap traffic.</p>
1391 </div>
1393 <!-- _______________________________________________________________________ -->
1394 <h4>
1396 </h4>
1398 <div>
1400 <p>
1402 retains a unique ID for each element inserted into the set. It internally
1403 contains a map and a vector, and it assigns a unique ID for each value inserted
1404 into the set.</p>
1406 <p>UniqueVector is very expensive: its cost is the sum of the cost of
1407 maintaining both the map and vector, it has high complexity, high constant
1408 factors, and produces a lot of malloc traffic. It should be avoided.</p>
1410 </div>
1413 <!-- _______________________________________________________________________ -->
1414 <h4>
1416 </h4>
1418 <div>
1420 <p>
1421 The STL provides several other options, such as std::multiset and the various
1422 "hash_set" like containers (whether from C++ TR1 or from the SGI library). We
1423 never use hash_set and unordered_set because they are generally very expensive
1424 (each insertion requires a malloc) and very non-portable.
1425 </p>
1427 <p>std::multiset is useful if you're not interested in elimination of
1428 duplicates, but has all the drawbacks of std::set. A sorted vector (where you
1429 don't delete duplicate entries) or some other approach is almost always
1430 better.</p>
1432 </div>
1434 </div>
1436 <!-- ======================================================================= -->
1437 <h3>
1439 </h3>
1441 <div>
1442 Map-like containers are useful when you want to associate data to a key. As
1443 usual, there are a lot of different ways to do this. :)
1445 <!-- _______________________________________________________________________ -->
1446 <h4>
1448 </h4>
1450 <div>
1452 <p>
1453 If your usage pattern follows a strict insert-then-query approach, you can
1455 for set-like containers</a>. The only difference is that your query function
1456 (which uses std::lower_bound to get efficient log(n) lookup) should only compare
1457 the key, not both the key and value. This yields the same advantages as sorted
1458 vectors for sets.
1459 </p>
1460 </div>
1462 <!-- _______________________________________________________________________ -->
1463 <h4>
1465 </h4>
1467 <div>
1469 <p>
1470 Strings are commonly used as keys in maps, and they are difficult to support
1471 efficiently: they are variable length, inefficient to hash and compare when
1472 long, expensive to copy, etc. StringMap is a specialized container designed to
1473 cope with these issues. It supports mapping an arbitrary range of bytes to an
1474 arbitrary other object.</p>
1476 <p>The StringMap implementation uses a quadratically-probed hash table, where
1477 the buckets store a pointer to the heap allocated entries (and some other
1478 stuff). The entries in the map must be heap allocated because the strings are
1479 variable length. The string data (key) and the element object (value) are
1480 stored in the same allocation with the string data immediately after the element
1481 object. This container guarantees the "<tt>(char*)(&Value+1)</tt>" points
1482 to the key string for a value.</p>
1484 <p>The StringMap is very fast for several reasons: quadratic probing is very
1485 cache efficient for lookups, the hash value of strings in buckets is not
1486 recomputed when looking up an element, StringMap rarely has to touch the
1487 memory for unrelated objects when looking up a value (even when hash collisions
1488 happen), hash table growth does not recompute the hash values for strings
1489 already in the table, and each pair in the map is store in a single allocation
1490 (the string data is stored in the same allocation as the Value of a pair).</p>
1492 <p>StringMap also provides query methods that take byte ranges, so it only ever
1493 copies a string if a value is inserted into the table.</p>
1494 </div>
1496 <!-- _______________________________________________________________________ -->
1497 <h4>
1499 </h4>
1501 <div>
1502 <p>
1503 IndexedMap is a specialized container for mapping small dense integers (or
1504 values that can be mapped to small dense integers) to some other type. It is
1505 internally implemented as a vector with a mapping function that maps the keys to
1506 the dense integer range.
1507 </p>
1509 <p>
1510 This is useful for cases like virtual registers in the LLVM code generator: they
1511 have a dense mapping that is offset by a compile-time constant (the first
1512 virtual register ID).</p>
1514 </div>
1516 <!-- _______________________________________________________________________ -->
1517 <h4>
1519 </h4>
1521 <div>
1523 <p>
1524 DenseMap is a simple quadratically probed hash table. It excels at supporting
1525 small keys and values: it uses a single allocation to hold all of the pairs that
1526 are currently inserted in the map. DenseMap is a great way to map pointers to
1527 pointers, or map other small types to each other.
