| blob | 49a76ee414767692a614a0b7e6bc69ba031eaa56 |
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>
164 <ul>
169 </ul></li>
171 <li><a href="#SymbolTable">The <tt>ValueSymbolTable</tt> and <tt>TypeSymbolTable</tt> classes</a></li>
172 <li><a href="#UserLayout">The <tt>User</tt> and owned <tt>Use</tt> classes' memory layout</a></li>
173 </ul></li>
176 <ul>
180 <ul>
182 <ul>
185 <ul>
187 <ul>
190 </ul>
191 </li>
192 </ul>
193 </li>
194 </ul>
195 </li>
198 </ul>
199 </li>
200 </ul>
201 </li>
202 </ol>
211 </div>
213 <!-- *********************************************************************** -->
214 <h2>
216 </h2>
217 <!-- *********************************************************************** -->
219 <div>
221 <p>This document is meant to highlight some of the important classes and
222 interfaces available in the LLVM source-base. This manual is not
223 intended to explain what LLVM is, how it works, and what LLVM code looks
224 like. It assumes that you know the basics of LLVM and are interested
225 in writing transformations or otherwise analyzing or manipulating the
226 code.</p>
228 <p>This document should get you oriented so that you can find your
229 way in the continuously growing source code that makes up the LLVM
230 infrastructure. Note that this manual is not intended to serve as a
231 replacement for reading the source code, so if you think there should be
232 a method in one of these classes to do something, but it's not listed,
234 are provided to make this as easy as possible.</p>
236 <p>The first section of this document describes general information that is
237 useful to know when working in the LLVM infrastructure, and the second describes
238 the Core LLVM classes. In the future this manual will be extended with
239 information describing how to use extension libraries, such as dominator
240 information, CFG traversal routines, and useful utilities like the <tt><a
243 </div>
245 <!-- *********************************************************************** -->
246 <h2>
248 </h2>
249 <!-- *********************************************************************** -->
251 <div>
253 <p>This section contains general information that is useful if you are working
254 in the LLVM source-base, but that isn't specific to any particular API.</p>
256 <!-- ======================================================================= -->
257 <h3>
259 </h3>
261 <div>
263 <p>LLVM makes heavy use of the C++ Standard Template Library (STL),
264 perhaps much more than you are used to, or have seen before. Because of
265 this, you might want to do a little background reading in the
266 techniques used and capabilities of the library. There are many good
267 pages that discuss the STL, and several books on the subject that you
268 can get, so it will not be discussed in this document.</p>
272 <ol>
275 C++ Library reference</a> - an excellent reference for the STL and other parts
276 of the standard C++ library.</li>
279 O'Reilly book in the making. It has a decent Standard Library
280 Reference that rivals Dinkumware's, and is unfortunately no longer free since the
281 book has been published.</li>
284 Questions</a></li>
287 Contains a useful <a
292 Page</a></li>
296 the book).</a></li>
298 </ol>
302 to write maintainable code more than where to put your curly braces.</p>
304 </div>
306 <!-- ======================================================================= -->
307 <h3>
309 </h3>
311 <div>
313 <ol>
315 static and shared libraries across platforms</a></li>
316 </ol>
318 </div>
320 </div>
322 <!-- *********************************************************************** -->
323 <h2>
325 </h2>
326 <!-- *********************************************************************** -->
328 <div>
330 <p>Here we highlight some LLVM APIs that are generally useful and good to
331 know about when writing transformations.</p>
333 <!-- ======================================================================= -->
334 <h3>
337 </h3>
339 <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 <!-- ======================================================================= -->
444 <h3>
447 </h3>
449 <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,
461 passing strings efficiently.</p>
463 <!-- _______________________________________________________________________ -->
464 <h4>
466 </h4>
468 <div>
471 (a character array and a length) and supports the common operations available
474 <p>It can be implicitly constructed using a C style null-terminated string,
479 iterator find(StringRef Key);
480 </pre>
488 </pre>
494 for more information.</p>
497 pointers to external memory it is not generally safe to store an instance of the
498 class (unless you know that the external storage will not be freed). StringRef is
499 small and pervasive enough in LLVM that it should always be passed by value.</p>
501 </div>
503 <!-- _______________________________________________________________________ -->
504 <h4>
506 </h4>
508 <div>
511 strings. For example, a common LLVM paradigm is to name one instruction based on
512 the name of another instruction with a suffix, for example:</p>
515 <pre>
517 </pre>
518 </div>
522 which points to temporary (stack allocated) objects. Twines can be implicitly
523 constructed as the result of the plus operator applied to strings (i.e., a C
525 actual concatenation of strings until it is actually required, at which point
526 it can be efficiently rendered directly into a character array. This avoids
527 unnecessary heap allocation involved in constructing the temporary results of
528 string concatenation. See
530 for more information.</p>
533 and should almost never be stored or mentioned directly. They are intended
534 solely for use when defining a function which should be able to efficiently
535 accept concatenated strings.</p>
537 </div>
539 </div>
541 <!-- ======================================================================= -->
542 <h3>
544 </h3>
546 <div>
548 <p>Often when working on your pass you will put a bunch of debugging printouts
549 and other code into your pass. After you get it working, you want to remove
550 it, but you may need it again in the future (to work out new bugs that you run
551 across).</p>
553 <p> Naturally, because of this, you don't want to delete the debug printouts,
554 but you don't want them to always be noisy. A standard compromise is to comment
555 them out, allowing you to enable them if you need them in the future.</p>
559 this problem. Basically, you can put arbitrary code into the argument of the
564 <pre>
566 </pre>
567 </div>
572 <pre>
576 I am here!
577 </pre>
578 </div>
581 to not have to create "yet another" command line option for the debug output for
583 so they do not cause a performance impact at all (for the same reason, they
584 should also not contain side-effects!).</p>
587 enable or disable it directly in gdb. Just use "<tt>set DebugFlag=0</tt>" or
588 "<tt>set DebugFlag=1</tt>" from the gdb if the program is running. If the
589 program hasn't been started yet, you can always just run it with
592 <!-- _______________________________________________________________________ -->
593 <h4>
596 </h4>
598 <div>
602 generator). If you want to enable debug information with more fine-grained
604 option as follows:</p>
607 <pre>
608 #undef DEBUG_TYPE
610 #define DEBUG_TYPE "foo"
612 #undef DEBUG_TYPE
613 #define DEBUG_TYPE "bar"
615 #undef DEBUG_TYPE
616 #define DEBUG_TYPE ""
618 </pre>
619 </div>
624 <pre>
628 No debug type
629 'foo' debug type
630 'bar' debug type
631 No debug type (2)
633 'foo' debug type
635 'bar' debug type
636 </pre>
637 </div>
640 a file, to specify the debug type for the entire module (if you do this before
643 "bar", because there is no system in place to ensure that names do not
644 conflict. If two different modules use the same string, they will all be turned
645 on when the name is specified. This allows, for example, all debug information
647 even if the source lives in multiple files.</p>
651 statement. It takes an additional first parameter, which is the type to use. For
652 example, the preceding example could be written as:</p>
656 <pre>
661 </pre>
662 </div>
664 </div>
666 </div>
668 <!-- ======================================================================= -->
669 <h3>
671 option</a>
672 </h3>
674 <div>
679 keep track of what the LLVM compiler is doing and how effective various
680 optimizations are. It is useful to see what optimizations are contributing to
681 making a particular program run faster.</p>
683 <p>Often you may run your pass on some big program, and you're interested to see
684 how many times it makes a certain transformation. Although you can do this with
685 hand inspection, or some ad-hoc method, this is a real pain and not very useful
687 keep track of this information, and the calculated information is presented in a
688 uniform manner with the rest of the passes being executed.</p>
691 it are as follows:</p>
693 <ol>
697 <pre>
698 #define <a href="#DEBUG_TYPE">DEBUG_TYPE</a> "mypassname" <i>// This goes before any #includes.</i>
699 STATISTIC(NumXForms, "The # of times I did stuff");
700 </pre>
701 </div>
704 specified by the first argument. The pass name is taken from the DEBUG_TYPE
705 macro, and the description is taken from the second argument. The variable
711 <pre>
713 </pre>
714 </div>
716 </li>
717 </ol>
723 <pre>
726 </pre>
727 </div>
730 suite, it gives a report that looks like this:</p>
733 <pre>
734 7646 bitcodewriter - Number of normal instructions
735 725 bitcodewriter - Number of oversized instructions
736 129996 bitcodewriter - Number of bitcode bytes written
737 2817 raise - Number of insts DCEd or constprop'd
738 3213 raise - Number of cast-of-self removed
739 5046 raise - Number of expression trees converted
740 75 raise - Number of other getelementptr's formed
741 138 raise - Number of load/store peepholes
742 42 deadtypeelim - Number of unused typenames removed from symtab
743 392 funcresolve - Number of varargs functions resolved
744 27 globaldce - Number of global variables removed
745 2 adce - Number of basic blocks removed
746 134 cee - Number of branches revectored
747 49 cee - Number of setcc instruction eliminated
748 532 gcse - Number of loads removed
749 2919 gcse - Number of instructions removed
750 86 indvars - Number of canonical indvars added
751 87 indvars - Number of aux indvars removed
752 25 instcombine - Number of dead inst eliminate
753 434 instcombine - Number of insts combined
754 248 licm - Number of load insts hoisted
755 1298 licm - Number of insts hoisted to a loop pre-header
756 3 licm - Number of insts hoisted to multiple loop preds (bad, no loop pre-header)
757 75 mem2reg - Number of alloca's promoted
758 1444 cfgsimplify - Number of blocks simplified
759 </pre>
760 </div>
762 <p>Obviously, with so many optimizations, having a unified framework for this
763 stuff is very nice. Making your pass fit well into the framework makes it more
764 maintainable and useful.</p>
766 </div>
768 <!-- ======================================================================= -->
769 <h3>
771 </h3>
773 <div>
775 <p>Several of the important data structures in LLVM are graphs: for example
779 DAGs</a>. In many cases, while debugging various parts of the compiler, it is
780 nice to instantly visualize these graphs.</p>
782 <p>LLVM provides several callbacks that are available in a debug build to do
784 the current LLVM tool will pop up a window containing the CFG for the function
785 where each basic block is a node in the graph, and each node contains the
786 instructions in the block. Similarly, there also exists
791 up a window. Alternatively, you can sprinkle calls to these functions in your
792 code in places you want to debug.</p>
794 <p>Getting this to work requires a small amount of configuration. On Unix
796 toolkit, and make sure 'dot' and 'gv' are in your path. If you are running on
797 Mac OS/X, download and install the Mac OS/X <a
800 it) to your path. Once in your system and path are set up, rerun the LLVM
801 configure script and rebuild LLVM to enable this functionality.</p>
807 specified color (choices of colors can be found at <a
809 complex node attributes can be provided with <tt>call
812 Attributes</a>.) If you want to restart and clear all the current graph
815 <p>Note that graph visualization features are compiled out of Release builds
816 to reduce file size. This means that you need a Debug+Asserts or
817 Release+Asserts build to use these features.</p>
819 </div>
821 </div>
823 <!-- *********************************************************************** -->
824 <h2>
826 </h2>
827 <!-- *********************************************************************** -->
829 <div>
832 and we commonly use STL data structures. This section describes the trade-offs
833 you should consider when you pick one.</p>
835 <p>
836 The first step is a choose your own adventure: do you want a sequential
837 container, a set-like container, or a map-like container? The most important
838 thing when choosing a container is the algorithmic properties of how you plan to
839 access the container. Based on that, you should use:</p>
841 <ul>
843 of an value based on another value. Map-like containers also support
844 efficient queries for containment (whether a key is in the map). Map-like
845 containers generally do not support efficient reverse mapping (values to
846 keys). If you need that, use two maps. Some map-like containers also
847 support efficient iteration through the keys in sorted order. Map-like
848 containers are the most expensive sort, only use them if you need one of
849 these capabilities.</li>
852 stuff into a container that automatically eliminates duplicates. Some
853 set-like containers support efficient iteration through the elements in
854 sorted order. Set-like containers are more expensive than sequential
855 containers.
