| blob | 060bbb6d91b0fa1073aa2e6f7f77c84979716202 |
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>
68 </ul></li>
70 <ul>
80 </ul></li>
82 <ul>
92 </ul></li>
94 <!--<ul>
95 todo
98 <ul>
102 </ul></li>
103 </ul>
104 </li>
106 <ul>
108 <ul>
116 class pointer</a> </li>
118 complex example</a> </li>
120 the same way</a> </li>
122 use-def chains</a> </li>
124 successors of blocks</a></li>
125 </ul>
126 </li>
128 <ul>
135 </ul>
136 </li>
138 <!--
139 <li>Working with the Control Flow Graph
140 <ul>
141 <li>Accessing predecessors and successors of a <tt>BasicBlock</tt>
142 <li>
143 <li>
144 </ul>
145 -->
146 </ul>
147 </li>
150 <ul>
152 </a></li>
157 </ul>
158 </li>
161 <ul>
163 <ul>
168 </ul></li>
170 <li><a href="#SymbolTable">The <tt>ValueSymbolTable</tt> and <tt>TypeSymbolTable</tt> classes</a></li>
171 <li><a href="#UserLayout">The <tt>User</tt> and owned <tt>Use</tt> classes' memory layout</a></li>
172 </ul></li>
175 <ul>
179 <ul>
181 <ul>
184 <ul>
186 <ul>
189 </ul>
190 </li>
191 </ul>
192 </li>
193 </ul>
194 </li>
197 </ul>
198 </li>
199 </ul>
200 </li>
201 </ol>
210 </div>
212 <!-- *********************************************************************** -->
213 <h2>
215 </h2>
216 <!-- *********************************************************************** -->
218 <div>
220 <p>This document is meant to highlight some of the important classes and
221 interfaces available in the LLVM source-base. This manual is not
222 intended to explain what LLVM is, how it works, and what LLVM code looks
223 like. It assumes that you know the basics of LLVM and are interested
224 in writing transformations or otherwise analyzing or manipulating the
225 code.</p>
227 <p>This document should get you oriented so that you can find your
228 way in the continuously growing source code that makes up the LLVM
229 infrastructure. Note that this manual is not intended to serve as a
230 replacement for reading the source code, so if you think there should be
231 a method in one of these classes to do something, but it's not listed,
233 are provided to make this as easy as possible.</p>
235 <p>The first section of this document describes general information that is
236 useful to know when working in the LLVM infrastructure, and the second describes
237 the Core LLVM classes. In the future this manual will be extended with
238 information describing how to use extension libraries, such as dominator
239 information, CFG traversal routines, and useful utilities like the <tt><a
242 </div>
244 <!-- *********************************************************************** -->
245 <h2>
247 </h2>
248 <!-- *********************************************************************** -->
250 <div>
252 <p>This section contains general information that is useful if you are working
253 in the LLVM source-base, but that isn't specific to any particular API.</p>
255 <!-- ======================================================================= -->
256 <h3>
258 </h3>
260 <div>
262 <p>LLVM makes heavy use of the C++ Standard Template Library (STL),
263 perhaps much more than you are used to, or have seen before. Because of
264 this, you might want to do a little background reading in the
265 techniques used and capabilities of the library. There are many good
266 pages that discuss the STL, and several books on the subject that you
267 can get, so it will not be discussed in this document.</p>
271 <ol>
274 C++ Library reference</a> - an excellent reference for the STL and other parts
275 of the standard C++ library.</li>
278 O'Reilly book in the making. It has a decent Standard Library
279 Reference that rivals Dinkumware's, and is unfortunately no longer free since the
280 book has been published.</li>
283 Questions</a></li>
286 Contains a useful <a
291 Page</a></li>
295 the book).</a></li>
297 </ol>
301 to write maintainable code more than where to put your curly braces.</p>
303 </div>
305 <!-- ======================================================================= -->
306 <h3>
308 </h3>
310 <div>
312 <ol>
314 static and shared libraries across platforms</a></li>
315 </ol>
317 </div>
319 </div>
321 <!-- *********************************************************************** -->
322 <h2>
324 </h2>
325 <!-- *********************************************************************** -->
327 <div>
329 <p>Here we highlight some LLVM APIs that are generally useful and good to
330 know about when writing transformations.</p>
332 <!-- ======================================================================= -->
333 <h3>
336 </h3>
338 <div>
340 <p>The LLVM source-base makes extensive use of a custom form of RTTI.
342 operator, but they don't have some drawbacks (primarily stemming from
344 have a v-table). Because they are used so often, you must know what they
345 do and how they work. All of these templates are defined in the <a
347 file (note that you very rarely have to include this file directly).</p>
349 <dl>
353 "<tt>instanceof</tt>" operator. It returns true or false depending on whether
354 a reference or pointer points to an instance of the specified class. This can
355 be very useful for constraint checking of various sorts (example below).</p>
356 </dd>
361 converts a pointer or reference from a base class to a derived class, causing
362 an assertion failure if it is not really an instance of the right type. This
363 should be used in cases where you have some information that makes you believe
368 <pre>
370 if (isa<<a href="#Constant">Constant</a>>(V) || isa<<a href="#Argument">Argument</a>>(V) || isa<<a href="#GlobalValue">GlobalValue</a>>(V))
371 return true;
375 }
376 </pre>
377 </div>
381 operator.</p>
383 </dd>
388 It checks to see if the operand is of the specified type, and if so, returns a
389 pointer to it (this operator does not work with references). If the operand is
390 not of the correct type, a null pointer is returned. Thus, this works very
394 statement like this:</p>
397 <pre>
398 if (<a href="#AllocationInst">AllocationInst</a> *AI = dyn_cast<<a href="#AllocationInst">AllocationInst</a>>(Val)) {
400 }
401 </pre>
402 </div>
406 statement, which is very convenient.</p>
411 blocks to check for lots of different variants of classes. If you find
412 yourself wanting to do this, it is much cleaner and more efficient to use the
415 </dd>
421 argument (which it then propagates). This can sometimes be useful, allowing
422 you to combine several null checks into one.</p></dd>
428 as an argument (which it then propagates). This can sometimes be useful,
429 allowing you to combine several null checks into one.</p></dd>
431 </dl>
433 <p>These five templates can be used with any classes, whether they have a
434 v-table or not. To add support for these templates, you simply need to add
436 to. Describing this is currently outside the scope of this document, but there
437 are lots of examples in the LLVM source base.</p>
439 </div>
442 <!-- ======================================================================= -->
443 <h3>
446 </h3>
448 <div>
450 <p>Although LLVM generally does not do much string manipulation, we do have
451 several important APIs which take strings. Two important examples are the
452 Value class -- which has names for instructions, functions, etc. -- and the
453 StringMap class which is used extensively in LLVM and Clang.</p>
455 <p>These are generic classes, and they need to be able to accept strings which
456 may have embedded null characters. Therefore, they cannot simply take
458 clients to perform a heap allocation which is usually unnecessary. Instead,
460 passing strings efficiently.</p>
462 <!-- _______________________________________________________________________ -->
463 <h4>
465 </h4>
467 <div>
470 (a character array and a length) and supports the common operations available
473 <p>It can be implicitly constructed using a C style null-terminated string,
478 iterator find(StringRef Key);
479 </pre>
487 </pre>
493 for more information.</p>
496 pointers to external memory it is not generally safe to store an instance of the
497 class (unless you know that the external storage will not be freed). StringRef is
498 small and pervasive enough in LLVM that it should always be passed by value.</p>
500 </div>
502 <!-- _______________________________________________________________________ -->
503 <h4>
505 </h4>
507 <div>
510 strings. For example, a common LLVM paradigm is to name one instruction based on
511 the name of another instruction with a suffix, for example:</p>
514 <pre>
516 </pre>
517 </div>
521 which points to temporary (stack allocated) objects. Twines can be implicitly
522 constructed as the result of the plus operator applied to strings (i.e., a C
524 actual concatenation of strings until it is actually required, at which point
525 it can be efficiently rendered directly into a character array. This avoids
526 unnecessary heap allocation involved in constructing the temporary results of
527 string concatenation. See
529 for more information.</p>
532 and should almost never be stored or mentioned directly. They are intended
533 solely for use when defining a function which should be able to efficiently
534 accept concatenated strings.</p>
536 </div>
538 </div>
540 <!-- ======================================================================= -->
541 <h3>
543 </h3>
545 <div>
547 <p>Often when working on your pass you will put a bunch of debugging printouts
548 and other code into your pass. After you get it working, you want to remove
549 it, but you may need it again in the future (to work out new bugs that you run
550 across).</p>
552 <p> Naturally, because of this, you don't want to delete the debug printouts,
553 but you don't want them to always be noisy. A standard compromise is to comment