1528 </p>
1530 <p>
1531 There are several aspects of DenseMap that you should be aware of, however. The
1532 iterators in a densemap are invalidated whenever an insertion occurs, unlike
1533 map. Also, because DenseMap allocates space for a large number of key/value
1534 pairs (it starts with 64 by default), it will waste a lot of space if your keys
1535 or values are large. Finally, you must implement a partial specialization of
1536 DenseMapInfo for the key that you want, if it isn't already supported. This
1537 is required to tell DenseMap about two special marker values (which can never be
1538 inserted into the map) that it needs internally.</p>
1540 </div>
1542 <!-- _______________________________________________________________________ -->
1543 <h4>
1545 </h4>
1547 <div>
1549 <p>
1551 Value*s (or subclasses) to another type. When a Value is deleted or RAUW'ed,
1552 ValueMap will update itself so the new version of the key is mapped to the same
1553 value, just as if the key were a WeakVH. You can configure exactly how this
1554 happens, and what else happens on these two events, by passing
1557 </div>
1559 <!-- _______________________________________________________________________ -->
1560 <h4>
1562 </h4>
1564 <div>
1566 <p> IntervalMap is a compact map for small keys and values. It maps key
1567 intervals instead of single keys, and it will automatically coalesce adjacent
1568 intervals. When then map only contains a few intervals, they are stored in the
1569 map object itself to avoid allocations.</p>
1571 <p> The IntervalMap iterators are quite big, so they should not be passed around
1572 as STL iterators. The heavyweight iterators allow a smaller data structure.</p>
1574 </div>
1576 <!-- _______________________________________________________________________ -->
1577 <h4>
1579 </h4>
1581 <div>
1583 <p>
1585 a single allocation per pair inserted into the map, it offers log(n) lookup with
1586 an extremely large constant factor, imposes a space penalty of 3 pointers per
1587 pair in the map, etc.</p>
1589 <p>std::map is most useful when your keys or values are very large, if you need
1590 to iterate over the collection in sorted order, or if you need stable iterators
1591 into the map (i.e. they don't get invalidated if an insertion or deletion of
1592 another element takes place).</p>
1594 </div>
1596 <!-- _______________________________________________________________________ -->
1597 <h4>
1599 </h4>
1601 <div>
1603 <p>IntEqClasses provides a compact representation of equivalence classes of
1605 equivalence class. Classes can be joined by passing two class representatives to
1606 the join(a, b) method. Two integers are in the same class when findLeader()
1607 returns the same representative.</p>
1609 <p>Once all equivalence classes are formed, the map can be compressed so each
1611 is the total number of equivalence classes. The map must be uncompressed before
1612 it can be edited again.</p>
1614 </div>
1616 <!-- _______________________________________________________________________ -->
1617 <h4>
1619 </h4>
1621 <div>
1623 <p>
1624 The STL provides several other options, such as std::multimap and the various
1625 "hash_map" like containers (whether from C++ TR1 or from the SGI library). We
1626 never use hash_set and unordered_set because they are generally very expensive
1627 (each insertion requires a malloc) and very non-portable.</p>
1629 <p>std::multimap is useful if you want to map a key to multiple values, but has
1630 all the drawbacks of std::map. A sorted vector or some other approach is almost
1631 always better.</p>
1633 </div>
1635 </div>
1637 <!-- ======================================================================= -->
1638 <h3>
1640 </h3>
1642 <div>
1644 <p>
1645 TODO: const char* vs stringref vs smallstring vs std::string. Describe twine,
1646 xref to #string_apis.
1647 </p>
1649 </div>
1651 <!-- ======================================================================= -->
1652 <h3>
1654 </h3>
1656 <div>
1657 <p>Unlike the other containers, there are only two bit storage containers, and
1658 choosing when to use each is relatively straightforward.</p>
1660 <p>One additional option is
1662 implementation in many common compilers (e.g. commonly available versions of
1663 GCC) is extremely inefficient and 2) the C++ standards committee is likely to
1664 deprecate this container and/or change it significantly somehow. In any case,
1665 please don't use it.</p>
1667 <!-- _______________________________________________________________________ -->
1668 <h4>
1670 </h4>
1672 <div>
1673 <p> The BitVector container provides a dynamic size set of bits for manipulation.
1674 It supports individual bit setting/testing, as well as set operations. The set
1675 operations take time O(size of bitvector), but operations are performed one word
1676 at a time, instead of one bit at a time. This makes the BitVector very fast for
1677 set operations compared to other containers. Use the BitVector when you expect
1678 the number of set bits to be high (IE a dense set).
1679 </p>
1680 </div>
1682 <!-- _______________________________________________________________________ -->
1683 <h4>
1685 </h4>
1687 <div>
1688 <p> The SmallBitVector container provides the same interface as BitVector, but
1689 it is optimized for the case where only a small number of bits, less than
1690 25 or so, are needed. It also transparently supports larger bit counts, but
1691 slightly less efficiently than a plain BitVector, so SmallBitVector should
1692 only be used when larger counts are rare.
1693 </p>
1695 <p>
1696 At this time, SmallBitVector does not support set operations (and, or, xor),
1697 and its operator[] does not provide an assignable lvalue.
1698 </p>
1699 </div>
1701 <!-- _______________________________________________________________________ -->
1702 <h4>
1704 </h4>
1706 <div>
1707 <p> The SparseBitVector container is much like BitVector, with one major
1708 difference: Only the bits that are set, are stored. This makes the
1709 SparseBitVector much more space efficient than BitVector when the set is sparse,
1710 as well as making set operations O(number of set bits) instead of O(size of
1711 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
1712 (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).
1713 </p>
1714 </div>
1716 </div>
1718 </div>
1720 <!-- *********************************************************************** -->
1721 <h2>
1723 </h2>
1724 <!-- *********************************************************************** -->
1726 <div>
1728 <p>This section describes how to perform some very simple transformations of
1729 LLVM code. This is meant to give examples of common idioms used, showing the
1731 you should also read about the main classes that you will be working with. The
1733 and descriptions of the main classes that you should know about.</p>
1735 <!-- NOTE: this section should be heavy on example code -->
1736 <!-- ======================================================================= -->
1737 <h3>
1739 </h3>
1741 <div>
1743 <p>The LLVM compiler infrastructure have many different data structures that may
1744 be traversed. Following the example of the C++ standard template library, the
1745 techniques used to traverse these various data structures are all basically the
1748 function returns an iterator pointing to one past the last valid element of the
1750 between the two operations.</p>
1752 <p>Because the pattern for iteration is common across many different aspects of
1753 the program representation, the standard template library algorithms may be used
1754 on them, and it is easier to remember how to iterate. First we show a few common
1755 examples of the data structures that need to be traversed. Other data
1756 structures are traversed in very similar ways.</p>
1758 <!-- _______________________________________________________________________ -->
1759 <h4>
1763 </h4>
1765 <div>
1768 transform in some way; in particular, you'd like to manipulate its
1775 <pre>
1782 </pre>
1783 </div>
1785 <p>Note that i can be used as if it were a pointer for the purposes of
1787 because the indirection operator is overloaded for the iterator
1791 </div>
1793 <!-- _______________________________________________________________________ -->
1794 <h4>
1798 </h4>
1800 <div>
1803 easy to iterate over the individual instructions that make up
1808 <pre>
1814 </pre>
1815 </div>
1817 <p>However, this isn't really the best way to print out the contents of a
1819 anything you'll care about, you could have just invoked the print routine on the
1822 </div>
1824 <!-- _______________________________________________________________________ -->
1825 <h4>
1829 </h4>
1831 <div>
1838 small example that shows how to dump all instructions in a function to the standard error stream:<p>
1841 <pre>
1845 for (inst_iterator I = inst_begin(F), E = inst_end(F); I != E; ++I)
1847 </pre>
1848 </div>
1851 work list with its initial contents. For example, if you wanted to
1853 F, all you would need to do is something like:</p>
1856 <pre>
1860 for (inst_iterator I = inst_begin(F), E = inst_end(F); I != E; ++I)
1861 worklist.insert(&*I);
1862 </pre>
1863 </div>
1868 </div>
1870 <!-- _______________________________________________________________________ -->
1871 <h4>
1873 vice-versa)</a>
1874 </h4>
1876 <div>
1878 <p>Sometimes, it'll be useful to grab a reference (or pointer) to a class
1879 instance when all you've got at hand is an iterator. Well, extracting
1880 a reference or a pointer from an iterator is very straight-forward.