856 </li>
859 the most efficient way to add elements and keeps track of the order they are
860 added to the collection. They permit duplicates and support efficient
861 iteration, but do not support efficient look-up based on a key.
862 </li>
865 container or reference structure that is used for character or byte
866 arrays.</li>
869 perform set operations on sets of numeric id's, while automatically
870 eliminating duplicates. Bit containers require a maximum of 1 bit for each
871 identifier you want to store.
872 </li>
873 </ul>
875 <p>
876 Once the proper category of container is determined, you can fine tune the
877 memory use, constant factors, and cache behaviors of access by intelligently
878 picking a member of the category. Note that constant factors and cache behavior
879 can be a big deal. If you have a vector that usually only contains a few
880 elements (but could contain many), for example, it's much better to use
882 . Doing so avoids (relatively) expensive malloc/free calls, which dwarf the
883 cost of adding the elements to the container. </p>
885 <!-- ======================================================================= -->
886 <h3>
888 </h3>
890 <div>
891 There are a variety of sequential containers available for you, based on your
892 needs. Pick the first in this section that will do what you want.
894 <!-- _______________________________________________________________________ -->
895 <h4>
897 </h4>
899 <div>
900 <p>The llvm::ArrayRef class is the preferred class to use in an interface that
901 accepts a sequential list of elements in memory and just reads from them. By
902 taking an ArrayRef, the API can be passed a fixed size array, an std::vector,
903 an llvm::SmallVector and anything else that is contiguous in memory.
904 </p>
905 </div>
909 <!-- _______________________________________________________________________ -->
910 <h4>
912 </h4>
914 <div>
915 <p>Fixed size arrays are very simple and very fast. They are good if you know
916 exactly how many elements you have, or you have a (low) upper bound on how many
917 you have.</p>
918 </div>
920 <!-- _______________________________________________________________________ -->
921 <h4>
923 </h4>
925 <div>
926 <p>Heap allocated arrays (new[] + delete[]) are also simple. They are good if
927 the number of elements is variable, if you know how many elements you will need
928 before the array is allocated, and if the array is usually large (if not,
930 allocated array is the cost of the new/delete (aka malloc/free). Also note that
931 if you are allocating an array of a type with a constructor, the constructor and
932 destructors will be run for every element in the array (re-sizable vectors only
933 construct those elements actually used).</p>
934 </div>
936 <!-- _______________________________________________________________________ -->
937 <h4>
939 </h4>
941 <div>
944 it supports efficient iteration, lays out elements in memory order (so you can
945 do pointer arithmetic between elements), supports efficient push_back/pop_back
946 operations, supports efficient random access to its elements, etc.</p>
948 <p>The advantage of SmallVector is that it allocates space for
950 the SmallVector is dynamically smaller than N, no malloc is performed. This can
951 be a big win in cases where the malloc/free call is far more expensive than the
952 code that fiddles around with the elements.</p>
955 predecessors/successors of a block is usually less than 8). On the other hand,
956 this makes the size of the SmallVector itself large, so you don't want to
957 allocate lots of them (doing so will waste a lot of space). As such,
958 SmallVectors are most useful when on the stack.</p>
960 <p>SmallVector also provides a nice portable and efficient replacement for
963 </div>
965 <!-- _______________________________________________________________________ -->
966 <h4>
968 </h4>
970 <div>
971 <p>
972 std::vector is well loved and respected. It is useful when SmallVector isn't:
973 when the size of the vector is often large (thus the small optimization will
974 rarely be a benefit) or if you will be allocating many instances of the vector
975 itself (which would waste space for elements that aren't in the container).
976 vector is also useful when interfacing with code that expects vectors :).
977 </p>
982 <pre>
983 for ( ... ) {
985 use V;
986 }
987 </pre>
988 </div>
993 <pre>
995 for ( ... ) {
996 use V;
997 V.clear();
998 }
999 </pre>
1000 </div>
1002 <p>Doing so will save (at least) one heap allocation and free per iteration of
1003 the loop.</p>
1005 </div>
1007 <!-- _______________________________________________________________________ -->
1008 <h4>
1010 </h4>
1012 <div>
1013 <p>std::deque is, in some senses, a generalized version of std::vector. Like
1014 std::vector, it provides constant time random access and other similar
1015 properties, but it also provides efficient access to the front of the list. It
1016 does not guarantee continuity of elements within memory.</p>
1018 <p>In exchange for this extra flexibility, std::deque has significantly higher
1019 constant factor costs than std::vector. If possible, use std::vector or
1020 something cheaper.</p>
1021 </div>
1023 <!-- _______________________________________________________________________ -->
1024 <h4>
1026 </h4>
1028 <div>
1029 <p>std::list is an extremely inefficient class that is rarely useful.
1030 It performs a heap allocation for every element inserted into it, thus having an
1031 extremely high constant factor, particularly for small data types. std::list
1032 also only supports bidirectional iteration, not random access iteration.</p>
1034 <p>In exchange for this high cost, std::list supports efficient access to both
1035 ends of the list (like std::deque, but unlike std::vector or SmallVector). In
1036 addition, the iterator invalidation characteristics of std::list are stronger
1037 than that of a vector class: inserting or removing an element into the list does
1038 not invalidate iterator or pointers to other elements in the list.</p>
1039 </div>
1041 <!-- _______________________________________________________________________ -->
1042 <h4>
1044 </h4>
1046 <div>
1048 intrusive, because it requires the element to store and provide access to the
1049 prev/next pointers for the list.</p>
1053 provides some novel characteristics. In particular, it can efficiently store
1054 polymorphic objects, the traits class is informed when an element is inserted or
1056 constant-time splice operation.</p>
1058 <p>These properties are exactly what we want for things like
1062 Related classes of interest are explained in the following subsections:
1063 <ul>
1068 </ul>
1069 </div>
1071 <!-- _______________________________________________________________________ -->
1072 <h4>
1074 </h4>
1076 <div>
1077 <p>
1078 Useful for storing a vector of values using only a few number of bits for each
1079 value. Apart from the standard operations of a vector-like container, it can
1080 also perform an 'or' set operation.
1081 </p>
1086 <pre>
1087 enum State {
1088 None = 0x0,
1089 FirstCondition = 0x1,
1090 SecondCondition = 0x2,
1091 Both = 0x3
1092 };
1094 State get() {
1096 Vec1.push_back(FirstCondition);
1099 Vec2.push_back(SecondCondition);
1101 Vec1 |= Vec2;
1102 return Vec1[0]; // returns 'Both'.