554 them out, allowing you to enable them if you need them in the future.</p>
558 this problem. Basically, you can put arbitrary code into the argument of the
563 <pre>
565 </pre>
566 </div>
571 <pre>
575 I am here!
576 </pre>
577 </div>
580 to not have to create "yet another" command line option for the debug output for
582 so they do not cause a performance impact at all (for the same reason, they
583 should also not contain side-effects!).</p>
586 enable or disable it directly in gdb. Just use "<tt>set DebugFlag=0</tt>" or
587 "<tt>set DebugFlag=1</tt>" from the gdb if the program is running. If the
588 program hasn't been started yet, you can always just run it with
591 <!-- _______________________________________________________________________ -->
592 <h4>
595 </h4>
597 <div>
601 generator). If you want to enable debug information with more fine-grained
603 option as follows:</p>
606 <pre>
607 #undef DEBUG_TYPE
609 #define DEBUG_TYPE "foo"
611 #undef DEBUG_TYPE
612 #define DEBUG_TYPE "bar"
614 #undef DEBUG_TYPE
615 #define DEBUG_TYPE ""
617 </pre>
618 </div>
623 <pre>
627 No debug type
628 'foo' debug type
629 'bar' debug type
630 No debug type (2)
632 'foo' debug type
634 'bar' debug type
635 </pre>
636 </div>
639 a file, to specify the debug type for the entire module (if you do this before
642 "bar", because there is no system in place to ensure that names do not
643 conflict. If two different modules use the same string, they will all be turned
644 on when the name is specified. This allows, for example, all debug information
646 even if the source lives in multiple files.</p>
650 statement. It takes an additional first parameter, which is the type to use. For
651 example, the preceding example could be written as:</p>
655 <pre>
660 </pre>
661 </div>
663 </div>
665 </div>
667 <!-- ======================================================================= -->
668 <h3>
670 option</a>
671 </h3>
673 <div>
678 keep track of what the LLVM compiler is doing and how effective various
679 optimizations are. It is useful to see what optimizations are contributing to
680 making a particular program run faster.</p>
682 <p>Often you may run your pass on some big program, and you're interested to see
683 how many times it makes a certain transformation. Although you can do this with
684 hand inspection, or some ad-hoc method, this is a real pain and not very useful
686 keep track of this information, and the calculated information is presented in a
687 uniform manner with the rest of the passes being executed.</p>
690 it are as follows:</p>
692 <ol>
696 <pre>
697 #define <a href="#DEBUG_TYPE">DEBUG_TYPE</a> "mypassname" <i>// This goes before any #includes.</i>
698 STATISTIC(NumXForms, "The # of times I did stuff");
699 </pre>
700 </div>
703 specified by the first argument. The pass name is taken from the DEBUG_TYPE
704 macro, and the description is taken from the second argument. The variable
710 <pre>
712 </pre>
713 </div>
715 </li>
716 </ol>
722 <pre>
725 </pre>
726 </div>
729 suite, it gives a report that looks like this:</p>
732 <pre>
733 7646 bitcodewriter - Number of normal instructions
734 725 bitcodewriter - Number of oversized instructions
735 129996 bitcodewriter - Number of bitcode bytes written
736 2817 raise - Number of insts DCEd or constprop'd
737 3213 raise - Number of cast-of-self removed
738 5046 raise - Number of expression trees converted
739 75 raise - Number of other getelementptr's formed
740 138 raise - Number of load/store peepholes
741 42 deadtypeelim - Number of unused typenames removed from symtab
742 392 funcresolve - Number of varargs functions resolved
743 27 globaldce - Number of global variables removed
744 2 adce - Number of basic blocks removed
745 134 cee - Number of branches revectored
746 49 cee - Number of setcc instruction eliminated
747 532 gcse - Number of loads removed
748 2919 gcse - Number of instructions removed
749 86 indvars - Number of canonical indvars added
750 87 indvars - Number of aux indvars removed
751 25 instcombine - Number of dead inst eliminate
752 434 instcombine - Number of insts combined
753 248 licm - Number of load insts hoisted
754 1298 licm - Number of insts hoisted to a loop pre-header
755 3 licm - Number of insts hoisted to multiple loop preds (bad, no loop pre-header)
756 75 mem2reg - Number of alloca's promoted
757 1444 cfgsimplify - Number of blocks simplified
758 </pre>
759 </div>
761 <p>Obviously, with so many optimizations, having a unified framework for this
762 stuff is very nice. Making your pass fit well into the framework makes it more
763 maintainable and useful.</p>
765 </div>
767 <!-- ======================================================================= -->
768 <h3>
770 </h3>
772 <div>
774 <p>Several of the important data structures in LLVM are graphs: for example
778 DAGs</a>. In many cases, while debugging various parts of the compiler, it is
779 nice to instantly visualize these graphs.</p>
781 <p>LLVM provides several callbacks that are available in a debug build to do
783 the current LLVM tool will pop up a window containing the CFG for the function
784 where each basic block is a node in the graph, and each node contains the
785 instructions in the block. Similarly, there also exists
790 up a window. Alternatively, you can sprinkle calls to these functions in your
791 code in places you want to debug.</p>
793 <p>Getting this to work requires a small amount of configuration. On Unix
795 toolkit, and make sure 'dot' and 'gv' are in your path. If you are running on
796 Mac OS/X, download and install the Mac OS/X <a
799 it) to your path. Once in your system and path are set up, rerun the LLVM
800 configure script and rebuild LLVM to enable this functionality.</p>
806 specified color (choices of colors can be found at <a
808 complex node attributes can be provided with <tt>call
811 Attributes</a>.) If you want to restart and clear all the current graph
814 </div>
816 </div>
818 <!-- *********************************************************************** -->
819 <h2>
821 </h2>
822 <!-- *********************************************************************** -->
824 <div>
827 and we commonly use STL data structures. This section describes the trade-offs
828 you should consider when you pick one.</p>
830 <p>
831 The first step is a choose your own adventure: do you want a sequential
832 container, a set-like container, or a map-like container? The most important
833 thing when choosing a container is the algorithmic properties of how you plan to
834 access the container. Based on that, you should use:</p>
836 <ul>
838 of an value based on another value. Map-like containers also support
839 efficient queries for containment (whether a key is in the map). Map-like
840 containers generally do not support efficient reverse mapping (values to
841 keys). If you need that, use two maps. Some map-like containers also
842 support efficient iteration through the keys in sorted order. Map-like
843 containers are the most expensive sort, only use them if you need one of
844 these capabilities.</li>
847 stuff into a container that automatically eliminates duplicates. Some
848 set-like containers support efficient iteration through the elements in
849 sorted order. Set-like containers are more expensive than sequential
850 containers.
851 </li>
854 the most efficient way to add elements and keeps track of the order they are
855 added to the collection. They permit duplicates and support efficient
856 iteration, but do not support efficient look-up based on a key.
857 </li>
860 container or reference structure that is used for character or byte
861 arrays.</li>
864 perform set operations on sets of numeric id's, while automatically
865 eliminating duplicates. Bit containers require a maximum of 1 bit for each
866 identifier you want to store.
867 </li>
868 </ul>
870 <p>
871 Once the proper category of container is determined, you can fine tune the
872 memory use, constant factors, and cache behaviors of access by intelligently
873 picking a member of the category. Note that constant factors and cache behavior
874 can be a big deal. If you have a vector that usually only contains a few
875 elements (but could contain many), for example, it's much better to use
877 . Doing so avoids (relatively) expensive malloc/free calls, which dwarf the
878 cost of adding the elements to the container. </p>
880 <!-- ======================================================================= -->
881 <h3>
883 </h3>
885 <div>
886 There are a variety of sequential containers available for you, based on your
887 needs. Pick the first in this section that will do what you want.
889 <!-- _______________________________________________________________________ -->
890 <h4>
892 </h4>
894 <div>
895 <p>The llvm::ArrayRef class is the preferred class to use in an interface that
896 accepts a sequential list of elements in memory and just reads from them. By
897 taking an ArrayRef, the API can be passed a fixed size array, an std::vector,
898 an llvm::SmallVector and anything else that is contiguous in memory.
899 </p>
900 </div>
904 <!-- _______________________________________________________________________ -->
905 <h4>
907 </h4>
909 <div>
910 <p>Fixed size arrays are very simple and very fast. They are good if you know
911 exactly how many elements you have, or you have a (low) upper bound on how many
912 you have.</p>
913 </div>
915 <!-- _______________________________________________________________________ -->
916 <h4>
918 </h4>
920 <div>
921 <p>Heap allocated arrays (new[] + delete[]) are also simple. They are good if
922 the number of elements is variable, if you know how many elements you will need
923 before the array is allocated, and if the array is usually large (if not,
925 allocated array is the cost of the new/delete (aka malloc/free). Also note that
926 if you are allocating an array of a type with a constructor, the constructor and
927 destructors will be run for every element in the array (re-sizable vectors only
928 construct those elements actually used).</p>
929 </div>
931 <!-- _______________________________________________________________________ -->
932 <h4>
934 </h4>
936 <div>
939 it supports efficient iteration, lays out elements in memory order (so you can
940 do pointer arithmetic between elements), supports efficient push_back/pop_back
941 operations, supports efficient random access to its elements, etc.</p>
943 <p>The advantage of SmallVector is that it allocates space for
945 the SmallVector is dynamically smaller than N, no malloc is performed. This can
946 be a big win in cases where the malloc/free call is far more expensive than the
947 code that fiddles around with the elements.</p>
950 predecessors/successors of a block is usually less than 8). On the other hand,
951 this makes the size of the SmallVector itself large, so you don't want to
952 allocate lots of them (doing so will waste a lot of space). As such,
953 SmallVectors are most useful when on the stack.</p>
955 <p>SmallVector also provides a nice portable and efficient replacement for
958 </div>
960 <!-- _______________________________________________________________________ -->
961 <h4>
963 </h4>
965 <div>
966 <p>
967 std::vector is well loved and respected. It is useful when SmallVector isn't:
968 when the size of the vector is often large (thus the small optimization will
969 rarely be a benefit) or if you will be allocating many instances of the vector
970 itself (which would waste space for elements that aren't in the container).
971 vector is also useful when interfacing with code that expects vectors :).
972 </p>
977 <pre>
978 for ( ... ) {
980 use V;
981 }
982 </pre>
983 </div>
988 <pre>
990 for ( ... ) {
991 use V;
992 V.clear();
993 }
994 </pre>
995 </div>
997 <p>Doing so will save (at least) one heap allocation and free per iteration of
998 the loop.</p>
1000 </div>
1002 <!-- _______________________________________________________________________ -->
1003 <h4>
1005 </h4>
1007 <div>
1008 <p>std::deque is, in some senses, a generalized version of std::vector. Like
1009 std::vector, it provides constant time random access and other similar
1010 properties, but it also provides efficient access to the front of the list. It
1011 does not guarantee continuity of elements within memory.</p>
1013 <p>In exchange for this extra flexibility, std::deque has significantly higher
1014 constant factor costs than std::vector. If possible, use std::vector or
1015 something cheaper.</p>
1016 </div>
1018 <!-- _______________________________________________________________________ -->
1019 <h4>
1021 </h4>
1023 <div>
1024 <p>std::list is an extremely inefficient class that is rarely useful.