1885 <pre>
1888 const Instruction& inst = *j;
1889 </pre>
1890 </div>
1892 <p>However, the iterators you'll be working with in the LLVM framework are
1893 special: they will automatically convert to a ptr-to-instance type whenever they
1894 need to. Instead of dereferencing the iterator and then taking the address of
1895 the result, you can simply assign the iterator to the proper pointer type and
1896 you get the dereference and address-of operation as a result of the assignment
1897 (behind the scenes, this is a result of overloading casting mechanisms). Thus
1898 the last line of the last example,</p>
1901 <pre>
1902 Instruction *pinst = &*i;
1903 </pre>
1904 </div>
1909 <pre>
1910 Instruction *pinst = i;
1911 </pre>
1912 </div>
1914 <p>It's also possible to turn a class pointer into the corresponding iterator,
1915 and this is a constant time operation (very efficient). The following code
1916 snippet illustrates use of the conversion constructors provided by LLVM
1917 iterators. By using these, you can explicitly grab the iterator of something
1918 without actually obtaining it via iteration over some structure:</p>
1921 <pre>
1922 void printNextInstruction(Instruction* inst) {
1923 BasicBlock::iterator it(inst);
1926 }
1927 </pre>
1928 </div>
1930 <p>Unfortunately, these implicit conversions come at a cost; they prevent
1931 these iterators from conforming to standard iterator conventions, and thus
1932 from being usable with standard algorithms and containers. For example, they
1934 from compiling:</p>
1937 <pre>
1939 </pre>
1940 </div>
1942 <p>Because of this, these implicit conversions may be removed some day,
1945 </div>
1947 <!--_______________________________________________________________________-->
1948 <h4>
1950 example</a>
1951 </h4>
1953 <div>
1955 <p>Say that you're writing a FunctionPass and would like to count all the
1959 much more straight-forward manner, but this example will allow us to explore how
1961 is what we want to do:</p>
1964 <pre>
1965 initialize callCounter to zero
1966 for each Function f in the Module
1967 for each BasicBlock b in f
1968 for each Instruction i in b
1969 if (i is a CallInst and calls the given function)
1970 increment callCounter
1971 </pre>
1972 </div>
1974 <p>And the actual code is (remember, because we're writing a
1979 <pre>
1980 Function* targetFunc = ...;
1982 class OurFunctionPass : public FunctionPass {
1983 public:
1984 OurFunctionPass(): callCounter(0) { }
1986 virtual runOnFunction(Function& F) {
1987 for (Function::iterator b = F.begin(), be = F.end(); b != be; ++b) {
1994 if (callInst->getCalledFunction() == targetFunc)
1995 ++callCounter;
1996 }
1997 }
1998 }
1999 }
2001 private:
2002 unsigned callCounter;
2003 };
2004 </pre>
2005 </div>
2007 </div>
2009 <!--_______________________________________________________________________-->
2010 <h4>
2012 </h4>
2014 <div>
2016 <p>You may have noticed that the previous example was a bit oversimplified in
2017 that it did not deal with call sites generated by 'invoke' instructions. In
2018 this, and in other situations, you may find that you want to treat
2021 less closely-related things. For these cases, LLVM provides a handy wrapper
2022 class called <a
2029 reference and it should not be dynamically allocated or deallocated using
2031 assignable and constructable, with costs equivalents to that of a bare pointer.