1103 }
1104 </pre>
1105 </div>
1107 </div>
1109 <!-- _______________________________________________________________________ -->
1110 <h4>
1112 </h4>
1114 <div>
1117 publicly derive from this traits class.</p>
1118 </div>
1120 <!-- _______________________________________________________________________ -->
1121 <h4>
1123 </h4>
1125 <div>
1127 supports a slightly narrower interface. Notably, inserters from
1131 used for a wide variety of customizations.</p>
1132 </div>
1134 <!-- _______________________________________________________________________ -->
1135 <h4>
1137 </h4>
1139 <div>
1142 in the default manner.</p>
1147 </div>
1149 <!-- _______________________________________________________________________ -->
1150 <h4>
1152 </h4>
1154 <div>
1156 citizen in the C++ ecosystem, it needs to support the standard container
1161 <p>The only sensible solution to this problem is to allocate a so-called
1163 iterator, providing the back-link to the last element. However conforming to the
1165 also must not be dereferenced.</p>
1168 how to allocate and store the sentinel. The corresponding policy is dictated
1170 whenever the need for a sentinel arises.</p>
1172 <p>While the default policy is sufficient in most cases, it may break down when
1175 is wasted. To alleviate the situation with numerous and voluminous
1181 arithmetic is used to obtain the sentinel, which is relative to the
1183 extra pointer, which serves as the back-link of the sentinel. This is the only
1184 field in the ghostly sentinel which can be legally accessed.</p>
1185 </div>
1187 <!-- _______________________________________________________________________ -->
1188 <h4>
1190 </h4>
1192 <div>
1195 <p>There are also various STL adapter classes such as std::queue,
1196 std::priority_queue, std::stack, etc. These provide simplified access to an
1197 underlying container but don't affect the cost of the container itself.</p>
1199 </div>
1201 </div>
1203 <!-- ======================================================================= -->
1204 <h3>
1206 </h3>
1208 <div>
1210 <p>Set-like containers are useful when you need to canonicalize multiple values
1211 into a single representation. There are several different choices for how to do
1212 this, providing various trade-offs.</p>
1214 <!-- _______________________________________________________________________ -->
1215 <h4>
1217 </h4>
1219 <div>
1221 <p>If you intend to insert a lot of elements, then do a lot of queries, a
1222 great approach is to use a vector (or other sequential container) with
1223 std::sort+std::unique to remove duplicates. This approach works really well if
1224 your usage pattern has these two distinct phases (insert then query), and can be
1226 </p>
1228 <p>
1229 This combination provides the several nice properties: the result data is
1230 contiguous in memory (good for cache locality), has few allocations, is easy to
1231 address (iterators in the final vector are just indices or pointers), and can be
1232 efficiently queried with a standard binary or radix search.</p>
1234 </div>
1236 <!-- _______________________________________________________________________ -->
1237 <h4>
1239 </h4>
1241 <div>
1243 <p>If you have a set-like data structure that is usually small and whose elements
1245 has space for N elements in place (thus, if the set is dynamically smaller than
1246 N, no malloc traffic is required) and accesses them with a simple linear search.
1247 When the set grows beyond 'N' elements, it allocates a more expensive representation that
1248 guarantees efficient access (for most types, it falls back to std::set, but for
1249 pointers it uses something far better, <a
1252 <p>The magic of this class is that it handles small sets extremely efficiently,
1253 but gracefully handles extremely large sets without loss of efficiency. The
1254 drawback is that the interface is quite small: it supports insertion, queries
1255 and erasing, but does not support iteration.</p>
1257 </div>
1259 <!-- _______________________________________________________________________ -->
1260 <h4>
1262 </h4>
1264 <div>
1266 <p>SmallPtrSet has all the advantages of <tt>SmallSet</tt> (and a <tt>SmallSet</tt> of pointers is
1268 more than 'N' insertions are performed, a single quadratically
1269 probed hash table is allocated and grows as needed, providing extremely
1270 efficient access (constant time insertion/deleting/queries with low constant
1271 factors) and is very stingy with malloc traffic.</p>
1274 whenever an insertion occurs. Also, the values visited by the iterators are not
1275 visited in sorted order.</p>
1277 </div>
1279 <!-- _______________________________________________________________________ -->
1280 <h4>
1282 </h4>
1284 <div>
1286 <p>
1287 DenseSet is a simple quadratically probed hash table. It excels at supporting
1288 small values: it uses a single allocation to hold all of the pairs that
1289 are currently inserted in the set. DenseSet is a great way to unique small
1290 values that are not simple pointers (use <a
1292 the same requirements for the value type that <a
1294 </p>
1296 </div>
1298 <!-- _______________________________________________________________________ -->
1299 <h4>
1301 </h4>
1303 <div>
1305 <p>
1306 FoldingSet is an aggregate class that is really good at uniquing
1307 expensive-to-create or polymorphic objects. It is a combination of a chained
1308 hash table with intrusive links (uniqued objects are required to inherit from
1310 its ID process.</p>
1313 a complex object (for example, a node in the code generator). The client has a
1314 description of *what* it wants to generate (it knows the opcode and all the
1315 operands), but we don't want to 'new' a node, then try inserting it into a set
1316 only to find out it already exists, at which point we would have to delete it
1317 and return the node that already exists.
1318 </p>
1320 <p>To support this style of client, FoldingSet perform a query with a
1321 FoldingSetNodeID (which wraps SmallVector) that can be used to describe the
1322 element that we want to query for. The query either returns the element
1323 matching the ID or it returns an opaque ID that indicates where insertion should
1324 take place. Construction of the ID usually does not require heap traffic.</p>
1326 <p>Because FoldingSet uses intrusive links, it can support polymorphic objects
1327 in the set (for example, you can have SDNode instances mixed with LoadSDNodes).
1328 Because the elements are individually allocated, pointers to the elements are
1329 stable: inserting or removing elements does not invalidate any pointers to other
1330 elements.
1331 </p>
1333 </div>
1335 <!-- _______________________________________________________________________ -->
1336 <h4>
1338 </h4>
1340 <div>
1343 many things but great at nothing. std::set allocates memory for each element
1344 inserted (thus it is very malloc intensive) and typically stores three pointers
1345 per element in the set (thus adding a large amount of per-element space
1346 overhead). It offers guaranteed log(n) performance, which is not particularly
1347 fast from a complexity standpoint (particularly if the elements of the set are
1348 expensive to compare, like strings), and has extremely high constant factors for
1349 lookup, insertion and removal.</p>
1351 <p>The advantages of std::set are that its iterators are stable (deleting or
1352 inserting an element from the set does not affect iterators or pointers to other
1353 elements) and that iteration over the set is guaranteed to be in sorted order.
1354 If the elements in the set are large, then the relative overhead of the pointers
1355 and malloc traffic is not a big deal, but if the elements of the set are small,
1356 std::set is almost never a good choice.</p>
1358 </div>
1360 <!-- _______________________________________________________________________ -->
1361 <h4>
1363 </h4>
1365 <div>
1368 Container</a>. The important property
1369 that this provides is efficient insertion with uniquing (duplicate elements are
1370 ignored) with iteration support. It implements this by inserting elements into
1371 both a set-like container and the sequential container, using the set-like
1372 container for uniquing and the sequential container for iteration.
1373 </p>
1375 <p>The difference between SetVector and other sets is that the order of
1376 iteration is guaranteed to match the order of insertion into the SetVector.
1377 This property is really important for things like sets of pointers. Because
1378 pointer values are non-deterministic (e.g. vary across runs of the program on
1379 different machines), iterating over the pointers in the set will
1380 not be in a well-defined order.</p>
1382 <p>
1383 The drawback of SetVector is that it requires twice as much space as a normal
1384 set and has the sum of constant factors from the set-like container and the
1385 sequential container that it uses. Use it *only* if you need to iterate over
1386 the elements in a deterministic order. SetVector is also expensive to delete
1387 elements out of (linear time), unless you use it's "pop_back" method, which is
1388 faster.
1389 </p>
1391 <p>SetVector is an adapter class that defaults to using std::vector and std::set
1392 for the underlying containers, so it is quite expensive. However,
1394 defaults to using a SmallVector and SmallSet of a specified size. If you use
1395 this, and if your sets are dynamically smaller than N, you will save a lot of
1396 heap traffic.</p>
1398 </div>
1400 <!-- _______________________________________________________________________ -->
1401 <h4>
1403 </h4>
1405 <div>
1407 <p>
1409 retains a unique ID for each element inserted into the set. It internally
1410 contains a map and a vector, and it assigns a unique ID for each value inserted
1411 into the set.</p>
1413 <p>UniqueVector is very expensive: its cost is the sum of the cost of
1414 maintaining both the map and vector, it has high complexity, high constant
1415 factors, and produces a lot of malloc traffic. It should be avoided.</p>
1417 </div>
1420 <!-- _______________________________________________________________________ -->
1421 <h4>
1423 </h4>
1425 <div>
1427 <p>
1428 The STL provides several other options, such as std::multiset and the various
1429 "hash_set" like containers (whether from C++ TR1 or from the SGI library). We
1430 never use hash_set and unordered_set because they are generally very expensive
1431 (each insertion requires a malloc) and very non-portable.
1432 </p>
1434 <p>std::multiset is useful if you're not interested in elimination of
1435 duplicates, but has all the drawbacks of std::set. A sorted vector (where you
1436 don't delete duplicate entries) or some other approach is almost always
1437 better.</p>
1439 </div>
1441 </div>
1443 <!-- ======================================================================= -->
1444 <h3>
1446 </h3>
1448 <div>
1449 Map-like containers are useful when you want to associate data to a key. As
1450 usual, there are a lot of different ways to do this. :)
1452 <!-- _______________________________________________________________________ -->
1453 <h4>
1455 </h4>
1457 <div>
1459 <p>
1460 If your usage pattern follows a strict insert-then-query approach, you can
1462 for set-like containers</a>. The only difference is that your query function
1463 (which uses std::lower_bound to get efficient log(n) lookup) should only compare
1464 the key, not both the key and value. This yields the same advantages as sorted
1465 vectors for sets.
1466 </p>
1467 </div>
1469 <!-- _______________________________________________________________________ -->
1470 <h4>
1472 </h4>
1474 <div>
1476 <p>
1477 Strings are commonly used as keys in maps, and they are difficult to support
1478 efficiently: they are variable length, inefficient to hash and compare when
1479 long, expensive to copy, etc. StringMap is a specialized container designed to
1480 cope with these issues. It supports mapping an arbitrary range of bytes to an
1481 arbitrary other object.</p>
1483 <p>The StringMap implementation uses a quadratically-probed hash table, where
1484 the buckets store a pointer to the heap allocated entries (and some other
1485 stuff). The entries in the map must be heap allocated because the strings are
1486 variable length. The string data (key) and the element object (value) are
1487 stored in the same allocation with the string data immediately after the element
1488 object. This container guarantees the "<tt>(char*)(&Value+1)</tt>" points
1489 to the key string for a value.</p>
1491 <p>The StringMap is very fast for several reasons: quadratic probing is very
1492 cache efficient for lookups, the hash value of strings in buckets is not
1493 recomputed when looking up an element, StringMap rarely has to touch the
1494 memory for unrelated objects when looking up a value (even when hash collisions
1495 happen), hash table growth does not recompute the hash values for strings
1496 already in the table, and each pair in the map is store in a single allocation
1497 (the string data is stored in the same allocation as the Value of a pair).</p>
1499 <p>StringMap also provides query methods that take byte ranges, so it only ever
1500 copies a string if a value is inserted into the table.</p>
1501 </div>
1503 <!-- _______________________________________________________________________ -->
1504 <h4>
1506 </h4>
1508 <div>
1509 <p>
1510 IndexedMap is a specialized container for mapping small dense integers (or
1511 values that can be mapped to small dense integers) to some other type. It is
1512 internally implemented as a vector with a mapping function that maps the keys to
1513 the dense integer range.