1025 It performs a heap allocation for every element inserted into it, thus having an
1026 extremely high constant factor, particularly for small data types. std::list
1027 also only supports bidirectional iteration, not random access iteration.</p>
1029 <p>In exchange for this high cost, std::list supports efficient access to both
1030 ends of the list (like std::deque, but unlike std::vector or SmallVector). In
1031 addition, the iterator invalidation characteristics of std::list are stronger
1032 than that of a vector class: inserting or removing an element into the list does
1033 not invalidate iterator or pointers to other elements in the list.</p>
1034 </div>
1036 <!-- _______________________________________________________________________ -->
1037 <h4>
1039 </h4>
1041 <div>
1043 intrusive, because it requires the element to store and provide access to the
1044 prev/next pointers for the list.</p>
1048 provides some novel characteristics. In particular, it can efficiently store
1049 polymorphic objects, the traits class is informed when an element is inserted or
1051 constant-time splice operation.</p>
1053 <p>These properties are exactly what we want for things like
1057 Related classes of interest are explained in the following subsections:
1058 <ul>
1063 </ul>
1064 </div>
1066 <!-- _______________________________________________________________________ -->
1067 <h4>
1069 </h4>
1071 <div>
1074 publicly derive from this traits class.</p>
1075 </div>
1077 <!-- _______________________________________________________________________ -->
1078 <h4>
1080 </h4>
1082 <div>
1084 supports a slightly narrower interface. Notably, inserters from
1088 used for a wide variety of customizations.</p>
1089 </div>
1091 <!-- _______________________________________________________________________ -->
1092 <h4>
1094 </h4>
1096 <div>
1099 in the default manner.</p>
1104 </div>
1106 <!-- _______________________________________________________________________ -->
1107 <h4>
1109 </h4>
1111 <div>
1113 citizen in the C++ ecosystem, it needs to support the standard container
1118 <p>The only sensible solution to this problem is to allocate a so-called
1120 iterator, providing the back-link to the last element. However conforming to the
1122 also must not be dereferenced.</p>
1125 how to allocate and store the sentinel. The corresponding policy is dictated
1127 whenever the need for a sentinel arises.</p>
1129 <p>While the default policy is sufficient in most cases, it may break down when
1132 is wasted. To alleviate the situation with numerous and voluminous
1138 arithmetic is used to obtain the sentinel, which is relative to the
1140 extra pointer, which serves as the back-link of the sentinel. This is the only
1141 field in the ghostly sentinel which can be legally accessed.</p>
1142 </div>
1144 <!-- _______________________________________________________________________ -->
1145 <h4>
1147 </h4>
1149 <div>
1152 <p>There are also various STL adapter classes such as std::queue,
1153 std::priority_queue, std::stack, etc. These provide simplified access to an
1154 underlying container but don't affect the cost of the container itself.</p>
1156 </div>
1158 </div>
1160 <!-- ======================================================================= -->
1161 <h3>
1163 </h3>
1165 <div>
1167 <p>Set-like containers are useful when you need to canonicalize multiple values
1168 into a single representation. There are several different choices for how to do
1169 this, providing various trade-offs.</p>
1171 <!-- _______________________________________________________________________ -->
1172 <h4>
1174 </h4>
1176 <div>
1178 <p>If you intend to insert a lot of elements, then do a lot of queries, a
1179 great approach is to use a vector (or other sequential container) with
1180 std::sort+std::unique to remove duplicates. This approach works really well if
1181 your usage pattern has these two distinct phases (insert then query), and can be
1183 </p>
1185 <p>
1186 This combination provides the several nice properties: the result data is
1187 contiguous in memory (good for cache locality), has few allocations, is easy to
1188 address (iterators in the final vector are just indices or pointers), and can be
1189 efficiently queried with a standard binary or radix search.</p>
1191 </div>
1193 <!-- _______________________________________________________________________ -->
1194 <h4>
1196 </h4>
1198 <div>
1200 <p>If you have a set-like data structure that is usually small and whose elements
1202 has space for N elements in place (thus, if the set is dynamically smaller than
1203 N, no malloc traffic is required) and accesses them with a simple linear search.
1204 When the set grows beyond 'N' elements, it allocates a more expensive representation that
1205 guarantees efficient access (for most types, it falls back to std::set, but for
1206 pointers it uses something far better, <a
1209 <p>The magic of this class is that it handles small sets extremely efficiently,
1210 but gracefully handles extremely large sets without loss of efficiency. The
1211 drawback is that the interface is quite small: it supports insertion, queries
1212 and erasing, but does not support iteration.</p>
1214 </div>
1216 <!-- _______________________________________________________________________ -->
1217 <h4>
1219 </h4>
1221 <div>
1223 <p>SmallPtrSet has all the advantages of <tt>SmallSet</tt> (and a <tt>SmallSet</tt> of pointers is
1225 more than 'N' insertions are performed, a single quadratically
1226 probed hash table is allocated and grows as needed, providing extremely
1227 efficient access (constant time insertion/deleting/queries with low constant
1228 factors) and is very stingy with malloc traffic.</p>
1231 whenever an insertion occurs. Also, the values visited by the iterators are not
1232 visited in sorted order.</p>
1234 </div>
1236 <!-- _______________________________________________________________________ -->
1237 <h4>
1239 </h4>
1241 <div>
1243 <p>
1244 DenseSet is a simple quadratically probed hash table. It excels at supporting
1245 small values: it uses a single allocation to hold all of the pairs that
1246 are currently inserted in the set. DenseSet is a great way to unique small
1247 values that are not simple pointers (use <a
1249 the same requirements for the value type that <a
1251 </p>
1253 </div>
1255 <!-- _______________________________________________________________________ -->
1256 <h4>
1258 </h4>
1260 <div>
1262 <p>
1263 FoldingSet is an aggregate class that is really good at uniquing
1264 expensive-to-create or polymorphic objects. It is a combination of a chained
1265 hash table with intrusive links (uniqued objects are required to inherit from
1267 its ID process.</p>
1270 a complex object (for example, a node in the code generator). The client has a
1271 description of *what* it wants to generate (it knows the opcode and all the
1272 operands), but we don't want to 'new' a node, then try inserting it into a set
1273 only to find out it already exists, at which point we would have to delete it
1274 and return the node that already exists.
1275 </p>
1277 <p>To support this style of client, FoldingSet perform a query with a
1278 FoldingSetNodeID (which wraps SmallVector) that can be used to describe the
1279 element that we want to query for. The query either returns the element
1280 matching the ID or it returns an opaque ID that indicates where insertion should
1281 take place. Construction of the ID usually does not require heap traffic.</p>
1283 <p>Because FoldingSet uses intrusive links, it can support polymorphic objects
1284 in the set (for example, you can have SDNode instances mixed with LoadSDNodes).
1285 Because the elements are individually allocated, pointers to the elements are
1286 stable: inserting or removing elements does not invalidate any pointers to other
1287 elements.
1288 </p>
1290 </div>
1292 <!-- _______________________________________________________________________ -->
1293 <h4>
1295 </h4>
1297 <div>
1300 many things but great at nothing. std::set allocates memory for each element
1301 inserted (thus it is very malloc intensive) and typically stores three pointers
1302 per element in the set (thus adding a large amount of per-element space
1303 overhead). It offers guaranteed log(n) performance, which is not particularly
1304 fast from a complexity standpoint (particularly if the elements of the set are
1305 expensive to compare, like strings), and has extremely high constant factors for
1306 lookup, insertion and removal.</p>
1308 <p>The advantages of std::set are that its iterators are stable (deleting or
1309 inserting an element from the set does not affect iterators or pointers to other
1310 elements) and that iteration over the set is guaranteed to be in sorted order.
1311 If the elements in the set are large, then the relative overhead of the pointers
1312 and malloc traffic is not a big deal, but if the elements of the set are small,
1313 std::set is almost never a good choice.</p>
1315 </div>
1317 <!-- _______________________________________________________________________ -->
1318 <h4>
1320 </h4>
1322 <div>
1325 Container</a>. The important property
1326 that this provides is efficient insertion with uniquing (duplicate elements are
1327 ignored) with iteration support. It implements this by inserting elements into
1328 both a set-like container and the sequential container, using the set-like
1329 container for uniquing and the sequential container for iteration.
1330 </p>
1332 <p>The difference between SetVector and other sets is that the order of
1333 iteration is guaranteed to match the order of insertion into the SetVector.
1334 This property is really important for things like sets of pointers. Because
1335 pointer values are non-deterministic (e.g. vary across runs of the program on
1336 different machines), iterating over the pointers in the set will
1337 not be in a well-defined order.</p>
1339 <p>
1340 The drawback of SetVector is that it requires twice as much space as a normal
1341 set and has the sum of constant factors from the set-like container and the
1342 sequential container that it uses. Use it *only* if you need to iterate over
1343 the elements in a deterministic order. SetVector is also expensive to delete
1344 elements out of (linear time), unless you use it's "pop_back" method, which is
1345 faster.
1346 </p>
1348 <p>SetVector is an adapter class that defaults to using std::vector and std::set
1349 for the underlying containers, so it is quite expensive. However,
1351 defaults to using a SmallVector and SmallSet of a specified size. If you use
1352 this, and if your sets are dynamically smaller than N, you will save a lot of
1353 heap traffic.</p>
1355 </div>
1357 <!-- _______________________________________________________________________ -->
1358 <h4>
1360 </h4>
1362 <div>
1364 <p>
1366 retains a unique ID for each element inserted into the set. It internally
1367 contains a map and a vector, and it assigns a unique ID for each value inserted
1368 into the set.</p>
1370 <p>UniqueVector is very expensive: its cost is the sum of the cost of
1371 maintaining both the map and vector, it has high complexity, high constant
1372 factors, and produces a lot of malloc traffic. It should be avoided.</p>
1374 </div>
1377 <!-- _______________________________________________________________________ -->
1378 <h4>
1380 </h4>
1382 <div>
1384 <p>
1385 The STL provides several other options, such as std::multiset and the various
1386 "hash_set" like containers (whether from C++ TR1 or from the SGI library). We
1387 never use hash_set and unordered_set because they are generally very expensive
1388 (each insertion requires a malloc) and very non-portable.
1389 </p>
1391 <p>std::multiset is useful if you're not interested in elimination of
1392 duplicates, but has all the drawbacks of std::set. A sorted vector (where you
1393 don't delete duplicate entries) or some other approach is almost always
1394 better.</p>
1396 </div>
1398 </div>
1400 <!-- ======================================================================= -->
1401 <h3>
1403 </h3>
1405 <div>
1406 Map-like containers are useful when you want to associate data to a key. As
1407 usual, there are a lot of different ways to do this. :)
1409 <!-- _______________________________________________________________________ -->
1410 <h4>
1412 </h4>
1414 <div>
1416 <p>
1417 If your usage pattern follows a strict insert-then-query approach, you can
1419 for set-like containers</a>. The only difference is that your query function
1420 (which uses std::lower_bound to get efficient log(n) lookup) should only compare
1421 the key, not both the key and value. This yields the same advantages as sorted
1422 vectors for sets.