2032 If you look at its definition, it has only a single pointer member.</p>
2034 </div>
2036 <!--_______________________________________________________________________-->
2037 <h4>
2039 </h4>
2041 <div>
2053 <pre>
2054 Function *F = ...;
2060 }
2061 </pre>
2062 </div>
2066 doing it only once in the loop body and reusing its result.</p>
2073 all of the values that a particular instruction uses (that is, the operands of
2077 <pre>
2078 Instruction *pi = ...;
2081 Value *v = *i;
2083 }
2084 </pre>
2085 </div>
2088 mutation free algorithms (such as analyses, etc.). For this purpose above
2095 </div>
2097 <!--_______________________________________________________________________-->
2098 <h4>
2100 successors of blocks</a>
2101 </h4>
2103 <div>
2105 <p>Iterating over the predecessors and successors of a block is quite easy
2107 this to iterate over all predecessors of BB:</p>
2110 <pre>
2111 #include "llvm/Support/CFG.h"
2112 BasicBlock *BB = ...;
2114 for (pred_iterator PI = pred_begin(BB), E = pred_end(BB); PI != E; ++PI) {
2115 BasicBlock *Pred = *PI;
2117 }
2118 </pre>
2119 </div>
2121 <p>Similarly, to iterate over successors use
2122 succ_iterator/succ_begin/succ_end.</p>
2124 </div>
2126 </div>
2128 <!-- ======================================================================= -->
2129 <h3>
2131 </h3>
2133 <div>
2135 <p>There are some primitive transformation operations present in the LLVM
2136 infrastructure that are worth knowing about. When performing
2137 transformations, it's fairly common to manipulate the contents of basic
2138 blocks. This section describes some of the common methods for doing so
2139 and gives example code.</p>
2141 <!--_______________________________________________________________________-->
2142 <h4>
2145 </h4>
2147 <div>
2152 constructor for the kind of instruction to instantiate and provide the necessary
2157 <pre>
2158 AllocaInst* ai = new AllocaInst(Type::Int32Ty);
2159 </pre>
2160 </div>
2164 subclass is likely to have varying default parameters which change the semantics
2165 of the instruction, so refer to the <a
2171 <p>It is very useful to name the values of instructions when you're able to, as
2172 this facilitates the debugging of your transformations. If you end up looking
2173 at generated LLVM machine code, you definitely want to have logical names
2174 associated with the results of instructions! By supplying a value for the
2176 associate a logical name with the result of the instruction's execution at
2177 run time. For example, say that I'm writing a transformation that dynamically
2178 allocates space for an integer on the stack, and that integer is going to be
2179 used as some kind of index by some other code. To accomplish this, I place an
2185 <pre>
2187 </pre>
2188 </div>
2191 execution value, which is a pointer to an integer on the run time stack.</p>
2198 <ul>
2199 <li>Insertion into an explicit instruction list
2206 <pre>
2207 BasicBlock *pb = ...;
2208 Instruction *pi = ...;
2209 Instruction *newInst = new Instruction(...);
2212 </pre>
2213 </div>
2217 classes provide constructors which take a pointer to a
2219 looked like: </p>
2222 <pre>
2223 BasicBlock *pb = ...;
2224 Instruction *newInst = new Instruction(...);
2227 </pre>
2228 </div>
2233 <pre>
2234 BasicBlock *pb = ...;
2235 Instruction *newInst = new Instruction(..., pb);
2236 </pre>
2237 </div>
2239 <p>which is much cleaner, especially if you are creating
2240 long instruction streams.</p></li>
2242 <li>Insertion into an implicit instruction list
2245 are implicitly associated with an existing instruction list: the instruction
2246 list of the enclosing basic block. Thus, we could have accomplished the same
2248 </p>
2251 <pre>
2252 Instruction *pi = ...;
2253 Instruction *newInst = new Instruction(...);
2256 </pre>
2257 </div>
2259 <p>In fact, this sequence of steps occurs so frequently that the
2261 constructors which take (as a default parameter) a pointer to an
2265 provided instruction, immediately before that instruction. Using an
2267 parameter, the above code becomes:</p>
2270 <pre>
2271 Instruction* pi = ...;
2272 Instruction* newInst = new Instruction(..., pi);
2273 </pre>
2274 </div>
2276 <p>which is much cleaner, especially if you're creating a lot of
2278 </ul>
2280 </div>
2282 <!--_______________________________________________________________________-->
2283 <h4>
2285 </h4>
2287 <div>
2289 <p>Deleting an instruction from an existing sequence of instructions that form a
2291 call the instruction's eraseFromParent() method. For example:</p>
2294 <pre>
2296 I->eraseFromParent();
2297 </pre>
2298 </div>
2300 <p>This unlinks the instruction from its containing basic block and deletes
2301 it. If you'd just like to unlink the instruction from its containing basic
2304 </div>
2306 <!--_______________________________________________________________________-->
2307 <h4>
2310 </h4>
2312 <div>
2316 <p>Including "<a href="/doxygen/BasicBlockUtils_8h-source.html">llvm/Transforms/Utils/BasicBlockUtils.h</a>"
2322 <ul>
2325 <p>This function replaces all uses of a given instruction with a value,
2326 and then removes the original instruction. The following example
2327 illustrates the replacement of the result of a particular
2329 pointer to an integer.</p>
2332 <pre>
2333 AllocaInst* instToReplace = ...;
2334 BasicBlock::iterator ii(instToReplace);
2337 Constant::getNullValue(PointerType::getUnqual(Type::Int32Ty)));
2338 </pre></div></li>
2342 <p>This function replaces a particular instruction with another
2343 instruction, inserting the new instruction into the basic block at the
2344 location where the old instruction was, and replacing any uses of the old
2345 instruction with the new instruction. The following example illustrates
2349 <pre>
2350 AllocaInst* instToReplace = ...;
2351 BasicBlock::iterator ii(instToReplace);
2355 </pre></div></li>
2356 </ul>
2364 information.</p>
2366 <!-- Value::replaceAllUsesWith User::replaceUsesOfWith Point out:
2367 include/llvm/Transforms/Utils/ especially BasicBlockUtils.h with:
2368 ReplaceInstWithValue, ReplaceInstWithInst -->
2370 </div>
2372 <!--_______________________________________________________________________-->
2373 <h4>
2375 </h4>
2377 <div>
2379 <p>Deleting a global variable from a module is just as easy as deleting an
2380 Instruction. First, you must have a pointer to the global variable that you wish
2381 to delete. You use this pointer to erase it from its parent, the module.
2382 For example:</p>
2385 <pre>
2388 GV->eraseFromParent();
2389 </pre>
2390 </div>
2392 </div>
2394 </div>
2396 <!-- ======================================================================= -->
2397 <h3>
2399 </h3>
2401 <div>
2403 <p>In generating IR, you may need some complex types. If you know these types
2406 has two forms depending on whether you're building types for cross-compilation
2409 out of types from
2411 namespace and pointers, functions, arrays, etc. built of
2413 whose size may depend on the host compiler. For example,</p>
2416 <pre>
2418 </pre>
2419 </div>
2424 <pre>
2426 params.push_back(PointerType::getUnqual(Type::Int32Ty));
2427 FunctionType *ft = FunctionType::get(Type::Int8Ty, params, false);
2428 </pre>
2429 </div>
2434 </div>
2436 </div>
2438 <!-- *********************************************************************** -->
2439 <h2>
2441 </h2>
2442 <!-- *********************************************************************** -->
2444 <div>
2445 <p>
2446 This section describes the interaction of the LLVM APIs with multithreading,
2447 both on the part of client applications, and in the JIT, in the hosted
2448 application.