1514 </p>
1516 <p>
1517 This is useful for cases like virtual registers in the LLVM code generator: they
1518 have a dense mapping that is offset by a compile-time constant (the first
1519 virtual register ID).</p>
1521 </div>
1523 <!-- _______________________________________________________________________ -->
1524 <h4>
1526 </h4>
1528 <div>
1530 <p>
1531 DenseMap is a simple quadratically probed hash table. It excels at supporting
1532 small keys and values: it uses a single allocation to hold all of the pairs that
1533 are currently inserted in the map. DenseMap is a great way to map pointers to
1534 pointers, or map other small types to each other.
1535 </p>
1537 <p>
1538 There are several aspects of DenseMap that you should be aware of, however. The
1539 iterators in a densemap are invalidated whenever an insertion occurs, unlike
1540 map. Also, because DenseMap allocates space for a large number of key/value
1541 pairs (it starts with 64 by default), it will waste a lot of space if your keys
1542 or values are large. Finally, you must implement a partial specialization of
1543 DenseMapInfo for the key that you want, if it isn't already supported. This
1544 is required to tell DenseMap about two special marker values (which can never be
1545 inserted into the map) that it needs internally.</p>
1547 </div>
1549 <!-- _______________________________________________________________________ -->
1550 <h4>
1552 </h4>
1554 <div>
1556 <p>
1558 Value*s (or subclasses) to another type. When a Value is deleted or RAUW'ed,
1559 ValueMap will update itself so the new version of the key is mapped to the same
1560 value, just as if the key were a WeakVH. You can configure exactly how this
1561 happens, and what else happens on these two events, by passing
1564 </div>
1566 <!-- _______________________________________________________________________ -->
1567 <h4>
1569 </h4>
1571 <div>
1573 <p> IntervalMap is a compact map for small keys and values. It maps key
1574 intervals instead of single keys, and it will automatically coalesce adjacent
1575 intervals. When then map only contains a few intervals, they are stored in the
1576 map object itself to avoid allocations.</p>
1578 <p> The IntervalMap iterators are quite big, so they should not be passed around
1579 as STL iterators. The heavyweight iterators allow a smaller data structure.</p>
1581 </div>
1583 <!-- _______________________________________________________________________ -->
1584 <h4>
1586 </h4>
1588 <div>
1590 <p>
1592 a single allocation per pair inserted into the map, it offers log(n) lookup with
1593 an extremely large constant factor, imposes a space penalty of 3 pointers per
1594 pair in the map, etc.</p>
1596 <p>std::map is most useful when your keys or values are very large, if you need
1597 to iterate over the collection in sorted order, or if you need stable iterators
1598 into the map (i.e. they don't get invalidated if an insertion or deletion of
1599 another element takes place).</p>
1601 </div>
1603 <!-- _______________________________________________________________________ -->
1604 <h4>
1606 </h4>
1608 <div>
1610 <p>IntEqClasses provides a compact representation of equivalence classes of
1612 equivalence class. Classes can be joined by passing two class representatives to
1613 the join(a, b) method. Two integers are in the same class when findLeader()
1614 returns the same representative.</p>
1616 <p>Once all equivalence classes are formed, the map can be compressed so each
1618 is the total number of equivalence classes. The map must be uncompressed before
1619 it can be edited again.</p>
1621 </div>
1623 <!-- _______________________________________________________________________ -->
1624 <h4>
1626 </h4>
1628 <div>
1630 <p>
1631 The STL provides several other options, such as std::multimap and the various
1632 "hash_map" like containers (whether from C++ TR1 or from the SGI library). We
1633 never use hash_set and unordered_set because they are generally very expensive
1634 (each insertion requires a malloc) and very non-portable.</p>
1636 <p>std::multimap is useful if you want to map a key to multiple values, but has
1637 all the drawbacks of std::map. A sorted vector or some other approach is almost
1638 always better.</p>
1640 </div>
1642 </div>
1644 <!-- ======================================================================= -->
1645 <h3>
1647 </h3>
1649 <div>
1651 <p>
1652 TODO: const char* vs stringref vs smallstring vs std::string. Describe twine,
1653 xref to #string_apis.
1654 </p>
1656 </div>
1658 <!-- ======================================================================= -->
1659 <h3>
1661 </h3>
1663 <div>
1664 <p>Unlike the other containers, there are only two bit storage containers, and
1665 choosing when to use each is relatively straightforward.</p>
1667 <p>One additional option is
1669 implementation in many common compilers (e.g. commonly available versions of
1670 GCC) is extremely inefficient and 2) the C++ standards committee is likely to
1671 deprecate this container and/or change it significantly somehow. In any case,
1672 please don't use it.</p>
1674 <!-- _______________________________________________________________________ -->
1675 <h4>
1677 </h4>
1679 <div>
1680 <p> The BitVector container provides a dynamic size set of bits for manipulation.
1681 It supports individual bit setting/testing, as well as set operations. The set
1682 operations take time O(size of bitvector), but operations are performed one word
1683 at a time, instead of one bit at a time. This makes the BitVector very fast for
1684 set operations compared to other containers. Use the BitVector when you expect
1685 the number of set bits to be high (IE a dense set).
1686 </p>
1687 </div>
1689 <!-- _______________________________________________________________________ -->
1690 <h4>
1692 </h4>
1694 <div>
1695 <p> The SmallBitVector container provides the same interface as BitVector, but
1696 it is optimized for the case where only a small number of bits, less than
1697 25 or so, are needed. It also transparently supports larger bit counts, but
1698 slightly less efficiently than a plain BitVector, so SmallBitVector should
1699 only be used when larger counts are rare.
1700 </p>
1702 <p>
1703 At this time, SmallBitVector does not support set operations (and, or, xor),
1704 and its operator[] does not provide an assignable lvalue.
1705 </p>
1706 </div>
1708 <!-- _______________________________________________________________________ -->
1709 <h4>
1711 </h4>
1713 <div>
1714 <p> The SparseBitVector container is much like BitVector, with one major
1715 difference: Only the bits that are set, are stored. This makes the
1716 SparseBitVector much more space efficient than BitVector when the set is sparse,
1717 as well as making set operations O(number of set bits) instead of O(size of
1718 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
1719 (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).
1720 </p>
1721 </div>
1723 </div>
1725 </div>
1727 <!-- *********************************************************************** -->
1728 <h2>
1730 </h2>
1731 <!-- *********************************************************************** -->
1733 <div>
1735 <p>This section describes how to perform some very simple transformations of
1736 LLVM code. This is meant to give examples of common idioms used, showing the
1738 you should also read about the main classes that you will be working with. The
1740 and descriptions of the main classes that you should know about.</p>
1742 <!-- NOTE: this section should be heavy on example code -->
1743 <!-- ======================================================================= -->
1744 <h3>
1746 </h3>
1748 <div>
1750 <p>The LLVM compiler infrastructure have many different data structures that may
1751 be traversed. Following the example of the C++ standard template library, the
1752 techniques used to traverse these various data structures are all basically the
1755 function returns an iterator pointing to one past the last valid element of the
1757 between the two operations.</p>
1759 <p>Because the pattern for iteration is common across many different aspects of
1760 the program representation, the standard template library algorithms may be used
1761 on them, and it is easier to remember how to iterate. First we show a few common
1762 examples of the data structures that need to be traversed. Other data
1763 structures are traversed in very similar ways.</p>
1765 <!-- _______________________________________________________________________ -->
1766 <h4>
1770 </h4>
1772 <div>
1775 transform in some way; in particular, you'd like to manipulate its
1782 <pre>
1789 </pre>
1790 </div>
1792 <p>Note that i can be used as if it were a pointer for the purposes of
1794 because the indirection operator is overloaded for the iterator
1798 </div>
1800 <!-- _______________________________________________________________________ -->
1801 <h4>
1805 </h4>
1807 <div>
1810 easy to iterate over the individual instructions that make up
1815 <pre>
1821 </pre>
1822 </div>
1824 <p>However, this isn't really the best way to print out the contents of a
1826 anything you'll care about, you could have just invoked the print routine on the
1829 </div>
1831 <!-- _______________________________________________________________________ -->
1832 <h4>
1836 </h4>
1838 <div>
1845 small example that shows how to dump all instructions in a function to the standard error stream:<p>
1848 <pre>
1852 for (inst_iterator I = inst_begin(F), E = inst_end(F); I != E; ++I)
1854 </pre>
1855 </div>
1858 work list with its initial contents. For example, if you wanted to
1860 F, all you would need to do is something like:</p>
1863 <pre>
1867 for (inst_iterator I = inst_begin(F), E = inst_end(F); I != E; ++I)
1868 worklist.insert(&*I);
1869 </pre>
1870 </div>
1875 </div>
1877 <!-- _______________________________________________________________________ -->
1878 <h4>
1880 vice-versa)</a>
1881 </h4>
1883 <div>
1885 <p>Sometimes, it'll be useful to grab a reference (or pointer) to a class
1886 instance when all you've got at hand is an iterator. Well, extracting
1887 a reference or a pointer from an iterator is very straight-forward.