1423 </p>
1424 </div>
1426 <!-- _______________________________________________________________________ -->
1427 <h4>
1429 </h4>
1431 <div>
1433 <p>
1434 Strings are commonly used as keys in maps, and they are difficult to support
1435 efficiently: they are variable length, inefficient to hash and compare when
1436 long, expensive to copy, etc. StringMap is a specialized container designed to
1437 cope with these issues. It supports mapping an arbitrary range of bytes to an
1438 arbitrary other object.</p>
1440 <p>The StringMap implementation uses a quadratically-probed hash table, where
1441 the buckets store a pointer to the heap allocated entries (and some other
1442 stuff). The entries in the map must be heap allocated because the strings are
1443 variable length. The string data (key) and the element object (value) are
1444 stored in the same allocation with the string data immediately after the element
1445 object. This container guarantees the "<tt>(char*)(&Value+1)</tt>" points
1446 to the key string for a value.</p>
1448 <p>The StringMap is very fast for several reasons: quadratic probing is very
1449 cache efficient for lookups, the hash value of strings in buckets is not
1450 recomputed when looking up an element, StringMap rarely has to touch the
1451 memory for unrelated objects when looking up a value (even when hash collisions
1452 happen), hash table growth does not recompute the hash values for strings
1453 already in the table, and each pair in the map is store in a single allocation
1454 (the string data is stored in the same allocation as the Value of a pair).</p>
1456 <p>StringMap also provides query methods that take byte ranges, so it only ever
1457 copies a string if a value is inserted into the table.</p>
1458 </div>
1460 <!-- _______________________________________________________________________ -->
1461 <h4>
1463 </h4>
1465 <div>
1466 <p>
1467 IndexedMap is a specialized container for mapping small dense integers (or
1468 values that can be mapped to small dense integers) to some other type. It is
1469 internally implemented as a vector with a mapping function that maps the keys to
1470 the dense integer range.
1471 </p>
1473 <p>
1474 This is useful for cases like virtual registers in the LLVM code generator: they
1475 have a dense mapping that is offset by a compile-time constant (the first
1476 virtual register ID).</p>
1478 </div>
1480 <!-- _______________________________________________________________________ -->
1481 <h4>
1483 </h4>
1485 <div>
1487 <p>
1488 DenseMap is a simple quadratically probed hash table. It excels at supporting
1489 small keys and values: it uses a single allocation to hold all of the pairs that
1490 are currently inserted in the map. DenseMap is a great way to map pointers to
1491 pointers, or map other small types to each other.
1492 </p>
1494 <p>
1495 There are several aspects of DenseMap that you should be aware of, however. The
1496 iterators in a densemap are invalidated whenever an insertion occurs, unlike
1497 map. Also, because DenseMap allocates space for a large number of key/value
1498 pairs (it starts with 64 by default), it will waste a lot of space if your keys
1499 or values are large. Finally, you must implement a partial specialization of
1500 DenseMapInfo for the key that you want, if it isn't already supported. This
1501 is required to tell DenseMap about two special marker values (which can never be
1502 inserted into the map) that it needs internally.</p>
1504 </div>
1506 <!-- _______________________________________________________________________ -->
1507 <h4>
1509 </h4>
1511 <div>
1513 <p>
1515 Value*s (or subclasses) to another type. When a Value is deleted or RAUW'ed,
1516 ValueMap will update itself so the new version of the key is mapped to the same
1517 value, just as if the key were a WeakVH. You can configure exactly how this
1518 happens, and what else happens on these two events, by passing
1521 </div>
1523 <!-- _______________________________________________________________________ -->
1524 <h4>
1526 </h4>
1528 <div>
1530 <p> IntervalMap is a compact map for small keys and values. It maps key
1531 intervals instead of single keys, and it will automatically coalesce adjacent
1532 intervals. When then map only contains a few intervals, they are stored in the
1533 map object itself to avoid allocations.</p>
1535 <p> The IntervalMap iterators are quite big, so they should not be passed around
1536 as STL iterators. The heavyweight iterators allow a smaller data structure.</p>
1538 </div>
1540 <!-- _______________________________________________________________________ -->
1541 <h4>
1543 </h4>
1545 <div>
1547 <p>
1549 a single allocation per pair inserted into the map, it offers log(n) lookup with
1550 an extremely large constant factor, imposes a space penalty of 3 pointers per
1551 pair in the map, etc.</p>
1553 <p>std::map is most useful when your keys or values are very large, if you need
1554 to iterate over the collection in sorted order, or if you need stable iterators
1555 into the map (i.e. they don't get invalidated if an insertion or deletion of
1556 another element takes place).</p>
1558 </div>
1560 <!-- _______________________________________________________________________ -->
1561 <h4>
1563 </h4>
1565 <div>
1567 <p>IntEqClasses provides a compact representation of equivalence classes of
1569 equivalence class. Classes can be joined by passing two class representatives to
1570 the join(a, b) method. Two integers are in the same class when findLeader()
1571 returns the same representative.</p>
1573 <p>Once all equivalence classes are formed, the map can be compressed so each
1575 is the total number of equivalence classes. The map must be uncompressed before
1576 it can be edited again.</p>
1578 </div>
1580 <!-- _______________________________________________________________________ -->
1581 <h4>
1583 </h4>
1585 <div>
1587 <p>
1588 The STL provides several other options, such as std::multimap and the various
1589 "hash_map" like containers (whether from C++ TR1 or from the SGI library). We
1590 never use hash_set and unordered_set because they are generally very expensive
1591 (each insertion requires a malloc) and very non-portable.</p>
1593 <p>std::multimap is useful if you want to map a key to multiple values, but has
1594 all the drawbacks of std::map. A sorted vector or some other approach is almost
1595 always better.</p>
1597 </div>
1599 </div>
1601 <!-- ======================================================================= -->
1602 <h3>
1604 </h3>
1606 <div>
1608 <p>
1609 TODO: const char* vs stringref vs smallstring vs std::string. Describe twine,
1610 xref to #string_apis.
1611 </p>
1613 </div>
1615 <!-- ======================================================================= -->
1616 <h3>
1618 </h3>
1620 <div>
1621 <p>Unlike the other containers, there are only two bit storage containers, and
1622 choosing when to use each is relatively straightforward.</p>
1624 <p>One additional option is
1626 implementation in many common compilers (e.g. commonly available versions of
1627 GCC) is extremely inefficient and 2) the C++ standards committee is likely to
1628 deprecate this container and/or change it significantly somehow. In any case,
1629 please don't use it.</p>
1631 <!-- _______________________________________________________________________ -->
1632 <h4>
1634 </h4>
1636 <div>
1637 <p> The BitVector container provides a dynamic size set of bits for manipulation.
1638 It supports individual bit setting/testing, as well as set operations. The set
1639 operations take time O(size of bitvector), but operations are performed one word
1640 at a time, instead of one bit at a time. This makes the BitVector very fast for
1641 set operations compared to other containers. Use the BitVector when you expect
1642 the number of set bits to be high (IE a dense set).
1643 </p>
1644 </div>
1646 <!-- _______________________________________________________________________ -->
1647 <h4>
1649 </h4>
1651 <div>
1652 <p> The SmallBitVector container provides the same interface as BitVector, but
1653 it is optimized for the case where only a small number of bits, less than
1654 25 or so, are needed. It also transparently supports larger bit counts, but
1655 slightly less efficiently than a plain BitVector, so SmallBitVector should
1656 only be used when larger counts are rare.
1657 </p>
1659 <p>
1660 At this time, SmallBitVector does not support set operations (and, or, xor),
1661 and its operator[] does not provide an assignable lvalue.
1662 </p>
1663 </div>
1665 <!-- _______________________________________________________________________ -->
1666 <h4>
1668 </h4>
1670 <div>
1671 <p> The SparseBitVector container is much like BitVector, with one major
1672 difference: Only the bits that are set, are stored. This makes the
1673 SparseBitVector much more space efficient than BitVector when the set is sparse,
1674 as well as making set operations O(number of set bits) instead of O(size of
1675 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
1676 (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).
1677 </p>
1678 </div>
1680 </div>
1682 </div>
1684 <!-- *********************************************************************** -->
1685 <h2>
1687 </h2>
1688 <!-- *********************************************************************** -->
1690 <div>
1692 <p>This section describes how to perform some very simple transformations of
1693 LLVM code. This is meant to give examples of common idioms used, showing the
1695 you should also read about the main classes that you will be working with. The
1697 and descriptions of the main classes that you should know about.</p>
1699 <!-- NOTE: this section should be heavy on example code -->
1700 <!-- ======================================================================= -->
1701 <h3>
1703 </h3>
1705 <div>
1707 <p>The LLVM compiler infrastructure have many different data structures that may
1708 be traversed. Following the example of the C++ standard template library, the
1709 techniques used to traverse these various data structures are all basically the
1712 function returns an iterator pointing to one past the last valid element of the
1714 between the two operations.</p>
1716 <p>Because the pattern for iteration is common across many different aspects of
1717 the program representation, the standard template library algorithms may be used
1718 on them, and it is easier to remember how to iterate. First we show a few common
1719 examples of the data structures that need to be traversed. Other data
1720 structures are traversed in very similar ways.</p>
1722 <!-- _______________________________________________________________________ -->
1723 <h4>
1727 </h4>
1729 <div>
1732 transform in some way; in particular, you'd like to manipulate its
1739 <pre>
1746 </pre>
1747 </div>
1749 <p>Note that i can be used as if it were a pointer for the purposes of
1751 because the indirection operator is overloaded for the iterator
1755 </div>
1757 <!-- _______________________________________________________________________ -->
1758 <h4>
1762 </h4>
1764 <div>
1767 easy to iterate over the individual instructions that make up
1772 <pre>
1778 </pre>
1779 </div>
1781 <p>However, this isn't really the best way to print out the contents of a
1783 anything you'll care about, you could have just invoked the print routine on the
1786 </div>
1788 <!-- _______________________________________________________________________ -->
1789 <h4>
1793 </h4>
1795 <div>
1802 small example that shows how to dump all instructions in a function to the standard error stream:<p>
1805 <pre>
1809 for (inst_iterator I = inst_begin(F), E = inst_end(F); I != E; ++I)
1811 </pre>
1812 </div>
1815 work list with its initial contents. For example, if you wanted to
1817 F, all you would need to do is something like:</p>
1820 <pre>
1824 for (inst_iterator I = inst_begin(F), E = inst_end(F); I != E; ++I)
1825 worklist.insert(&*I);
1826 </pre>
1827 </div>
1832 </div>
1834 <!-- _______________________________________________________________________ -->
1835 <h4>
1837 vice-versa)</a>
1838 </h4>
1840 <div>
1842 <p>Sometimes, it'll be useful to grab a reference (or pointer) to a class
1843 instance when all you've got at hand is an iterator. Well, extracting
1844 a reference or a pointer from an iterator is very straight-forward.