2449 </p>
2451 <p>
2452 Note that LLVM's support for multithreading is still relatively young. Up
2453 through version 2.5, the execution of threaded hosted applications was
2454 supported, but not threaded client access to the APIs. While this use case is
2456 ensure proper operation in multithreaded mode.
2457 </p>
2459 <p>
2460 Note that, on Unix-like platforms, LLVM requires the presence of GCC's atomic
2461 intrinsics in order to support threaded operation. If you need a
2462 multhreading-capable LLVM on a platform without a suitably modern system
2463 compiler, consider compiling LLVM and LLVM-GCC in single-threaded mode, and
2464 using the resultant compiler to build a copy of LLVM with multithreading
2465 support.
2466 </p>
2468 <!-- ======================================================================= -->
2469 <h3>
2471 </h3>
2473 <div>
2475 <p>
2476 In order to properly protect its internal data structures while avoiding
2477 excessive locking overhead in the single-threaded case, the LLVM must intialize
2478 certain data structures necessary to provide guards around its internals. To do
2480 making any concurrent LLVM API calls. To subsequently tear down these
2483 mode.
2484 </p>
2486 <p>
2488 say that no other LLVM API calls may be executing at any time during the
2490 </tt>. It's is the client's responsibility to enforce this isolation.
2491 </p>
2493 <p>
2495 failure of the initialization. Failure typically indicates that your copy of
2496 LLVM was built without multithreading support, typically because GCC atomic
2497 intrinsics were not found in your system compiler. In this case, the LLVM API
2500 must be taken</a> to ensure that side exits and the like do not accidentally
2501 result in concurrent LLVM API calls.
2502 </p>
2503 </div>
2505 <!-- ======================================================================= -->
2506 <h3>
2508 </h3>
2510 <div>
2511 <p>
2513 to deallocate memory used for internal structures. This will also invoke
2517 </p>
2519 <p>
2520 Note that, if you use scope-based shutdown, you can use the
2522 destructor.
2523 </div>
2525 <!-- ======================================================================= -->
2526 <h3>
2528 </h3>
2530 <div>
2531 <p>
2533 initialization of static resources, such as the global type tables. Before the
2536 however, it uses double-checked locking to implement thread-safe lazy
2537 initialization.
2538 </p>
2540 <p>
2541 Note that, because no other threads are allowed to issue LLVM API calls before
2544 </p>
2546 <p>
2548 APIs provide access to the global lock used to implement the double-checked
2549 locking for lazy initialization. These should only be used internally to LLVM,
2550 and only if you know what you're doing!
2551 </p>
2552 </div>
2554 <!-- ======================================================================= -->
2555 <h3>
2557 </h3>
2559 <div>
2560 <p>
2562 to operate multiple, isolated instances of LLVM concurrently within the same
2563 address space. For instance, in a hypothetical compile-server, the compilation
2564 of an individual translation unit is conceptually independent from all the
2565 others, and it would be desirable to be able to compile incoming translation
2566 units concurrently on independent server threads. Fortunately,
2568 </p>
2570 <p>
2577 safe to compile on multiple threads simultaneously, as long as no two threads
2578 operate on entities within the same context.
2579 </p>
2581 <p>
2582 In practice, very few places in the API require the explicit specification of a
2585 other entities can determine what context they belong to by looking at their
2587 maintain this interface design.
2588 </p>
2590 <p>
2594 isolation is not a concern.
2595 </p>
2596 </div>
2598 <!-- ======================================================================= -->
2599 <h3>
2601 </h3>
2603 <div>
2604 <p>
2605 LLVM's "eager" JIT compiler is safe to use in threaded programs. Multiple
2608 run code output by the JIT concurrently. The user must still ensure that only
2610 might be modifying it. One way to do that is to always hold the JIT lock while
2614 </p>
2616 <p>When the JIT is configured to compile lazily (using
2619 updating call sites after a function is lazily-jitted. It's still possible to
2620 use the lazy JIT in a threaded program if you ensure that only one thread at a
2621 time can call any particular lazy stub and that the JIT lock guards any IR
2622 access, but we suggest using only the eager JIT in threaded programs.
2623 </p>
2624 </div>
2626 </div>
2628 <!-- *********************************************************************** -->
2629 <h2>
2631 </h2>
2632 <!-- *********************************************************************** -->
2634 <div>
2635 <p>
2636 This section describes some of the advanced or obscure API's that most clients
2637 do not need to be aware of. These API's tend manage the inner workings of the
2638 LLVM system, and only need to be accessed in unusual circumstances.
2639 </p>
2642 <!-- ======================================================================= -->
2643 <h3>
2645 </h3>
2647 <div>
2653 </p>
2656 by most clients. It should only be used when iteration over the symbol table
2657 names themselves are required, which is very special purpose. Note that not
2658 all LLVM
2660 an empty name) do not exist in the symbol table.
2661 </p>
2663 <p>Symbol tables support iteration over the values in the symbol
2668 appropriate symbol table.</p>
2670 </div>
2674 <!-- ======================================================================= -->
2675 <h3>
2677 </h3>
2679 <div>
2685 addition and removal.</p>
2687 <!-- ______________________________________________________________________ -->
2688 <h4>
2691 </a>
2692 </h4>
2694 <div>
2695 <p>
2697 or refer to them out-of-line by means of a pointer. A mixed variant
2700 </p>
2702 <p>
2704 <ul>
2705 <li><p>Layout a)
2707 object and there are a fixed number of them.</p>
2709 <li><p>Layout b)
2712 number of them.</p>
2713 </ul>
2714 <p>
2715 As of v2.4 each layout still possesses a direct pointer to the
2717 we stick to this redundancy for the sake of simplicity.
2719 has. (Theoretically this information can also be calculated
2720 given the scheme presented below.)</p>
2721 <p>
2723 enforce the following memory layouts:</p>
2725 <ul>
2726 <li><p>Layout a) is modelled by prepending the <tt>User</tt> object by the <tt>Use[]</tt> array.</p>
2728 <pre>
2729 ...---.---.---.---.-------...