1892 <pre>
1895 const Instruction& inst = *j;
1896 </pre>
1897 </div>
1899 <p>However, the iterators you'll be working with in the LLVM framework are
1900 special: they will automatically convert to a ptr-to-instance type whenever they
1901 need to. Instead of dereferencing the iterator and then taking the address of
1902 the result, you can simply assign the iterator to the proper pointer type and
1903 you get the dereference and address-of operation as a result of the assignment
1904 (behind the scenes, this is a result of overloading casting mechanisms). Thus
1905 the last line of the last example,</p>
1908 <pre>
1909 Instruction *pinst = &*i;
1910 </pre>
1911 </div>
1916 <pre>
1917 Instruction *pinst = i;
1918 </pre>
1919 </div>
1921 <p>It's also possible to turn a class pointer into the corresponding iterator,
1922 and this is a constant time operation (very efficient). The following code
1923 snippet illustrates use of the conversion constructors provided by LLVM
1924 iterators. By using these, you can explicitly grab the iterator of something
1925 without actually obtaining it via iteration over some structure:</p>
1928 <pre>
1929 void printNextInstruction(Instruction* inst) {
1930 BasicBlock::iterator it(inst);
1933 }
1934 </pre>
1935 </div>
1937 <p>Unfortunately, these implicit conversions come at a cost; they prevent
1938 these iterators from conforming to standard iterator conventions, and thus
1939 from being usable with standard algorithms and containers. For example, they
1941 from compiling:</p>
1944 <pre>
1946 </pre>
1947 </div>
1949 <p>Because of this, these implicit conversions may be removed some day,
1952 </div>
1954 <!--_______________________________________________________________________-->
1955 <h4>
1957 example</a>
1958 </h4>
1960 <div>
1962 <p>Say that you're writing a FunctionPass and would like to count all the
1966 much more straight-forward manner, but this example will allow us to explore how
1968 is what we want to do:</p>
1971 <pre>
1972 initialize callCounter to zero
1973 for each Function f in the Module
1974 for each BasicBlock b in f
1975 for each Instruction i in b
1976 if (i is a CallInst and calls the given function)
1977 increment callCounter
1978 </pre>
1979 </div>
1981 <p>And the actual code is (remember, because we're writing a
1986 <pre>
1987 Function* targetFunc = ...;
1989 class OurFunctionPass : public FunctionPass {
1990 public:
1991 OurFunctionPass(): callCounter(0) { }
1993 virtual runOnFunction(Function& F) {
1994 for (Function::iterator b = F.begin(), be = F.end(); b != be; ++b) {
2001 if (callInst->getCalledFunction() == targetFunc)
2002 ++callCounter;
2003 }
2004 }
2005 }
2006 }
2008 private:
2009 unsigned callCounter;
2010 };
2011 </pre>
2012 </div>
2014 </div>
2016 <!--_______________________________________________________________________-->
2017 <h4>
2019 </h4>
2021 <div>
2023 <p>You may have noticed that the previous example was a bit oversimplified in
2024 that it did not deal with call sites generated by 'invoke' instructions. In
2025 this, and in other situations, you may find that you want to treat
2028 less closely-related things. For these cases, LLVM provides a handy wrapper
2029 class called <a
2036 reference and it should not be dynamically allocated or deallocated using
2038 assignable and constructable, with costs equivalents to that of a bare pointer.
2039 If you look at its definition, it has only a single pointer member.</p>
2041 </div>
2043 <!--_______________________________________________________________________-->
2044 <h4>
2046 </h4>
2048 <div>
2060 <pre>
2061 Function *F = ...;
2067 }
2068 </pre>
2069 </div>
2073 doing it only once in the loop body and reusing its result.</p>
2080 all of the values that a particular instruction uses (that is, the operands of
2084 <pre>
2085 Instruction *pi = ...;
2088 Value *v = *i;
2090 }
2091 </pre>
2092 </div>
2095 mutation free algorithms (such as analyses, etc.). For this purpose above
2102 </div>
2104 <!--_______________________________________________________________________-->
2105 <h4>
2107 successors of blocks</a>
2108 </h4>
2110 <div>
2112 <p>Iterating over the predecessors and successors of a block is quite easy
2114 this to iterate over all predecessors of BB:</p>
2117 <pre>
2118 #include "llvm/Support/CFG.h"
2119 BasicBlock *BB = ...;
2121 for (pred_iterator PI = pred_begin(BB), E = pred_end(BB); PI != E; ++PI) {
2122 BasicBlock *Pred = *PI;
2124 }
2125 </pre>
2126 </div>
2128 <p>Similarly, to iterate over successors use
2129 succ_iterator/succ_begin/succ_end.</p>
2131 </div>
2133 </div>
2135 <!-- ======================================================================= -->
2136 <h3>
2138 </h3>
2140 <div>
2142 <p>There are some primitive transformation operations present in the LLVM
2143 infrastructure that are worth knowing about. When performing
2144 transformations, it's fairly common to manipulate the contents of basic
2145 blocks. This section describes some of the common methods for doing so
2146 and gives example code.</p>
2148 <!--_______________________________________________________________________-->
2149 <h4>
2152 </h4>
2154 <div>
2159 constructor for the kind of instruction to instantiate and provide the necessary
2164 <pre>
2165 AllocaInst* ai = new AllocaInst(Type::Int32Ty);
2166 </pre>
2167 </div>
2171 subclass is likely to have varying default parameters which change the semantics
2172 of the instruction, so refer to the <a
2178 <p>It is very useful to name the values of instructions when you're able to, as
2179 this facilitates the debugging of your transformations. If you end up looking
2180 at generated LLVM machine code, you definitely want to have logical names
2181 associated with the results of instructions! By supplying a value for the
2183 associate a logical name with the result of the instruction's execution at
2184 run time. For example, say that I'm writing a transformation that dynamically
2185 allocates space for an integer on the stack, and that integer is going to be
2186 used as some kind of index by some other code. To accomplish this, I place an
2192 <pre>
2194 </pre>
2195 </div>
2198 execution value, which is a pointer to an integer on the run time stack.</p>
2205 <ul>
2206 <li>Insertion into an explicit instruction list
2213 <pre>
2214 BasicBlock *pb = ...;
2215 Instruction *pi = ...;
2216 Instruction *newInst = new Instruction(...);
2219 </pre>
2220 </div>
2224 classes provide constructors which take a pointer to a
2226 looked like: </p>
2229 <pre>
2230 BasicBlock *pb = ...;
2231 Instruction *newInst = new Instruction(...);
2234 </pre>
2235 </div>
2240 <pre>
2241 BasicBlock *pb = ...;
2242 Instruction *newInst = new Instruction(..., pb);
2243 </pre>
2244 </div>
2246 <p>which is much cleaner, especially if you are creating
2247 long instruction streams.</p></li>
2249 <li>Insertion into an implicit instruction list
2252 are implicitly associated with an existing instruction list: the instruction
2253 list of the enclosing basic block. Thus, we could have accomplished the same
2255 </p>
2258 <pre>
2259 Instruction *pi = ...;
2260 Instruction *newInst = new Instruction(...);
2263 </pre>
2264 </div>
2266 <p>In fact, this sequence of steps occurs so frequently that the
2268 constructors which take (as a default parameter) a pointer to an
2272 provided instruction, immediately before that instruction. Using an
2274 parameter, the above code becomes:</p>
2277 <pre>
2278 Instruction* pi = ...;
2279 Instruction* newInst = new Instruction(..., pi);
2280 </pre>
2281 </div>
2283 <p>which is much cleaner, especially if you're creating a lot of
2285 </ul>
2287 </div>
2289 <!--_______________________________________________________________________-->
2290 <h4>
2292 </h4>
2294 <div>
2296 <p>Deleting an instruction from an existing sequence of instructions that form a
2298 call the instruction's eraseFromParent() method. For example:</p>
2301 <pre>
2303 I->eraseFromParent();
2304 </pre>
2305 </div>
2307 <p>This unlinks the instruction from its containing basic block and deletes
2308 it. If you'd just like to unlink the instruction from its containing basic
2311 </div>
2313 <!--_______________________________________________________________________-->
2314 <h4>
2317 </h4>
2319 <div>
2323 <p>Including "<a href="/doxygen/BasicBlockUtils_8h-source.html">llvm/Transforms/Utils/BasicBlockUtils.h</a>"
2329 <ul>
2332 <p>This function replaces all uses of a given instruction with a value,
2333 and then removes the original instruction. The following example
2334 illustrates the replacement of the result of a particular
2336 pointer to an integer.</p>
2339 <pre>
2340 AllocaInst* instToReplace = ...;
2341 BasicBlock::iterator ii(instToReplace);
2344 Constant::getNullValue(PointerType::getUnqual(Type::Int32Ty)));
2345 </pre></div></li>
2349 <p>This function replaces a particular instruction with another
2350 instruction, inserting the new instruction into the basic block at the
2351 location where the old instruction was, and replacing any uses of the old
2352 instruction with the new instruction. The following example illustrates
2356 <pre>
2357 AllocaInst* instToReplace = ...;
2358 BasicBlock::iterator ii(instToReplace);
2362 </pre></div></li>
2363 </ul>
2371 information.</p>
2373 <!-- Value::replaceAllUsesWith User::replaceUsesOfWith Point out:
2374 include/llvm/Transforms/Utils/ especially BasicBlockUtils.h with:
2375 ReplaceInstWithValue, ReplaceInstWithInst -->
2377 </div>
2379 <!--_______________________________________________________________________-->
2380 <h4>
2382 </h4>
2384 <div>
2386 <p>Deleting a global variable from a module is just as easy as deleting an
2387 Instruction. First, you must have a pointer to the global variable that you wish
2388 to delete. You use this pointer to erase it from its parent, the module.