1849 <pre>
1852 const Instruction& inst = *j;
1853 </pre>
1854 </div>
1856 <p>However, the iterators you'll be working with in the LLVM framework are
1857 special: they will automatically convert to a ptr-to-instance type whenever they
1858 need to. Instead of dereferencing the iterator and then taking the address of
1859 the result, you can simply assign the iterator to the proper pointer type and
1860 you get the dereference and address-of operation as a result of the assignment
1861 (behind the scenes, this is a result of overloading casting mechanisms). Thus
1862 the last line of the last example,</p>
1865 <pre>
1866 Instruction *pinst = &*i;
1867 </pre>
1868 </div>
1873 <pre>
1874 Instruction *pinst = i;
1875 </pre>
1876 </div>
1878 <p>It's also possible to turn a class pointer into the corresponding iterator,
1879 and this is a constant time operation (very efficient). The following code
1880 snippet illustrates use of the conversion constructors provided by LLVM
1881 iterators. By using these, you can explicitly grab the iterator of something
1882 without actually obtaining it via iteration over some structure:</p>
1885 <pre>
1886 void printNextInstruction(Instruction* inst) {
1887 BasicBlock::iterator it(inst);
1890 }
1891 </pre>
1892 </div>
1894 <p>Unfortunately, these implicit conversions come at a cost; they prevent
1895 these iterators from conforming to standard iterator conventions, and thus
1896 from being usable with standard algorithms and containers. For example, they
1898 from compiling:</p>
1901 <pre>
1903 </pre>
1904 </div>
1906 <p>Because of this, these implicit conversions may be removed some day,
1909 </div>
1911 <!--_______________________________________________________________________-->
1912 <h4>
1914 example</a>
1915 </h4>
1917 <div>
1919 <p>Say that you're writing a FunctionPass and would like to count all the
1923 much more straight-forward manner, but this example will allow us to explore how
1925 is what we want to do:</p>
1928 <pre>
1929 initialize callCounter to zero
1930 for each Function f in the Module
1931 for each BasicBlock b in f
1932 for each Instruction i in b
1933 if (i is a CallInst and calls the given function)
1934 increment callCounter
1935 </pre>
1936 </div>
1938 <p>And the actual code is (remember, because we're writing a
1943 <pre>
1944 Function* targetFunc = ...;
1946 class OurFunctionPass : public FunctionPass {
1947 public:
1948 OurFunctionPass(): callCounter(0) { }
1950 virtual runOnFunction(Function& F) {
1951 for (Function::iterator b = F.begin(), be = F.end(); b != be; ++b) {
1958 if (callInst->getCalledFunction() == targetFunc)
1959 ++callCounter;
1960 }
1961 }
1962 }
1963 }
1965 private:
1966 unsigned callCounter;
1967 };
1968 </pre>
1969 </div>
1971 </div>
1973 <!--_______________________________________________________________________-->
1974 <h4>
1976 </h4>
1978 <div>
1980 <p>You may have noticed that the previous example was a bit oversimplified in
1981 that it did not deal with call sites generated by 'invoke' instructions. In
1982 this, and in other situations, you may find that you want to treat
1985 less closely-related things. For these cases, LLVM provides a handy wrapper
1986 class called <a
1993 reference and it should not be dynamically allocated or deallocated using
1995 assignable and constructable, with costs equivalents to that of a bare pointer.
1996 If you look at its definition, it has only a single pointer member.</p>
1998 </div>
2000 <!--_______________________________________________________________________-->
2001 <h4>
2003 </h4>
2005 <div>
2017 <pre>
2018 Function *F = ...;
2024 }
2025 </pre>
2026 </div>
2030 doing it only once in the loop body and reusing its result.</p>
2037 all of the values that a particular instruction uses (that is, the operands of
2041 <pre>
2042 Instruction *pi = ...;
2045 Value *v = *i;
2047 }
2048 </pre>
2049 </div>
2052 mutation free algorithms (such as analyses, etc.). For this purpose above
2059 </div>
2061 <!--_______________________________________________________________________-->
2062 <h4>
2064 successors of blocks</a>
2065 </h4>
2067 <div>
2069 <p>Iterating over the predecessors and successors of a block is quite easy
2071 this to iterate over all predecessors of BB:</p>
2074 <pre>
2075 #include "llvm/Support/CFG.h"
2076 BasicBlock *BB = ...;
2078 for (pred_iterator PI = pred_begin(BB), E = pred_end(BB); PI != E; ++PI) {
2079 BasicBlock *Pred = *PI;
2081 }
2082 </pre>
2083 </div>
2085 <p>Similarly, to iterate over successors use
2086 succ_iterator/succ_begin/succ_end.</p>
2088 </div>
2090 </div>
2092 <!-- ======================================================================= -->
2093 <h3>
2095 </h3>
2097 <div>
2099 <p>There are some primitive transformation operations present in the LLVM
2100 infrastructure that are worth knowing about. When performing
2101 transformations, it's fairly common to manipulate the contents of basic
2102 blocks. This section describes some of the common methods for doing so
2103 and gives example code.</p>
2105 <!--_______________________________________________________________________-->
2106 <h4>
2109 </h4>
2111 <div>
2116 constructor for the kind of instruction to instantiate and provide the necessary
2121 <pre>
2122 AllocaInst* ai = new AllocaInst(Type::Int32Ty);
2123 </pre>
2124 </div>
2128 subclass is likely to have varying default parameters which change the semantics
2129 of the instruction, so refer to the <a
2135 <p>It is very useful to name the values of instructions when you're able to, as
2136 this facilitates the debugging of your transformations. If you end up looking
2137 at generated LLVM machine code, you definitely want to have logical names
2138 associated with the results of instructions! By supplying a value for the
2140 associate a logical name with the result of the instruction's execution at
2141 run time. For example, say that I'm writing a transformation that dynamically
2142 allocates space for an integer on the stack, and that integer is going to be
2143 used as some kind of index by some other code. To accomplish this, I place an
2149 <pre>
2151 </pre>
2152 </div>
2155 execution value, which is a pointer to an integer on the run time stack.</p>
2162 <ul>
2163 <li>Insertion into an explicit instruction list
2170 <pre>
2171 BasicBlock *pb = ...;
2172 Instruction *pi = ...;
2173 Instruction *newInst = new Instruction(...);
2176 </pre>
2177 </div>
2181 classes provide constructors which take a pointer to a
2183 looked like: </p>
2186 <pre>
2187 BasicBlock *pb = ...;
2188 Instruction *newInst = new Instruction(...);
2191 </pre>
2192 </div>
2197 <pre>
2198 BasicBlock *pb = ...;
2199 Instruction *newInst = new Instruction(..., pb);
2200 </pre>
2201 </div>
2203 <p>which is much cleaner, especially if you are creating
2204 long instruction streams.</p></li>
2206 <li>Insertion into an implicit instruction list
2209 are implicitly associated with an existing instruction list: the instruction
2210 list of the enclosing basic block. Thus, we could have accomplished the same
2212 </p>
2215 <pre>
2216 Instruction *pi = ...;
2217 Instruction *newInst = new Instruction(...);
2220 </pre>
2221 </div>
2223 <p>In fact, this sequence of steps occurs so frequently that the
2225 constructors which take (as a default parameter) a pointer to an
2229 provided instruction, immediately before that instruction. Using an
2231 parameter, the above code becomes:</p>
2234 <pre>
2235 Instruction* pi = ...;
2236 Instruction* newInst = new Instruction(..., pi);
2237 </pre>
2238 </div>
2240 <p>which is much cleaner, especially if you're creating a lot of
2242 </ul>
2244 </div>
2246 <!--_______________________________________________________________________-->
2247 <h4>
2249 </h4>
2251 <div>
2253 <p>Deleting an instruction from an existing sequence of instructions that form a
2255 call the instruction's eraseFromParent() method. For example:</p>
2258 <pre>
2260 I->eraseFromParent();
2261 </pre>
2262 </div>
2264 <p>This unlinks the instruction from its containing basic block and deletes
2265 it. If you'd just like to unlink the instruction from its containing basic
2268 </div>
2270 <!--_______________________________________________________________________-->
2271 <h4>
2274 </h4>
2276 <div>
2280 <p>Including "<a href="/doxygen/BasicBlockUtils_8h-source.html">llvm/Transforms/Utils/BasicBlockUtils.h</a>"
2286 <ul>
2289 <p>This function replaces all uses of a given instruction with a value,
2290 and then removes the original instruction. The following example
2291 illustrates the replacement of the result of a particular
2293 pointer to an integer.</p>
2296 <pre>
2297 AllocaInst* instToReplace = ...;
2298 BasicBlock::iterator ii(instToReplace);
2301 Constant::getNullValue(PointerType::getUnqual(Type::Int32Ty)));
2302 </pre></div></li>
2306 <p>This function replaces a particular instruction with another
2307 instruction, inserting the new instruction into the basic block at the
2308 location where the old instruction was, and replacing any uses of the old
2309 instruction with the new instruction. The following example illustrates
2313 <pre>
2314 AllocaInst* instToReplace = ...;
2315 BasicBlock::iterator ii(instToReplace);
2319 </pre></div></li>
2320 </ul>
2328 information.</p>
2330 <!-- Value::replaceAllUsesWith User::replaceUsesOfWith Point out:
2331 include/llvm/Transforms/Utils/ especially BasicBlockUtils.h with:
2332 ReplaceInstWithValue, ReplaceInstWithInst -->
2334 </div>
2336 <!--_______________________________________________________________________-->
2337 <h4>
2339 </h4>
2341 <div>
2343 <p>Deleting a global variable from a module is just as easy as deleting an
2344 Instruction. First, you must have a pointer to the global variable that you wish
2345 to delete. You use this pointer to erase it from its parent, the module.