2730 | P | P | P | P | User
2731 '''---'---'---'---'-------'''
2732 </pre>
2735 <pre>
2736 .-------...
2737 | User
2738 '-------'''
2739 |
2740 v
2741 .---.---.---.---...
2742 | P | P | P | P |
2743 '---'---'---'---'''
2744 </pre>
2745 </ul>
2749 </div>
2751 <!-- ______________________________________________________________________ -->
2752 <h4>
2754 </h4>
2756 <div>
2757 <p>
2760 recover it. This is accomplished by the following scheme:</p>
2762 A bit-encoding in the 2 LSBits (least significant bits) of the <tt>Use::Prev</tt> allows to find the
2764 <ul>
2769 </ul>
2770 <p>
2773 we have to walk to the next stop picking up digits
2774 and calculating the offset:</p>
2775 <pre>
2776 .---.---.---.---.---.---.---.---.---.---.---.---.---.---.---.---.----------------
2778 '---'---'---'---'---'---'---'---'---'---'---'---'---'---'---'---'----------------
2780 | | | | |__>
2781 | | | |__________>
2782 | | |______________________>
2783 | |______________________________________>
2784 |__________________________________________________________>
2785 </pre>
2786 <p>
2787 Only the significant number of bits need to be stored between the
2791 </div>
2793 <!-- ______________________________________________________________________ -->
2794 <h4>
2796 </h4>
2798 <div>
2799 <p>
2800 The following literate Haskell fragment demonstrates the concept:</p>
2803 <pre>
2804 > import Test.QuickCheck
2805 >
2810 >
2816 >
2817 > takeLast n ss = reverse $ take n $ reverse ss
2818 >
2820 >
2821 </pre>
2822 </div>
2823 <p>
2825 <p>
2826 The reverse algorithm computes the length of the string just by examining
2827 a certain prefix:</p>
2830 <pre>
2835 >
2838 > decode walk acc _ = walk + acc
2839 >
2840 </pre>
2841 </div>
2842 <p>
2844 <p>
2848 <pre>
2850 > testcaseLength = length testcase
2851 >
2853 > where arr = takeLast n testcase
2854 >
2855 </pre>
2856 </div>
2857 <p>
2860 <pre>
2861 *Main> quickCheck identityProp
2862 OK, passed 100 tests.
2863 </pre>
2864 <p>
2865 Let's be a bit more exhaustive:</p>
2868 <pre>
2869 >
2871 >
2872 </pre>
2873 </div>
2874 <p>
2877 <pre>
2878 *Main> deepCheck identityProp
2879 OK, passed 500 tests.
2880 </pre>
2882 </div>
2884 <!-- ______________________________________________________________________ -->
2885 <h4>
2887 </h4>
2889 <div>
2891 <p>
2895 tag bits.</p>
2896 <p>
2900 the LSBit set. (Portability is relying on the fact that all known compilers place the
2903 </div>
2905 </div>
2907 </div>
2909 <!-- *********************************************************************** -->
2910 <h2>
2912 </h2>
2913 <!-- *********************************************************************** -->
2915 <div>
2919 <p>The Core LLVM classes are the primary means of representing the program
2920 being inspected or transformed. The core LLVM classes are defined in
2924 <!-- ======================================================================= -->
2925 <h3>
2927 </h3>
2929 <div>
2935 subclasses. They are hidden because they offer no useful functionality beyond
2939 named, but this is not a requirement. There exists exactly
2940 one instance of a given shape at any one time. This allows type equality to
2941 be performed with address equality of the Type Instance. That is, given two
2943 </p>
2945 <!-- _______________________________________________________________________ -->
2946 <h4>
2948 </h4>
2950 <div>
2952 <ul>
2956 floating point types.</li>
2959 that don't have a size are abstract types, labels and void.</li>
2961 </ul>
2962 </div>
2964 <!-- _______________________________________________________________________ -->
2965 <h4>
2967 </h4>
2968 <div>
2969 <dl>
2971 <dd>Subclass of DerivedType that represents integer types of any bit width.
2974 <ul>
2976 type of a specific bit width.</li>
2978 type.</li>
2979 </ul>
2980 </dd>
2982 <dd>This is subclassed by ArrayType and PointerType
2983 <ul>
2985 of the elements in the sequential type. </li>
2986 </ul>
2987 </dd>
2989 <dd>This is a subclass of SequentialType and defines the interface for array
2990 types.
2991 <ul>
2993 elements in the array. </li>
2994 </ul>
2995 </dd>
2999 <dd>Subclass of SequentialType for vector types. A
3000 vector type is similar to an ArrayType but is distinguished because it is
3001 a first class type whereas ArrayType is not. Vector types are used for
3002 vector operations and are usually small vectors of of an integer or floating
3003 point type.</dd>
3007 <dd>Subclass of DerivedTypes for function types.