2389 For example:</p>
2392 <pre>
2395 GV->eraseFromParent();
2396 </pre>
2397 </div>
2399 </div>
2401 </div>
2403 <!-- ======================================================================= -->
2404 <h3>
2406 </h3>
2408 <div>
2410 <p>In generating IR, you may need some complex types. If you know these types
2413 has two forms depending on whether you're building types for cross-compilation
2416 out of types from
2418 namespace and pointers, functions, arrays, etc. built of
2420 whose size may depend on the host compiler. For example,</p>
2423 <pre>
2425 </pre>
2426 </div>
2431 <pre>
2433 params.push_back(PointerType::getUnqual(Type::Int32Ty));
2434 FunctionType *ft = FunctionType::get(Type::Int8Ty, params, false);
2435 </pre>
2436 </div>
2441 </div>
2443 </div>
2445 <!-- *********************************************************************** -->
2446 <h2>
2448 </h2>
2449 <!-- *********************************************************************** -->
2451 <div>
2452 <p>
2453 This section describes the interaction of the LLVM APIs with multithreading,
2454 both on the part of client applications, and in the JIT, in the hosted
2455 application.
2456 </p>
2458 <p>
2459 Note that LLVM's support for multithreading is still relatively young. Up
2460 through version 2.5, the execution of threaded hosted applications was
2461 supported, but not threaded client access to the APIs. While this use case is
2463 ensure proper operation in multithreaded mode.
2464 </p>
2466 <p>
2467 Note that, on Unix-like platforms, LLVM requires the presence of GCC's atomic
2468 intrinsics in order to support threaded operation. If you need a
2469 multhreading-capable LLVM on a platform without a suitably modern system
2470 compiler, consider compiling LLVM and LLVM-GCC in single-threaded mode, and
2471 using the resultant compiler to build a copy of LLVM with multithreading
2472 support.
2473 </p>
2475 <!-- ======================================================================= -->
2476 <h3>
2478 </h3>
2480 <div>
2482 <p>
2483 In order to properly protect its internal data structures while avoiding
2484 excessive locking overhead in the single-threaded case, the LLVM must intialize
2485 certain data structures necessary to provide guards around its internals. To do
2487 making any concurrent LLVM API calls. To subsequently tear down these
2490 mode.
2491 </p>
2493 <p>
2495 say that no other LLVM API calls may be executing at any time during the
2497 </tt>. It's is the client's responsibility to enforce this isolation.
2498 </p>
2500 <p>
2502 failure of the initialization. Failure typically indicates that your copy of
2503 LLVM was built without multithreading support, typically because GCC atomic
2504 intrinsics were not found in your system compiler. In this case, the LLVM API
2507 must be taken</a> to ensure that side exits and the like do not accidentally
2508 result in concurrent LLVM API calls.
2509 </p>
2510 </div>
2512 <!-- ======================================================================= -->
2513 <h3>
2515 </h3>
2517 <div>
2518 <p>
2520 to deallocate memory used for internal structures. This will also invoke
2524 </p>
2526 <p>
2527 Note that, if you use scope-based shutdown, you can use the
2529 destructor.
2530 </div>
2532 <!-- ======================================================================= -->
2533 <h3>
2535 </h3>
2537 <div>
2538 <p>
2540 initialization of static resources, such as the global type tables. Before the
2543 however, it uses double-checked locking to implement thread-safe lazy
2544 initialization.
2545 </p>
2547 <p>
2548 Note that, because no other threads are allowed to issue LLVM API calls before
2551 </p>
2553 <p>
2555 APIs provide access to the global lock used to implement the double-checked
2556 locking for lazy initialization. These should only be used internally to LLVM,
2557 and only if you know what you're doing!
2558 </p>
2559 </div>
2561 <!-- ======================================================================= -->
2562 <h3>
2564 </h3>
2566 <div>
2567 <p>
2569 to operate multiple, isolated instances of LLVM concurrently within the same
2570 address space. For instance, in a hypothetical compile-server, the compilation
2571 of an individual translation unit is conceptually independent from all the
2572 others, and it would be desirable to be able to compile incoming translation
2573 units concurrently on independent server threads. Fortunately,
2575 </p>
2577 <p>
2584 safe to compile on multiple threads simultaneously, as long as no two threads
2585 operate on entities within the same context.
2586 </p>
2588 <p>
2589 In practice, very few places in the API require the explicit specification of a
2592 other entities can determine what context they belong to by looking at their
2594 maintain this interface design.
2595 </p>
2597 <p>
2601 isolation is not a concern.
2602 </p>
2603 </div>
2605 <!-- ======================================================================= -->
2606 <h3>
2608 </h3>
2610 <div>
2611 <p>
2612 LLVM's "eager" JIT compiler is safe to use in threaded programs. Multiple
2615 run code output by the JIT concurrently. The user must still ensure that only
2617 might be modifying it. One way to do that is to always hold the JIT lock while
2621 </p>
2623 <p>When the JIT is configured to compile lazily (using
2626 updating call sites after a function is lazily-jitted. It's still possible to
2627 use the lazy JIT in a threaded program if you ensure that only one thread at a
2628 time can call any particular lazy stub and that the JIT lock guards any IR
2629 access, but we suggest using only the eager JIT in threaded programs.
2630 </p>
2631 </div>
2633 </div>
2635 <!-- *********************************************************************** -->
2636 <h2>
2638 </h2>
2639 <!-- *********************************************************************** -->
2641 <div>
2642 <p>
2643 This section describes some of the advanced or obscure API's that most clients
2644 do not need to be aware of. These API's tend manage the inner workings of the
2645 LLVM system, and only need to be accessed in unusual circumstances.
2646 </p>
2648 <!-- ======================================================================= -->
2649 <h3>
2651 </h3>
2653 <div>
2655 <p>
2656 The LLVM type system has a very simple goal: allow clients to compare types for
2657 structural equality with a simple pointer comparison (aka a shallow compare).
2658 This goal makes clients much simpler and faster, and is used throughout the LLVM
2659 system.
2660 </p>
2662 <p>
2663 Unfortunately achieving this goal is not a simple matter. In particular,
2664 recursive types and late resolution of opaque types makes the situation very
2665 difficult to handle. Fortunately, for the most part, our implementation makes
2666 most clients able to be completely unaware of the nasty internal details. The
2667 primary case where clients are exposed to the inner workings of it are when
2668 building a recursive type. In addition to this case, the LLVM bitcode reader,
2669 assembly parser, and linker also have to be aware of the inner workings of this
2670 system.
2671 </p>
2673 <p>
2674 For our purposes below, we need three concepts. First, an "Opaque Type" is
2677 opaque type as part of its type graph (for example "<tt>{ opaque, i32 }</tt>").
2678 Third, a concrete type is a type that is not an abstract type (e.g. "<tt>{ i32,
2679 float }</tt>").
2680 </p>
2682 <!-- ______________________________________________________________________ -->
2683 <h4>
2685 </h4>
2687 <div>
2689 <p>
2690 Because the most common question is "how do I build a recursive type with LLVM",
2691 we answer it now and explain it as we go. Here we include enough to cause this
2692 to be emitted to an output .ll file:
2693 </p>
2696 <pre>
2697 %mylist = type { %mylist*, i32 }
2698 </pre>
2699 </div>
2701 <p>
2702 To build this, use the following LLVM APIs:
2703 </p>
2706 <pre>
2710 Elts.push_back(PointerType::getUnqual(StructTy));
2711 Elts.push_back(Type::Int32Ty);
2712 StructType *NewSTy = StructType::get(Elts);
2716 cast<OpaqueType>(StructTy.get())-><a href="#refineAbstractTypeTo">refineAbstractTypeTo</a>(NewSTy);
2718 // <i>NewSTy is potentially invalidated, but StructTy (a <a href="#PATypeHolder">PATypeHolder</a>) is</i>
2724 </pre>
2725 </div>
2727 <p>
2728 This code shows the basic approach used to build recursive types: build a
2729 non-recursive type using 'opaque', then use type unification to close the cycle.
2730 The type unification step is performed by the <tt><a
2732 described next. After that, we describe the <a
2734 </p>
2736 </div>
2738 <!-- ______________________________________________________________________ -->
2739 <h4>
2741 </h4>
2743 <div>
2744 <p>
2746 While this method is actually a member of the DerivedType class, it is most
2747 often used on OpaqueType instances. Type unification is actually a recursive
2748 process. After unification, types can become structurally isomorphic to
2749 existing types, and all duplicates are deleted (to preserve pointer equality).
2750 </p>
2752 <p>
2753 In the example above, the OpaqueType object is definitely deleted.
2754 Additionally, if there is an "{ \2*, i32}" type already created in the system,
2756 a type is deleted, any "Type*" pointers in the program are invalidated. As
2759 types can never move or be deleted). To deal with this, the <a
2761 reference to a possibly refined type, and the <a
2763 complex datastructures.
2764 </p>
2766 </div>
2768 <!-- ______________________________________________________________________ -->
2769 <h4>
2771 </h4>
2773 <div>
2774 <p>
2775 PATypeHolder is a form of a "smart pointer" for Type objects. When VMCore
2776 happily goes about nuking types that become isomorphic to existing types, it
2777 automatically updates all PATypeHolder objects to point to the new type. In the
2778 example above, this allows the code to maintain a pointer to the resultant
2779 resolved recursive type, even though the Type*'s are potentially invalidated.
2780 </p>
2782 <p>
2783 PATypeHolder is an extremely light-weight object that uses a lazy union-find
2784 implementation to update pointers. For example the pointer from a Value to its
2785 Type is maintained by PATypeHolder objects.
2786 </p>
2788 </div>
2790 <!-- ______________________________________________________________________ -->
2791 <h4>
2793 </h4>
2795 <div>
2797 <p>
2798 Some data structures need more to perform more complex updates when types get
2799 resolved. To support this, a class can derive from the AbstractTypeUser class.
2800 This class
2801 allows it to get callbacks when certain types are resolved. To register to get
2802 callbacks for a particular type, the DerivedType::{add/remove}AbstractTypeUser
2803 methods can be called on a type. Note that these methods only work for <i>
2804 abstract</i> types. Concrete types (those that do not include any opaque
2805 objects) can never be refined.