2346 For example:</p>
2349 <pre>
2352 GV->eraseFromParent();
2353 </pre>
2354 </div>
2356 </div>
2358 </div>
2360 <!-- ======================================================================= -->
2361 <h3>
2363 </h3>
2365 <div>
2367 <p>In generating IR, you may need some complex types. If you know these types
2370 has two forms depending on whether you're building types for cross-compilation
2373 out of types from
2375 namespace and pointers, functions, arrays, etc. built of
2377 whose size may depend on the host compiler. For example,</p>
2380 <pre>
2382 </pre>
2383 </div>
2388 <pre>
2390 params.push_back(PointerType::getUnqual(Type::Int32Ty));
2391 FunctionType *ft = FunctionType::get(Type::Int8Ty, params, false);
2392 </pre>
2393 </div>
2398 </div>
2400 </div>
2402 <!-- *********************************************************************** -->
2403 <h2>
2405 </h2>
2406 <!-- *********************************************************************** -->
2408 <div>
2409 <p>
2410 This section describes the interaction of the LLVM APIs with multithreading,
2411 both on the part of client applications, and in the JIT, in the hosted
2412 application.
2413 </p>
2415 <p>
2416 Note that LLVM's support for multithreading is still relatively young. Up
2417 through version 2.5, the execution of threaded hosted applications was
2418 supported, but not threaded client access to the APIs. While this use case is
2420 ensure proper operation in multithreaded mode.
2421 </p>
2423 <p>
2424 Note that, on Unix-like platforms, LLVM requires the presence of GCC's atomic
2425 intrinsics in order to support threaded operation. If you need a
2426 multhreading-capable LLVM on a platform without a suitably modern system
2427 compiler, consider compiling LLVM and LLVM-GCC in single-threaded mode, and
2428 using the resultant compiler to build a copy of LLVM with multithreading
2429 support.
2430 </p>
2432 <!-- ======================================================================= -->
2433 <h3>
2435 </h3>
2437 <div>
2439 <p>
2440 In order to properly protect its internal data structures while avoiding
2441 excessive locking overhead in the single-threaded case, the LLVM must intialize
2442 certain data structures necessary to provide guards around its internals. To do
2444 making any concurrent LLVM API calls. To subsequently tear down these
2447 mode.
2448 </p>
2450 <p>
2452 say that no other LLVM API calls may be executing at any time during the
2454 </tt>. It's is the client's responsibility to enforce this isolation.
2455 </p>
2457 <p>
2459 failure of the initialization. Failure typically indicates that your copy of
2460 LLVM was built without multithreading support, typically because GCC atomic
2461 intrinsics were not found in your system compiler. In this case, the LLVM API
2464 must be taken</a> to ensure that side exits and the like do not accidentally
2465 result in concurrent LLVM API calls.
2466 </p>
2467 </div>
2469 <!-- ======================================================================= -->
2470 <h3>
2472 </h3>
2474 <div>
2475 <p>
2477 to deallocate memory used for internal structures. This will also invoke
2481 </p>
2483 <p>
2484 Note that, if you use scope-based shutdown, you can use the
2486 destructor.
2487 </div>
2489 <!-- ======================================================================= -->
2490 <h3>
2492 </h3>
2494 <div>
2495 <p>
2497 initialization of static resources, such as the global type tables. Before the
2500 however, it uses double-checked locking to implement thread-safe lazy
2501 initialization.
2502 </p>
2504 <p>
2505 Note that, because no other threads are allowed to issue LLVM API calls before
2508 </p>
2510 <p>
2512 APIs provide access to the global lock used to implement the double-checked
2513 locking for lazy initialization. These should only be used internally to LLVM,
2514 and only if you know what you're doing!
2515 </p>
2516 </div>
2518 <!-- ======================================================================= -->
2519 <h3>
2521 </h3>
2523 <div>
2524 <p>
2526 to operate multiple, isolated instances of LLVM concurrently within the same
2527 address space. For instance, in a hypothetical compile-server, the compilation
2528 of an individual translation unit is conceptually independent from all the
2529 others, and it would be desirable to be able to compile incoming translation
2530 units concurrently on independent server threads. Fortunately,
2532 </p>
2534 <p>
2541 safe to compile on multiple threads simultaneously, as long as no two threads
2542 operate on entities within the same context.
2543 </p>
2545 <p>
2546 In practice, very few places in the API require the explicit specification of a
2549 other entities can determine what context they belong to by looking at their
2551 maintain this interface design.
2552 </p>
2554 <p>
2558 isolation is not a concern.
2559 </p>
2560 </div>
2562 <!-- ======================================================================= -->
2563 <h3>
2565 </h3>
2567 <div>
2568 <p>
2569 LLVM's "eager" JIT compiler is safe to use in threaded programs. Multiple
2572 run code output by the JIT concurrently. The user must still ensure that only
2574 might be modifying it. One way to do that is to always hold the JIT lock while
2578 </p>
2580 <p>When the JIT is configured to compile lazily (using
2583 updating call sites after a function is lazily-jitted. It's still possible to
2584 use the lazy JIT in a threaded program if you ensure that only one thread at a
2585 time can call any particular lazy stub and that the JIT lock guards any IR
2586 access, but we suggest using only the eager JIT in threaded programs.
2587 </p>
2588 </div>
2590 </div>
2592 <!-- *********************************************************************** -->
2593 <h2>
2595 </h2>
2596 <!-- *********************************************************************** -->
2598 <div>
2599 <p>
2600 This section describes some of the advanced or obscure API's that most clients
2601 do not need to be aware of. These API's tend manage the inner workings of the
2602 LLVM system, and only need to be accessed in unusual circumstances.
2603 </p>
2605 <!-- ======================================================================= -->
2606 <h3>
2608 </h3>
2610 <div>
2612 <p>
2613 The LLVM type system has a very simple goal: allow clients to compare types for
2614 structural equality with a simple pointer comparison (aka a shallow compare).
2615 This goal makes clients much simpler and faster, and is used throughout the LLVM
2616 system.
2617 </p>
2619 <p>
2620 Unfortunately achieving this goal is not a simple matter. In particular,
2621 recursive types and late resolution of opaque types makes the situation very
2622 difficult to handle. Fortunately, for the most part, our implementation makes
2623 most clients able to be completely unaware of the nasty internal details. The
2624 primary case where clients are exposed to the inner workings of it are when
2625 building a recursive type. In addition to this case, the LLVM bitcode reader,
2626 assembly parser, and linker also have to be aware of the inner workings of this
2627 system.
2628 </p>
2630 <p>
2631 For our purposes below, we need three concepts. First, an "Opaque Type" is
2634 opaque type as part of its type graph (for example "<tt>{ opaque, i32 }</tt>").
2635 Third, a concrete type is a type that is not an abstract type (e.g. "<tt>{ i32,
2636 float }</tt>").
2637 </p>
2639 <!-- ______________________________________________________________________ -->
2640 <h4>
2642 </h4>
2644 <div>
2646 <p>
2647 Because the most common question is "how do I build a recursive type with LLVM",
2648 we answer it now and explain it as we go. Here we include enough to cause this
2649 to be emitted to an output .ll file:
2650 </p>
2653 <pre>
2654 %mylist = type { %mylist*, i32 }
2655 </pre>
2656 </div>
2658 <p>
2659 To build this, use the following LLVM APIs:
2660 </p>
2663 <pre>
2667 Elts.push_back(PointerType::getUnqual(StructTy));
2668 Elts.push_back(Type::Int32Ty);
2669 StructType *NewSTy = StructType::get(Elts);
2673 cast<OpaqueType>(StructTy.get())-><a href="#refineAbstractTypeTo">refineAbstractTypeTo</a>(NewSTy);
2675 // <i>NewSTy is potentially invalidated, but StructTy (a <a href="#PATypeHolder">PATypeHolder</a>) is</i>
2681 </pre>
2682 </div>
2684 <p>
2685 This code shows the basic approach used to build recursive types: build a
2686 non-recursive type using 'opaque', then use type unification to close the cycle.
2687 The type unification step is performed by the <tt><a
2689 described next. After that, we describe the <a
2691 </p>
2693 </div>
2695 <!-- ______________________________________________________________________ -->
2696 <h4>
2698 </h4>
2700 <div>
2701 <p>
2703 While this method is actually a member of the DerivedType class, it is most
2704 often used on OpaqueType instances. Type unification is actually a recursive
2705 process. After unification, types can become structurally isomorphic to
2706 existing types, and all duplicates are deleted (to preserve pointer equality).
2707 </p>
2709 <p>
2710 In the example above, the OpaqueType object is definitely deleted.
2711 Additionally, if there is an "{ \2*, i32}" type already created in the system,
2713 a type is deleted, any "Type*" pointers in the program are invalidated. As
2716 types can never move or be deleted). To deal with this, the <a
2718 reference to a possibly refined type, and the <a
2720 complex datastructures.
2721 </p>
2723 </div>
2725 <!-- ______________________________________________________________________ -->
2726 <h4>
2728 </h4>
2730 <div>
2731 <p>
2732 PATypeHolder is a form of a "smart pointer" for Type objects. When VMCore
2733 happily goes about nuking types that become isomorphic to existing types, it
2734 automatically updates all PATypeHolder objects to point to the new type. In the
2735 example above, this allows the code to maintain a pointer to the resultant
2736 resolved recursive type, even though the Type*'s are potentially invalidated.