3008 <ul>
3010 function</li>
3012 return type of the function.</li>
3014 the type of the ith parameter.</li>
3016 number of formal parameters.</li>
3017 </ul>
3018 </dd>
3019 </dl>
3020 </div>
3022 </div>
3024 <!-- ======================================================================= -->
3025 <h3>
3027 </h3>
3029 <div>
3036 programs. An LLVM module is effectively either a translation unit of the
3037 original program or a combination of several translation units merged by the
3042 helpful member functions that try to make common operations easy.</p>
3044 <!-- _______________________________________________________________________ -->
3045 <h4>
3047 </h4>
3049 <div>
3051 <ul>
3053 </ul>
3056 provide a name for it (probably based on the name of the translation unit).</p>
3058 <ul>
3065 <p>These are forwarding methods that make it easy to access the contents of
3067 list.</p></li>
3072 necessary to use when you need to update the list or perform a complex
3073 action that doesn't have a forwarding method.</p>
3076 </ul>
3078 <hr>
3080 <ul>
3088 <p> These are forwarding methods that make it easy to access the contents of
3096 use when you need to update the list or perform a complex action that
3097 doesn't have a forwarding method.</p>
3100 </ul>
3102 <hr>
3104 <ul>
3111 </ul>
3113 <hr>
3115 <ul>
3128 external declaration for the function and return it.</p></li>
3135 string.</p></li>
3142 name, true is returned and the <a
3144 </ul>
3146 </div>
3148 </div>
3150 <!-- ======================================================================= -->
3151 <h3>
3153 </h3>
3155 <div>
3158 <br>
3162 base. It represents a typed value that may be used (among other things) as an
3170 for a program. For example, an incoming argument to a function (represented
3172 every instruction in the function that references the argument. To keep track
3178 methods, shown below.</p>
3182 method. In addition, all LLVM values can be named. The "name" of the
3186 <pre>
3188 </pre>
3189 </div>
3192 that the name of any value may be missing (an empty string), so names should
3194 debugging printouts), they should not be used to keep track of values or map
3198 <p>One important aspect of LLVM is that there is no distinction between an SSA
3199 variable and the operation that produces it. Because of this, any reference to
3200 the value produced by an instruction (or the value available as an incoming
3201 argument, for example) is represented as a direct pointer to the instance of
3202 the class that
3203 represents this value. Although this may take some getting used to, it
3204 simplifies the representation and makes it easier to manipulate.</p>
3206 <!-- _______________________________________________________________________ -->
3207 <h4>
3209 </h4>
3211 <div>
3213 <ul>
3215 use-list<br>
3217 the use-list<br>
3219 value.<br>
3222 the use-list.<br>
3224 use-list.<br>
3226 element in the list.
3227 <p> These methods are the interface to access the def-use
3228 information in LLVM. As with all other iterators in LLVM, the naming
3230 </li>
3233 </li>
3239 </li>
3244 "<tt>V</tt>" instead. For example, if you detect that an instruction always
3245 produces a constant value (for example through constant folding), you can
3246 replace all uses of the instruction with the constant like this:</p>
3249 <pre>
3250 Inst->replaceAllUsesWith(ConstVal);
3251 </pre>
3252 </div>
3254 </ul>
3256 </div>
3258 </div>
3260 <!-- ======================================================================= -->
3261 <h3>
3263 </h3>
3265 <div>
3267 <p>
3280 Single Assignment (SSA) form, there can only be one definition referred to,
3281 allowing this direct connection. This connection provides the use-def
3282 information in LLVM.</p>
3284 <!-- _______________________________________________________________________ -->
3285 <h4>
3287 </h4>
3289 <div>
3292 an index access interface and through an iterator based interface.</p>
3294 <ul>
3298 convenient form for direct access.</p></li>
3301 list<br>
3303 the operand list.<br>
3305 operand list.
3306 <p> Together, these methods make up the iterator based interface to
3308 </ul>
3310 </div>
3312 </div>
3314 <!-- ======================================================================= -->
3315 <h3>
3317 </h3>
3319 <div>
3328 instructions. It provides only a few methods, but is a very commonly used
3330 opcode (instruction type) and the parent <a
3332 into. To represent a specific type of instruction, one of many subclasses of
3341 file contains some meta-data about the various different types of instructions
3342 in LLVM. It describes the enum values that are used as opcodes (for example
3347 this file confuses doxygen, so these enum values don't show up correctly in the
3350 <!-- _______________________________________________________________________ -->
3351 <h4>
3354 </a>
3355 </h4>
3356 <div>
3357 <ul>
3359 <p>This subclasses represents all two operand instructions whose operands
3360 must be the same type, except for the comparison instructions.</p></li>
3363 common operations on cast instructions.</p>
3365 <p>This subclass respresents the two comparison instructions,
3369 <p>This subclass is the parent of all terminator instructions (those which
3370 can terminate a block).</p>
3371 </ul>
3372 </div>
3374 <!-- _______________________________________________________________________ -->
3375 <h4>
3378 </a>
3379 </h4>
3381 <div>
3383 <ul>
3388 <p>Returns true if the instruction writes to memory, i.e. it is a
3393 <p>Returns another instance of the specified instruction, identical
3394 in all ways to the original except that the instruction has no parent
3396 and it has no name</p></li>
3397 </ul>
3399 </div>
3401 </div>
3403 <!-- ======================================================================= -->
3404 <h3>
3406 </h3>
3408 <div>
3410 <p>Constant represents a base class for different types of constants. It
3411 is subclassed by ConstantInt, ConstantArray, etc. for representing
3413 a subclass, which represents the address of a global variable or function.
3414 </p>
3416 <!-- _______________________________________________________________________ -->
3418 <div>
3419 <ul>
3420 <li>ConstantInt : This subclass of Constant represents an integer constant of
3421 any width.
3422 <ul>
3424 value of this constant, an APInt value.</li>
3426 value to an int64_t via sign extension. If the value (not the bit width)
3427 of the APInt is too large to fit in an int64_t, an assertion will result.
3428 For this reason, use of this method is discouraged.</li>
3430 value to a uint64_t via zero extension. IF the value (not the bit width)
3431 of the APInt is too large to fit in a uint64_t, an assertion will result.
3432 For this reason, use of this method is discouraged.</li>
3435 The type is implied as the IntegerType that corresponds to the bit width
3438 Returns the ConstantInt object that represents the value provided by
3440 </ul>
3441 </li>
3442 <li>ConstantFP : This class represents a floating point constant.
3443 <ul>
3445 this constant. </li>
3446 </ul>
3447 </li>
3448 <li>ConstantArray : This represents a constant array.
3449 <ul>
3451 a vector of component constants that makeup this array. </li>
3452 </ul>
3453 </li>
3454 <li>ConstantStruct : This represents a constant struct.
3455 <ul>
3457 a vector of component constants that makeup this array. </li>
3458 </ul>
3459 </li>
3460 <li>GlobalValue : This represents either a global variable or a function. In
3461 either case, the value is a constant fixed address (after linking).