2806 </p>
2807 </div>
2809 </div>
2811 <!-- ======================================================================= -->
2812 <h3>
2815 </h3>
2817 <div>
2825 names for types.</p>
2828 by most clients. It should only be used when iteration over the symbol table
2829 names themselves are required, which is very special purpose. Note that not
2830 all LLVM
2832 an empty name) do not exist in the symbol table.
2833 </p>
2835 <p>These symbol tables support iteration over the values/types in the symbol
2840 appropriate symbol table. For types, use the Module::addTypeName method to
2841 insert entries into the symbol table.</p>
2843 </div>
2847 <!-- ======================================================================= -->
2848 <h3>
2850 </h3>
2852 <div>
2858 addition and removal.</p>
2860 <!-- ______________________________________________________________________ -->
2861 <h4>
2864 </a>
2865 </h4>
2867 <div>
2868 <p>
2870 or refer to them out-of-line by means of a pointer. A mixed variant
2873 </p>
2875 <p>
2877 <ul>
2878 <li><p>Layout a)
2880 object and there are a fixed number of them.</p>
2882 <li><p>Layout b)
2885 number of them.</p>
2886 </ul>
2887 <p>
2888 As of v2.4 each layout still possesses a direct pointer to the
2890 we stick to this redundancy for the sake of simplicity.
2892 has. (Theoretically this information can also be calculated
2893 given the scheme presented below.)</p>
2894 <p>
2896 enforce the following memory layouts:</p>
2898 <ul>
2899 <li><p>Layout a) is modelled by prepending the <tt>User</tt> object by the <tt>Use[]</tt> array.</p>
2901 <pre>
2902 ...---.---.---.---.-------...
2903 | P | P | P | P | User
2904 '''---'---'---'---'-------'''
2905 </pre>
2908 <pre>
2909 .-------...
2910 | User
2911 '-------'''
2912 |
2913 v
2914 .---.---.---.---...
2915 | P | P | P | P |
2916 '---'---'---'---'''
2917 </pre>
2918 </ul>
2922 </div>
2924 <!-- ______________________________________________________________________ -->
2925 <h4>
2927 </h4>
2929 <div>
2930 <p>
2933 recover it. This is accomplished by the following scheme:</p>
2935 A bit-encoding in the 2 LSBits (least significant bits) of the <tt>Use::Prev</tt> allows to find the
2937 <ul>
2942 </ul>
2943 <p>
2946 we have to walk to the next stop picking up digits
2947 and calculating the offset:</p>
2948 <pre>
2949 .---.---.---.---.---.---.---.---.---.---.---.---.---.---.---.---.----------------
2951 '---'---'---'---'---'---'---'---'---'---'---'---'---'---'---'---'----------------
2953 | | | | |__>
2954 | | | |__________>
2955 | | |______________________>
2956 | |______________________________________>
2957 |__________________________________________________________>
2958 </pre>
2959 <p>
2960 Only the significant number of bits need to be stored between the
2964 </div>
2966 <!-- ______________________________________________________________________ -->
2967 <h4>
2969 </h4>
2971 <div>
2972 <p>
2973 The following literate Haskell fragment demonstrates the concept:</p>
2976 <pre>
2977 > import Test.QuickCheck
2978 >
2983 >
2989 >
2990 > takeLast n ss = reverse $ take n $ reverse ss
2991 >
2993 >
2994 </pre>
2995 </div>
2996 <p>
2998 <p>
2999 The reverse algorithm computes the length of the string just by examining
3000 a certain prefix:</p>
3003 <pre>
3008 >
3011 > decode walk acc _ = walk + acc
3012 >
3013 </pre>
3014 </div>
3015 <p>
3017 <p>
3021 <pre>
3023 > testcaseLength = length testcase
3024 >
3026 > where arr = takeLast n testcase
3027 >
3028 </pre>
3029 </div>
3030 <p>
3033 <pre>
3034 *Main> quickCheck identityProp
3035 OK, passed 100 tests.
3036 </pre>
3037 <p>
3038 Let's be a bit more exhaustive:</p>
3041 <pre>
3042 >
3044 >
3045 </pre>
3046 </div>
3047 <p>
3050 <pre>
3051 *Main> deepCheck identityProp
3052 OK, passed 500 tests.
3053 </pre>
3055 </div>
3057 <!-- ______________________________________________________________________ -->
3058 <h4>
3060 </h4>
3062 <div>
3064 <p>
3068 tag bits.</p>
3069 <p>
3073 the LSBit set. (Portability is relying on the fact that all known compilers place the
3076 </div>
3078 </div>
3080 </div>
3082 <!-- *********************************************************************** -->
3083 <h2>
3085 </h2>
3086 <!-- *********************************************************************** -->
3088 <div>
3092 <p>The Core LLVM classes are the primary means of representing the program
3093 being inspected or transformed. The core LLVM classes are defined in
3097 <!-- ======================================================================= -->
3098 <h3>
3100 </h3>
3102 <div>
3108 subclasses. They are hidden because they offer no useful functionality beyond
3112 named, but this is not a requirement. There exists exactly
3113 one instance of a given shape at any one time. This allows type equality to
3114 be performed with address equality of the Type Instance. That is, given two
3116 </p>
3118 <!-- _______________________________________________________________________ -->
3119 <h4>
3121 </h4>
3123 <div>
3125 <ul>
3129 floating point types.</li>
3132 an OpaqueType anywhere in its definition).</li>
3135 that don't have a size are abstract types, labels and void.</li>
3137 </ul>
3138 </div>
3140 <!-- _______________________________________________________________________ -->
3141 <h4>
3143 </h4>
3144 <div>
3145 <dl>
3147 <dd>Subclass of DerivedType that represents integer types of any bit width.
3150 <ul>
3152 type of a specific bit width.</li>
3154 type.</li>
3155 </ul>
3156 </dd>
3158 <dd>This is subclassed by ArrayType and PointerType
3159 <ul>
3161 of the elements in the sequential type. </li>
3162 </ul>
3163 </dd>
3165 <dd>This is a subclass of SequentialType and defines the interface for array
3166 types.
3167 <ul>
3169 elements in the array. </li>
3170 </ul>
3171 </dd>
3175 <dd>Subclass of SequentialType for vector types. A
3176 vector type is similar to an ArrayType but is distinguished because it is
3177 a first class type whereas ArrayType is not. Vector types are used for
3178 vector operations and are usually small vectors of of an integer or floating
3179 point type.</dd>
3183 <dd>Subclass of DerivedTypes for function types.
3184 <ul>
3186 function</li>
3188 return type of the function.</li>
3190 the type of the ith parameter.</li>
3192 number of formal parameters.</li>
3193 </ul>
3194 </dd>
3196 <dd>Sublcass of DerivedType for abstract types. This class
3197 defines no content and is used as a placeholder for some other type. Note
3198 that OpaqueType is used (temporarily) during type resolution for forward
3199 references of types. Once the referenced type is resolved, the OpaqueType
3200 is replaced with the actual type. OpaqueType can also be used for data
3201 abstraction. At link time opaque types can be resolved to actual types
3202 of the same name.</dd>
3203 </dl>
3204 </div>
3206 </div>
3208 <!-- ======================================================================= -->
3209 <h3>
3211 </h3>
3213 <div>
3220 programs. An LLVM module is effectively either a translation unit of the
3221 original program or a combination of several translation units merged by the
3226 helpful member functions that try to make common operations easy.</p>
3228 <!-- _______________________________________________________________________ -->
3229 <h4>
3231 </h4>
3233 <div>
3235 <ul>
3237 </ul>
3240 provide a name for it (probably based on the name of the translation unit).</p>
3242 <ul>
3249 <p>These are forwarding methods that make it easy to access the contents of
3251 list.</p></li>
3256 necessary to use when you need to update the list or perform a complex
3257 action that doesn't have a forwarding method.</p>
3260 </ul>
3262 <hr>
3264 <ul>
3272 <p> These are forwarding methods that make it easy to access the contents of
3280 use when you need to update the list or perform a complex action that
3281 doesn't have a forwarding method.</p>
3284 </ul>
3286 <hr>
3288 <ul>
3295 </ul>
3297 <hr>
3299 <ul>
3312 external declaration for the function and return it.</p></li>
3319 string.</p></li>
3326 name, true is returned and the <a
3328 </ul>
3330 </div>
3332 </div>
3334 <!-- ======================================================================= -->
3335 <h3>
3337 </h3>
3339 <div>
3342 <br>
3346 base. It represents a typed value that may be used (among other things) as an
3354 for a program. For example, an incoming argument to a function (represented
3356 every instruction in the function that references the argument. To keep track
3362 methods, shown below.</p>
3366 method. In addition, all LLVM values can be named. The "name" of the
3370 <pre>
3372 </pre>
3373 </div>
3376 that the name of any value may be missing (an empty string), so names should
3378 debugging printouts), they should not be used to keep track of values or map
3382 <p>One important aspect of LLVM is that there is no distinction between an SSA
3383 variable and the operation that produces it. Because of this, any reference to
3384 the value produced by an instruction (or the value available as an incoming
3385 argument, for example) is represented as a direct pointer to the instance of
3386 the class that
3387 represents this value. Although this may take some getting used to, it
3388 simplifies the representation and makes it easier to manipulate.</p>
3390 <!-- _______________________________________________________________________ -->
3391 <h4>
3393 </h4>
3395 <div>
3397 <ul>
3399 use-list<br>
3401 the use-list<br>
3403 value.<br>
3406 the use-list.<br>
3408 use-list.<br>
3410 element in the list.