2737 </p>
2739 <p>
2740 PATypeHolder is an extremely light-weight object that uses a lazy union-find
2741 implementation to update pointers. For example the pointer from a Value to its
2742 Type is maintained by PATypeHolder objects.
2743 </p>
2745 </div>
2747 <!-- ______________________________________________________________________ -->
2748 <h4>
2750 </h4>
2752 <div>
2754 <p>
2755 Some data structures need more to perform more complex updates when types get
2756 resolved. To support this, a class can derive from the AbstractTypeUser class.
2757 This class
2758 allows it to get callbacks when certain types are resolved. To register to get
2759 callbacks for a particular type, the DerivedType::{add/remove}AbstractTypeUser
2760 methods can be called on a type. Note that these methods only work for <i>
2761 abstract</i> types. Concrete types (those that do not include any opaque
2762 objects) can never be refined.
2763 </p>
2764 </div>
2766 </div>
2768 <!-- ======================================================================= -->
2769 <h3>
2772 </h3>
2774 <div>
2782 names for types.</p>
2785 by most clients. It should only be used when iteration over the symbol table
2786 names themselves are required, which is very special purpose. Note that not
2787 all LLVM
2789 an empty name) do not exist in the symbol table.
2790 </p>
2792 <p>These symbol tables support iteration over the values/types in the symbol
2797 appropriate symbol table. For types, use the Module::addTypeName method to
2798 insert entries into the symbol table.</p>
2800 </div>
2804 <!-- ======================================================================= -->
2805 <h3>
2807 </h3>
2809 <div>
2815 addition and removal.</p>
2817 <!-- ______________________________________________________________________ -->
2818 <h4>
2821 </a>
2822 </h4>
2824 <div>
2825 <p>
2827 or refer to them out-of-line by means of a pointer. A mixed variant
2830 </p>
2832 <p>
2834 <ul>
2835 <li><p>Layout a)
2837 object and there are a fixed number of them.</p>
2839 <li><p>Layout b)
2842 number of them.</p>
2843 </ul>
2844 <p>
2845 As of v2.4 each layout still possesses a direct pointer to the
2847 we stick to this redundancy for the sake of simplicity.
2849 has. (Theoretically this information can also be calculated
2850 given the scheme presented below.)</p>
2851 <p>
2853 enforce the following memory layouts:</p>
2855 <ul>
2856 <li><p>Layout a) is modelled by prepending the <tt>User</tt> object by the <tt>Use[]</tt> array.</p>
2858 <pre>
2859 ...---.---.---.---.-------...
2860 | P | P | P | P | User
2861 '''---'---'---'---'-------'''
2862 </pre>
2865 <pre>
2866 .-------...
2867 | User
2868 '-------'''
2869 |
2870 v
2871 .---.---.---.---...
2872 | P | P | P | P |
2873 '---'---'---'---'''
2874 </pre>
2875 </ul>
2879 </div>
2881 <!-- ______________________________________________________________________ -->
2882 <h4>
2884 </h4>
2886 <div>
2887 <p>
2890 recover it. This is accomplished by the following scheme:</p>
2892 A bit-encoding in the 2 LSBits (least significant bits) of the <tt>Use::Prev</tt> allows to find the
2894 <ul>
2899 </ul>
2900 <p>
2903 we have to walk to the next stop picking up digits
2904 and calculating the offset:</p>
2905 <pre>
2906 .---.---.---.---.---.---.---.---.---.---.---.---.---.---.---.---.----------------
2908 '---'---'---'---'---'---'---'---'---'---'---'---'---'---'---'---'----------------
2910 | | | | |__>
2911 | | | |__________>
2912 | | |______________________>
2913 | |______________________________________>
2914 |__________________________________________________________>
2915 </pre>
2916 <p>
2917 Only the significant number of bits need to be stored between the
2921 </div>
2923 <!-- ______________________________________________________________________ -->
2924 <h4>
2926 </h4>
2928 <div>
2929 <p>
2930 The following literate Haskell fragment demonstrates the concept:</p>
2933 <pre>
2934 > import Test.QuickCheck
2935 >
2940 >
2946 >
2947 > takeLast n ss = reverse $ take n $ reverse ss
2948 >
2950 >
2951 </pre>
2952 </div>
2953 <p>
2955 <p>
2956 The reverse algorithm computes the length of the string just by examining
2957 a certain prefix:</p>
2960 <pre>
2965 >
2968 > decode walk acc _ = walk + acc
2969 >
2970 </pre>
2971 </div>
2972 <p>
2974 <p>
2978 <pre>
2980 > testcaseLength = length testcase
2981 >
2983 > where arr = takeLast n testcase
2984 >
2985 </pre>
2986 </div>
2987 <p>
2990 <pre>
2991 *Main> quickCheck identityProp
2992 OK, passed 100 tests.
2993 </pre>
2994 <p>
2995 Let's be a bit more exhaustive:</p>
2998 <pre>
2999 >
3001 >
3002 </pre>
3003 </div>
3004 <p>
3007 <pre>
3008 *Main> deepCheck identityProp
3009 OK, passed 500 tests.
3010 </pre>
3012 </div>
3014 <!-- ______________________________________________________________________ -->
3015 <h4>
3017 </h4>
3019 <div>
3021 <p>
3025 tag bits.</p>
3026 <p>
3030 the LSBit set. (Portability is relying on the fact that all known compilers place the
3033 </div>
3035 </div>
3037 </div>
3039 <!-- *********************************************************************** -->
3040 <h2>
3042 </h2>
3043 <!-- *********************************************************************** -->
3045 <div>
3049 <p>The Core LLVM classes are the primary means of representing the program
3050 being inspected or transformed. The core LLVM classes are defined in
3054 <!-- ======================================================================= -->
3055 <h3>
3057 </h3>
3059 <div>
3065 subclasses. They are hidden because they offer no useful functionality beyond
3069 named, but this is not a requirement. There exists exactly
3070 one instance of a given shape at any one time. This allows type equality to
3071 be performed with address equality of the Type Instance. That is, given two
3073 </p>
3075 <!-- _______________________________________________________________________ -->
3076 <h4>
3078 </h4>
3080 <div>
3082 <ul>
3086 floating point types.</li>
3089 an OpaqueType anywhere in its definition).</li>
3092 that don't have a size are abstract types, labels and void.</li>
3094 </ul>
3095 </div>
3097 <!-- _______________________________________________________________________ -->
3098 <h4>
3100 </h4>
3101 <div>
3102 <dl>
3104 <dd>Subclass of DerivedType that represents integer types of any bit width.
3107 <ul>
3109 type of a specific bit width.</li>
3111 type.</li>
3112 </ul>
3113 </dd>
3115 <dd>This is subclassed by ArrayType and PointerType
3116 <ul>
3118 of the elements in the sequential type. </li>
3119 </ul>
3120 </dd>
3122 <dd>This is a subclass of SequentialType and defines the interface for array
3123 types.
3124 <ul>
3126 elements in the array. </li>
3127 </ul>
3128 </dd>
3132 <dd>Subclass of SequentialType for vector types. A
3133 vector type is similar to an ArrayType but is distinguished because it is
3134 a first class type whereas ArrayType is not. Vector types are used for
3135 vector operations and are usually small vectors of of an integer or floating
3136 point type.</dd>
3140 <dd>Subclass of DerivedTypes for function types.
3141 <ul>
3143 function</li>
3145 return type of the function.</li>
3147 the type of the ith parameter.</li>
3149 number of formal parameters.</li>
3150 </ul>
3151 </dd>
3153 <dd>Sublcass of DerivedType for abstract types. This class
3154 defines no content and is used as a placeholder for some other type. Note
3155 that OpaqueType is used (temporarily) during type resolution for forward
3156 references of types. Once the referenced type is resolved, the OpaqueType
3157 is replaced with the actual type. OpaqueType can also be used for data
3158 abstraction. At link time opaque types can be resolved to actual types
3159 of the same name.</dd>
3160 </dl>
3161 </div>
3163 </div>
3165 <!-- ======================================================================= -->
3166 <h3>
3168 </h3>
3170 <div>
3177 programs. An LLVM module is effectively either a translation unit of the
3178 original program or a combination of several translation units merged by the
3183 helpful member functions that try to make common operations easy.</p>
3185 <!-- _______________________________________________________________________ -->
3186 <h4>
3188 </h4>
3190 <div>
3192 <ul>
3194 </ul>
3197 provide a name for it (probably based on the name of the translation unit).</p>
3199 <ul>
3206 <p>These are forwarding methods that make it easy to access the contents of
3208 list.</p></li>
3213 necessary to use when you need to update the list or perform a complex
3214 action that doesn't have a forwarding method.</p>
3217 </ul>
3219 <hr>
3221 <ul>
3229 <p> These are forwarding methods that make it easy to access the contents of
3237 use when you need to update the list or perform a complex action that
3238 doesn't have a forwarding method.</p>
3241 </ul>
3243 <hr>
3245 <ul>
3252 </ul>
3254 <hr>
3256 <ul>
3269 external declaration for the function and return it.</p></li>
3276 string.</p></li>
3283 name, true is returned and the <a
3285 </ul>
3287 </div>
3289 </div>
3291 <!-- ======================================================================= -->
3292 <h3>
3294 </h3>
3296 <div>
3299 <br>
3303 base. It represents a typed value that may be used (among other things) as an
3311 for a program. For example, an incoming argument to a function (represented
3313 every instruction in the function that references the argument. To keep track
3319 methods, shown below.</p>
3323 method. In addition, all LLVM values can be named. The "name" of the
3327 <pre>
3329 </pre>
3330 </div>
3333 that the name of any value may be missing (an empty string), so names should
3335 debugging printouts), they should not be used to keep track of values or map
3339 <p>One important aspect of LLVM is that there is no distinction between an SSA
3340 variable and the operation that produces it. Because of this, any reference to
3341 the value produced by an instruction (or the value available as an incoming
3342 argument, for example) is represented as a direct pointer to the instance of
3343 the class that
3344 represents this value. Although this may take some getting used to, it
3345 simplifies the representation and makes it easier to manipulate.</p>
3347 <!-- _______________________________________________________________________ -->
3348 <h4>
3350 </h4>
3352 <div>
3354 <ul>
3356 use-list<br>
3358 the use-list<br>
3360 value.<br>
3363 the use-list.<br>
3365 use-list.<br>
3367 element in the list.