3462 </li>
3463 </ul>
3464 </div>
3466 </div>
3468 <!-- ======================================================================= -->
3469 <h3>
3471 </h3>
3473 <div>
3478 Class</a><br>
3485 Because they are visible at global scope, they are also subject to linking with
3486 other globals defined in different translation units. To control the linking
3493 unit, and does not participate in linking. If it has external linkage, it is
3494 visible to external code, and does participate in linking. In addition to
3500 global is always a pointer to its contents. It is important to remember this
3505 the address of the first element of this array and the value of the
3513 <!-- _______________________________________________________________________ -->
3514 <h4>
3517 </a>
3518 </h4>
3520 <div>
3522 <ul>
3527 <p> </p>
3528 </li>
3531 GlobalValue is currently embedded into.</p></li>
3532 </ul>
3534 </div>
3536 </div>
3538 <!-- ======================================================================= -->
3539 <h3>
3541 </h3>
3543 <div>
3554 actually one of the more complex classes in the LLVM hierarchy because it must
3562 ordering of the blocks in the function, which indicate how the code will be
3563 laid out by the backend. Additionally, the first <a
3566 block. There are no implicit exit nodes, and in fact there may be multiple exit
3570 function hasn't been linked in yet.</p>
3580 LLVM feature that is only used when you have to look up a value by name. Aside
3582 internally to make sure that there are not conflicts between the names of <a
3589 is its address (after linking) which is guaranteed to be constant.</p>
3591 <!-- _______________________________________________________________________ -->
3592 <h4>
3595 </a>
3596 </h4>
3598 <div>
3600 <ul>
3605 the the program. The constructor must specify the type of the function to
3606 create and what type of linkage the function should have. The <a
3608 specifies the formal arguments and return value for the function. The same
3611 in which the function is defined. If this argument is provided, the function
3612 will automatically be inserted into that module's list of
3613 functions.</p></li>
3618 function is "external", it does not have a body, and thus must be resolved
3619 by linking with a function defined in a different translation unit.</p></li>
3627 <p>These are forwarding methods that make it easy to access the contents of
3629 list.</p></li>
3634 is necessary to use when you need to update the list or perform a complex
3635 action that doesn't have a forwarding method.</p></li>
3638 iterator<br>
3644 <p>These are forwarding methods that make it easy to access the contents of
3646 list.</p></li>
3651 necessary to use when you need to update the list or perform a complex
3652 action that doesn't have a forwarding method.</p></li>
3657 function. Because the entry block for the function is always the first
3666 function.</p></li>
3672 </ul>
3674 </div>
3676 </div>
3678 <!-- ======================================================================= -->
3679 <h3>
3681 </h3>
3683 <div>
3687 <br>
3689 Class</a><br>
3695 <p>Global variables are represented with the (surprise surprise)
3698 always referenced by their address (global values must live in memory, so their
3699 "name" refers to their constant address). See
3701 variables may have an initial value (which must be a
3703 they may be marked as "constant" themselves (indicating that their contents
3704 never change at runtime).</p>
3706 <!-- _______________________________________________________________________ -->
3707 <h4>
3710 </a>
3711 </h4>
3713 <div>
3715 <ul>
3720 <p>Create a new global variable of the specified type. If
3722 unchanging for the program. The Linkage parameter specifies the type of
3723 linkage (internal, external, weak, linkonce, appending) for the variable.
3724 If the linkage is InternalLinkage, WeakAnyLinkage, WeakODRLinkage,
3725 LinkOnceAnyLinkage or LinkOnceODRLinkage, then the resultant
3726 global variable will have internal linkage. AppendingLinkage concatenates
3727 together all instances (in different translation units) of the variable
3728 into a single variable but is only applicable to arrays. See
3730 further details on linkage types. Optionally an initializer, a name, and the
3731 module to put the variable into may be specified for the global variable as
3732 well.</p></li>
3736 <p>Returns true if this is a global variable that is known not to
3737 be modified at runtime.</p></li>
3746 to call this method if there is no initializer.</p></li>
3747 </ul>
3749 </div>
3751 </div>
3753 <!-- ======================================================================= -->
3754 <h3>
3756 </h3>
3758 <div>
3763 Class</a><br>
3766 <p>This class represents a single entry single exit section of the code,
3767 commonly known as a basic block by the compiler community. The
3770 Matching the language definition, the last element of this list of instructions
3771 is always a terminator instruction (a subclass of the <a
3774 <p>In addition to tracking the list of instructions that make up the block, the
3783 <!-- _______________________________________________________________________ -->
3784 <h4>
3787 </a>
3788 </h4>
3790 <div>
3791 <ul>
3797 insertion into a function. The constructor optionally takes a name for the new
3800 automatically inserted at the end of the specified <a
3808 STL-style functions for accessing the instruction list.
3810 <p>These methods and typedefs are forwarding functions that have the same
3811 semantics as the standard library methods of the same names. These methods
3812 expose the underlying instruction list of a basic block in a way that is easy to
3813 manipulate. To get the full complement of container operations (including
3815 method.</p></li>
3819 <p>This method is used to get access to the underlying container that actually
3820 holds the Instructions. This method must be used when there isn't a forwarding
3822 to perform. Because there are no forwarding functions for "updating"
3823 operations, you need to use this if you want to update the contents of a
3829 embedded into, or a null pointer if it is homeless.</p></li>
3833 <p> Returns a pointer to the terminator instruction that appears at the end of
3835 instruction in the block is not a terminator, then a null pointer is
3836 returned.</p></li>
3838 </ul>
3840 </div>
3842 </div>
3844 <!-- ======================================================================= -->
3845 <h3>
3847 </h3>
3849 <div>
3851 <p>This subclass of Value defines the interface for incoming formal
3852 arguments to a function. A Function maintains a list of its formal
3853 arguments. An argument has a pointer to the parent Function.</p>
3855 </div>
3857 </div>
3859 <!-- *********************************************************************** -->
3860 <hr>
3861 <address>
3870 Last modified: $Date$
3871 </address>
3873 </body>
3874 </html>