3411 <p> These methods are the interface to access the def-use
3412 information in LLVM. As with all other iterators in LLVM, the naming
3414 </li>
3417 </li>
3423 </li>
3428 "<tt>V</tt>" instead. For example, if you detect that an instruction always
3429 produces a constant value (for example through constant folding), you can
3430 replace all uses of the instruction with the constant like this:</p>
3433 <pre>
3434 Inst->replaceAllUsesWith(ConstVal);
3435 </pre>
3436 </div>
3438 </ul>
3440 </div>
3442 </div>
3444 <!-- ======================================================================= -->
3445 <h3>
3447 </h3>
3449 <div>
3451 <p>
3464 Single Assignment (SSA) form, there can only be one definition referred to,
3465 allowing this direct connection. This connection provides the use-def
3466 information in LLVM.</p>
3468 <!-- _______________________________________________________________________ -->
3469 <h4>
3471 </h4>
3473 <div>
3476 an index access interface and through an iterator based interface.</p>
3478 <ul>
3482 convenient form for direct access.</p></li>
3485 list<br>
3487 the operand list.<br>
3489 operand list.
3490 <p> Together, these methods make up the iterator based interface to
3492 </ul>
3494 </div>
3496 </div>
3498 <!-- ======================================================================= -->
3499 <h3>
3501 </h3>
3503 <div>
3512 instructions. It provides only a few methods, but is a very commonly used
3514 opcode (instruction type) and the parent <a
3516 into. To represent a specific type of instruction, one of many subclasses of
3525 file contains some meta-data about the various different types of instructions
3526 in LLVM. It describes the enum values that are used as opcodes (for example
3531 this file confuses doxygen, so these enum values don't show up correctly in the
3534 <!-- _______________________________________________________________________ -->
3535 <h4>
3538 </a>
3539 </h4>
3540 <div>
3541 <ul>
3543 <p>This subclasses represents all two operand instructions whose operands
3544 must be the same type, except for the comparison instructions.</p></li>
3547 common operations on cast instructions.</p>
3549 <p>This subclass respresents the two comparison instructions,
3553 <p>This subclass is the parent of all terminator instructions (those which
3554 can terminate a block).</p>
3555 </ul>
3556 </div>
3558 <!-- _______________________________________________________________________ -->
3559 <h4>
3562 </a>
3563 </h4>
3565 <div>
3567 <ul>
3572 <p>Returns true if the instruction writes to memory, i.e. it is a
3577 <p>Returns another instance of the specified instruction, identical
3578 in all ways to the original except that the instruction has no parent
3580 and it has no name</p></li>
3581 </ul>
3583 </div>
3585 </div>
3587 <!-- ======================================================================= -->
3588 <h3>
3590 </h3>
3592 <div>
3594 <p>Constant represents a base class for different types of constants. It
3595 is subclassed by ConstantInt, ConstantArray, etc. for representing
3597 a subclass, which represents the address of a global variable or function.
3598 </p>
3600 <!-- _______________________________________________________________________ -->
3602 <div>
3603 <ul>
3604 <li>ConstantInt : This subclass of Constant represents an integer constant of
3605 any width.
3606 <ul>
3608 value of this constant, an APInt value.</li>
3610 value to an int64_t via sign extension. If the value (not the bit width)
3611 of the APInt is too large to fit in an int64_t, an assertion will result.
3612 For this reason, use of this method is discouraged.</li>
3614 value to a uint64_t via zero extension. IF the value (not the bit width)
3615 of the APInt is too large to fit in a uint64_t, an assertion will result.
3616 For this reason, use of this method is discouraged.</li>
3619 The type is implied as the IntegerType that corresponds to the bit width
3622 Returns the ConstantInt object that represents the value provided by
3624 </ul>
3625 </li>
3626 <li>ConstantFP : This class represents a floating point constant.
3627 <ul>
3629 this constant. </li>
3630 </ul>
3631 </li>
3632 <li>ConstantArray : This represents a constant array.
3633 <ul>
3635 a vector of component constants that makeup this array. </li>
3636 </ul>
3637 </li>
3638 <li>ConstantStruct : This represents a constant struct.
3639 <ul>
3641 a vector of component constants that makeup this array. </li>
3642 </ul>
3643 </li>
3644 <li>GlobalValue : This represents either a global variable or a function. In
3645 either case, the value is a constant fixed address (after linking).
3646 </li>
3647 </ul>
3648 </div>
3650 </div>
3652 <!-- ======================================================================= -->
3653 <h3>
3655 </h3>
3657 <div>
3662 Class</a><br>
3669 Because they are visible at global scope, they are also subject to linking with
3670 other globals defined in different translation units. To control the linking
3677 unit, and does not participate in linking. If it has external linkage, it is
3678 visible to external code, and does participate in linking. In addition to
3684 global is always a pointer to its contents. It is important to remember this
3689 the address of the first element of this array and the value of the
3697 <!-- _______________________________________________________________________ -->
3698 <h4>
3701 </a>
3702 </h4>
3704 <div>
3706 <ul>
3711 <p> </p>
3712 </li>
3715 GlobalValue is currently embedded into.</p></li>
3716 </ul>
3718 </div>
3720 </div>
3722 <!-- ======================================================================= -->
3723 <h3>
3725 </h3>
3727 <div>
3738 actually one of the more complex classes in the LLVM hierarchy because it must
3746 ordering of the blocks in the function, which indicate how the code will be
3747 laid out by the backend. Additionally, the first <a
3750 block. There are no implicit exit nodes, and in fact there may be multiple exit
3754 function hasn't been linked in yet.</p>
3764 LLVM feature that is only used when you have to look up a value by name. Aside
3766 internally to make sure that there are not conflicts between the names of <a
3773 is its address (after linking) which is guaranteed to be constant.</p>
3775 <!-- _______________________________________________________________________ -->
3776 <h4>
3779 </a>
3780 </h4>
3782 <div>
3784 <ul>
3789 the the program. The constructor must specify the type of the function to
3790 create and what type of linkage the function should have. The <a
3792 specifies the formal arguments and return value for the function. The same
3795 in which the function is defined. If this argument is provided, the function
3796 will automatically be inserted into that module's list of
3797 functions.</p></li>
3802 function is "external", it does not have a body, and thus must be resolved
3803 by linking with a function defined in a different translation unit.</p></li>
3811 <p>These are forwarding methods that make it easy to access the contents of
3813 list.</p></li>
3818 is necessary to use when you need to update the list or perform a complex
3819 action that doesn't have a forwarding method.</p></li>
3822 iterator<br>
3828 <p>These are forwarding methods that make it easy to access the contents of
3830 list.</p></li>
3835 necessary to use when you need to update the list or perform a complex
3836 action that doesn't have a forwarding method.</p></li>
3841 function. Because the entry block for the function is always the first
3850 function.</p></li>
3856 </ul>
3858 </div>
3860 </div>
3862 <!-- ======================================================================= -->
3863 <h3>
3865 </h3>
3867 <div>
3871 <br>
3873 Class</a><br>
3879 <p>Global variables are represented with the (surprise surprise)
3882 always referenced by their address (global values must live in memory, so their
3883 "name" refers to their constant address). See
3885 variables may have an initial value (which must be a
3887 they may be marked as "constant" themselves (indicating that their contents
3888 never change at runtime).</p>
3890 <!-- _______________________________________________________________________ -->
3891 <h4>
3894 </a>
3895 </h4>
3897 <div>
3899 <ul>
3904 <p>Create a new global variable of the specified type. If
3906 unchanging for the program. The Linkage parameter specifies the type of
3907 linkage (internal, external, weak, linkonce, appending) for the variable.
3908 If the linkage is InternalLinkage, WeakAnyLinkage, WeakODRLinkage,
3909 LinkOnceAnyLinkage or LinkOnceODRLinkage, then the resultant
3910 global variable will have internal linkage. AppendingLinkage concatenates
3911 together all instances (in different translation units) of the variable
3912 into a single variable but is only applicable to arrays. See
3914 further details on linkage types. Optionally an initializer, a name, and the
3915 module to put the variable into may be specified for the global variable as
3916 well.</p></li>
3920 <p>Returns true if this is a global variable that is known not to
3921 be modified at runtime.</p></li>
3930 to call this method if there is no initializer.</p></li>
3931 </ul>
3933 </div>
3935 </div>
3937 <!-- ======================================================================= -->
3938 <h3>
3940 </h3>
3942 <div>
3947 Class</a><br>
3950 <p>This class represents a single entry single exit section of the code,
3951 commonly known as a basic block by the compiler community. The
3954 Matching the language definition, the last element of this list of instructions
3955 is always a terminator instruction (a subclass of the <a
3958 <p>In addition to tracking the list of instructions that make up the block, the
3967 <!-- _______________________________________________________________________ -->
3968 <h4>
3971 </a>
3972 </h4>
3974 <div>
3975 <ul>
3981 insertion into a function. The constructor optionally takes a name for the new
3984 automatically inserted at the end of the specified <a
3992 STL-style functions for accessing the instruction list.
3994 <p>These methods and typedefs are forwarding functions that have the same
3995 semantics as the standard library methods of the same names. These methods
3996 expose the underlying instruction list of a basic block in a way that is easy to
3997 manipulate. To get the full complement of container operations (including
3999 method.</p></li>
4003 <p>This method is used to get access to the underlying container that actually
4004 holds the Instructions. This method must be used when there isn't a forwarding
4006 to perform. Because there are no forwarding functions for "updating"
4007 operations, you need to use this if you want to update the contents of a
4013 embedded into, or a null pointer if it is homeless.</p></li>
4017 <p> Returns a pointer to the terminator instruction that appears at the end of
4019 instruction in the block is not a terminator, then a null pointer is
4020 returned.</p></li>
4022 </ul>
4024 </div>
4026 </div>
4028 <!-- ======================================================================= -->
4029 <h3>
4031 </h3>
4033 <div>
4035 <p>This subclass of Value defines the interface for incoming formal
4036 arguments to a function. A Function maintains a list of its formal
4037 arguments. An argument has a pointer to the parent Function.</p>
4039 </div>
4041 </div>
4043 <!-- *********************************************************************** -->
4044 <hr>
4045 <address>
4054 Last modified: $Date$
4055 </address>
4057 </body>
4058 </html>