3368 <p> These methods are the interface to access the def-use
3369 information in LLVM. As with all other iterators in LLVM, the naming
3371 </li>
3374 </li>
3380 </li>
3385 "<tt>V</tt>" instead. For example, if you detect that an instruction always
3386 produces a constant value (for example through constant folding), you can
3387 replace all uses of the instruction with the constant like this:</p>
3390 <pre>
3391 Inst->replaceAllUsesWith(ConstVal);
3392 </pre>
3393 </div>
3395 </ul>
3397 </div>
3399 </div>
3401 <!-- ======================================================================= -->
3402 <h3>
3404 </h3>
3406 <div>
3408 <p>
3421 Single Assignment (SSA) form, there can only be one definition referred to,
3422 allowing this direct connection. This connection provides the use-def
3423 information in LLVM.</p>
3425 <!-- _______________________________________________________________________ -->
3426 <h4>
3428 </h4>
3430 <div>
3433 an index access interface and through an iterator based interface.</p>
3435 <ul>
3439 convenient form for direct access.</p></li>
3442 list<br>
3444 the operand list.<br>
3446 operand list.
3447 <p> Together, these methods make up the iterator based interface to
3449 </ul>
3451 </div>
3453 </div>
3455 <!-- ======================================================================= -->
3456 <h3>
3458 </h3>
3460 <div>
3469 instructions. It provides only a few methods, but is a very commonly used
3471 opcode (instruction type) and the parent <a
3473 into. To represent a specific type of instruction, one of many subclasses of
3482 file contains some meta-data about the various different types of instructions
3483 in LLVM. It describes the enum values that are used as opcodes (for example
3488 this file confuses doxygen, so these enum values don't show up correctly in the
3491 <!-- _______________________________________________________________________ -->
3492 <h4>
3495 </a>
3496 </h4>
3497 <div>
3498 <ul>
3500 <p>This subclasses represents all two operand instructions whose operands
3501 must be the same type, except for the comparison instructions.</p></li>
3504 common operations on cast instructions.</p>
3506 <p>This subclass respresents the two comparison instructions,
3510 <p>This subclass is the parent of all terminator instructions (those which
3511 can terminate a block).</p>
3512 </ul>
3513 </div>
3515 <!-- _______________________________________________________________________ -->
3516 <h4>
3519 </a>
3520 </h4>
3522 <div>
3524 <ul>
3529 <p>Returns true if the instruction writes to memory, i.e. it is a
3534 <p>Returns another instance of the specified instruction, identical
3535 in all ways to the original except that the instruction has no parent
3537 and it has no name</p></li>
3538 </ul>
3540 </div>
3542 </div>
3544 <!-- ======================================================================= -->
3545 <h3>
3547 </h3>
3549 <div>
3551 <p>Constant represents a base class for different types of constants. It
3552 is subclassed by ConstantInt, ConstantArray, etc. for representing
3554 a subclass, which represents the address of a global variable or function.
3555 </p>
3557 <!-- _______________________________________________________________________ -->
3559 <div>
3560 <ul>
3561 <li>ConstantInt : This subclass of Constant represents an integer constant of
3562 any width.
3563 <ul>
3565 value of this constant, an APInt value.</li>
3567 value to an int64_t via sign extension. If the value (not the bit width)
3568 of the APInt is too large to fit in an int64_t, an assertion will result.
3569 For this reason, use of this method is discouraged.</li>
3571 value to a uint64_t via zero extension. IF the value (not the bit width)
3572 of the APInt is too large to fit in a uint64_t, an assertion will result.
3573 For this reason, use of this method is discouraged.</li>
3576 The type is implied as the IntegerType that corresponds to the bit width
3579 Returns the ConstantInt object that represents the value provided by
3581 </ul>
3582 </li>
3583 <li>ConstantFP : This class represents a floating point constant.
3584 <ul>
3586 this constant. </li>
3587 </ul>
3588 </li>
3589 <li>ConstantArray : This represents a constant array.
3590 <ul>
3592 a vector of component constants that makeup this array. </li>
3593 </ul>
3594 </li>
3595 <li>ConstantStruct : This represents a constant struct.
3596 <ul>
3598 a vector of component constants that makeup this array. </li>
3599 </ul>
3600 </li>
3601 <li>GlobalValue : This represents either a global variable or a function. In
3602 either case, the value is a constant fixed address (after linking).
3603 </li>
3604 </ul>
3605 </div>
3607 </div>
3609 <!-- ======================================================================= -->
3610 <h3>
3612 </h3>
3614 <div>
3619 Class</a><br>
3626 Because they are visible at global scope, they are also subject to linking with
3627 other globals defined in different translation units. To control the linking
3634 unit, and does not participate in linking. If it has external linkage, it is
3635 visible to external code, and does participate in linking. In addition to
3641 global is always a pointer to its contents. It is important to remember this
3646 the address of the first element of this array and the value of the
3654 <!-- _______________________________________________________________________ -->
3655 <h4>
3658 </a>
3659 </h4>
3661 <div>
3663 <ul>
3668 <p> </p>
3669 </li>
3672 GlobalValue is currently embedded into.</p></li>
3673 </ul>
3675 </div>
3677 </div>
3679 <!-- ======================================================================= -->
3680 <h3>
3682 </h3>
3684 <div>
3695 actually one of the more complex classes in the LLVM hierarchy because it must
3703 ordering of the blocks in the function, which indicate how the code will be
3704 laid out by the backend. Additionally, the first <a
3707 block. There are no implicit exit nodes, and in fact there may be multiple exit
3711 function hasn't been linked in yet.</p>
3721 LLVM feature that is only used when you have to look up a value by name. Aside
3723 internally to make sure that there are not conflicts between the names of <a
3730 is its address (after linking) which is guaranteed to be constant.</p>
3732 <!-- _______________________________________________________________________ -->
3733 <h4>
3736 </a>
3737 </h4>
3739 <div>
3741 <ul>
3746 the the program. The constructor must specify the type of the function to
3747 create and what type of linkage the function should have. The <a
3749 specifies the formal arguments and return value for the function. The same
3752 in which the function is defined. If this argument is provided, the function
3753 will automatically be inserted into that module's list of
3754 functions.</p></li>
3759 function is "external", it does not have a body, and thus must be resolved
3760 by linking with a function defined in a different translation unit.</p></li>
3768 <p>These are forwarding methods that make it easy to access the contents of
3770 list.</p></li>
3775 is necessary to use when you need to update the list or perform a complex
3776 action that doesn't have a forwarding method.</p></li>
3779 iterator<br>
3785 <p>These are forwarding methods that make it easy to access the contents of
3787 list.</p></li>
3792 necessary to use when you need to update the list or perform a complex
3793 action that doesn't have a forwarding method.</p></li>
3798 function. Because the entry block for the function is always the first
3807 function.</p></li>
3813 </ul>
3815 </div>
3817 </div>
3819 <!-- ======================================================================= -->
3820 <h3>
3822 </h3>
3824 <div>
3828 <br>
3830 Class</a><br>
3836 <p>Global variables are represented with the (surprise surprise)
3839 always referenced by their address (global values must live in memory, so their
3840 "name" refers to their constant address). See
3842 variables may have an initial value (which must be a
3844 they may be marked as "constant" themselves (indicating that their contents
3845 never change at runtime).</p>
3847 <!-- _______________________________________________________________________ -->
3848 <h4>
3851 </a>
3852 </h4>
3854 <div>
3856 <ul>
3861 <p>Create a new global variable of the specified type. If
3863 unchanging for the program. The Linkage parameter specifies the type of
3864 linkage (internal, external, weak, linkonce, appending) for the variable.
3865 If the linkage is InternalLinkage, WeakAnyLinkage, WeakODRLinkage,
3866 LinkOnceAnyLinkage or LinkOnceODRLinkage, then the resultant
3867 global variable will have internal linkage. AppendingLinkage concatenates
3868 together all instances (in different translation units) of the variable
3869 into a single variable but is only applicable to arrays. See
3871 further details on linkage types. Optionally an initializer, a name, and the
3872 module to put the variable into may be specified for the global variable as
3873 well.</p></li>
3877 <p>Returns true if this is a global variable that is known not to
3878 be modified at runtime.</p></li>
3887 to call this method if there is no initializer.</p></li>
3888 </ul>
3890 </div>
3892 </div>
3894 <!-- ======================================================================= -->
3895 <h3>
3897 </h3>
3899 <div>
3904 Class</a><br>
3907 <p>This class represents a single entry single exit section of the code,
3908 commonly known as a basic block by the compiler community. The
3911 Matching the language definition, the last element of this list of instructions
3912 is always a terminator instruction (a subclass of the <a
3915 <p>In addition to tracking the list of instructions that make up the block, the
3924 <!-- _______________________________________________________________________ -->
3925 <h4>
3928 </a>
3929 </h4>
3931 <div>
3932 <ul>
3938 insertion into a function. The constructor optionally takes a name for the new
3941 automatically inserted at the end of the specified <a
3949 STL-style functions for accessing the instruction list.
3951 <p>These methods and typedefs are forwarding functions that have the same
3952 semantics as the standard library methods of the same names. These methods
3953 expose the underlying instruction list of a basic block in a way that is easy to
3954 manipulate. To get the full complement of container operations (including
3956 method.</p></li>
3960 <p>This method is used to get access to the underlying container that actually
3961 holds the Instructions. This method must be used when there isn't a forwarding
3963 to perform. Because there are no forwarding functions for "updating"
3964 operations, you need to use this if you want to update the contents of a
3970 embedded into, or a null pointer if it is homeless.</p></li>
3974 <p> Returns a pointer to the terminator instruction that appears at the end of
3976 instruction in the block is not a terminator, then a null pointer is
3977 returned.</p></li>
3979 </ul>
3981 </div>
3983 </div>
3985 <!-- ======================================================================= -->
3986 <h3>
3988 </h3>
3990 <div>
3992 <p>This subclass of Value defines the interface for incoming formal
3993 arguments to a function. A Function maintains a list of its formal
3994 arguments. An argument has a pointer to the parent Function.</p>
3996 </div>
3998 </div>
4000 <!-- *********************************************************************** -->
4001 <hr>
4002 <address>
4011 Last modified: $Date$
4012 </address>
4014 </body>
4015 </html>