1 <!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.01//EN"
2 "http://www.w3.org/TR/html4/strict.dtd">
5 <meta http-equiv="Content-type" content="text/html;charset=UTF-8">
6 <title>LLVM Programmer's Manual</title>
7 <link rel="stylesheet" href="llvm.css" type="text/css">
11 <div class="doc_title">
12 LLVM Programmer's Manual
16 <li><a href="#introduction">Introduction</a></li>
17 <li><a href="#general">General Information</a>
19 <li><a href="#stl">The C++ Standard Template Library</a></li>
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>
28 <li><a href="#apis">Important and useful LLVM APIs</a>
30 <li><a href="#isa">The <tt>isa<></tt>, <tt>cast<></tt>
31 and <tt>dyn_cast<></tt> templates</a> </li>
32 <li><a href="#string_apis">Passing strings (the <tt>StringRef</tt>
33 and <tt>Twine</tt> classes)</a>
35 <li><a href="#StringRef">The <tt>StringRef</tt> class</a> </li>
36 <li><a href="#Twine">The <tt>Twine</tt> class</a> </li>
39 <li><a href="#DEBUG">The <tt>DEBUG()</tt> macro and <tt>-debug</tt>
42 <li><a href="#DEBUG_TYPE">Fine grained debug info with <tt>DEBUG_TYPE</tt>
43 and the <tt>-debug-only</tt> option</a> </li>
46 <li><a href="#Statistic">The <tt>Statistic</tt> class & <tt>-stats</tt>
49 <li>The <tt>InstVisitor</tt> template
50 <li>The general graph API
52 <li><a href="#ViewGraph">Viewing graphs while debugging code</a></li>
55 <li><a href="#datastructure">Picking the Right Data Structure for a Task</a>
57 <li><a href="#ds_sequential">Sequential Containers (std::vector, std::list, etc)</a>
59 <li><a href="#dss_fixedarrays">Fixed Size Arrays</a></li>
60 <li><a href="#dss_heaparrays">Heap Allocated Arrays</a></li>
61 <li><a href="#dss_smallvector">"llvm/ADT/SmallVector.h"</a></li>
62 <li><a href="#dss_vector"><vector></a></li>
63 <li><a href="#dss_deque"><deque></a></li>
64 <li><a href="#dss_list"><list></a></li>
65 <li><a href="#dss_ilist">llvm/ADT/ilist.h</a></li>
66 <li><a href="#dss_other">Other Sequential Container Options</a></li>
68 <li><a href="#ds_set">Set-Like Containers (std::set, SmallSet, SetVector, etc)</a>
70 <li><a href="#dss_sortedvectorset">A sorted 'vector'</a></li>
71 <li><a href="#dss_smallset">"llvm/ADT/SmallSet.h"</a></li>
72 <li><a href="#dss_smallptrset">"llvm/ADT/SmallPtrSet.h"</a></li>
73 <li><a href="#dss_denseset">"llvm/ADT/DenseSet.h"</a></li>
74 <li><a href="#dss_FoldingSet">"llvm/ADT/FoldingSet.h"</a></li>
75 <li><a href="#dss_set"><set></a></li>
76 <li><a href="#dss_setvector">"llvm/ADT/SetVector.h"</a></li>
77 <li><a href="#dss_uniquevector">"llvm/ADT/UniqueVector.h"</a></li>
78 <li><a href="#dss_otherset">Other Set-Like ContainerOptions</a></li>
80 <li><a href="#ds_map">Map-Like Containers (std::map, DenseMap, etc)</a>
82 <li><a href="#dss_sortedvectormap">A sorted 'vector'</a></li>
83 <li><a href="#dss_stringmap">"llvm/ADT/StringMap.h"</a></li>
84 <li><a href="#dss_indexedmap">"llvm/ADT/IndexedMap.h"</a></li>
85 <li><a href="#dss_densemap">"llvm/ADT/DenseMap.h"</a></li>
86 <li><a href="#dss_map"><map></a></li>
87 <li><a href="#dss_othermap">Other Map-Like Container Options</a></li>
89 <li><a href="#ds_string">String-like containers</a>
93 <li><a href="#ds_bit">BitVector-like containers</a>
95 <li><a href="#dss_bitvector">A dense bitvector</a></li>
96 <li><a href="#dss_sparsebitvector">A sparse bitvector</a></li>
100 <li><a href="#common">Helpful Hints for Common Operations</a>
102 <li><a href="#inspection">Basic Inspection and Traversal Routines</a>
104 <li><a href="#iterate_function">Iterating over the <tt>BasicBlock</tt>s
105 in a <tt>Function</tt></a> </li>
106 <li><a href="#iterate_basicblock">Iterating over the <tt>Instruction</tt>s
107 in a <tt>BasicBlock</tt></a> </li>
108 <li><a href="#iterate_institer">Iterating over the <tt>Instruction</tt>s
109 in a <tt>Function</tt></a> </li>
110 <li><a href="#iterate_convert">Turning an iterator into a
111 class pointer</a> </li>
112 <li><a href="#iterate_complex">Finding call sites: a more
113 complex example</a> </li>
114 <li><a href="#calls_and_invokes">Treating calls and invokes
115 the same way</a> </li>
116 <li><a href="#iterate_chains">Iterating over def-use &
117 use-def chains</a> </li>
118 <li><a href="#iterate_preds">Iterating over predecessors &
119 successors of blocks</a></li>
122 <li><a href="#simplechanges">Making simple changes</a>
124 <li><a href="#schanges_creating">Creating and inserting new
125 <tt>Instruction</tt>s</a> </li>
126 <li><a href="#schanges_deleting">Deleting <tt>Instruction</tt>s</a> </li>
127 <li><a href="#schanges_replacing">Replacing an <tt>Instruction</tt>
128 with another <tt>Value</tt></a> </li>
129 <li><a href="#schanges_deletingGV">Deleting <tt>GlobalVariable</tt>s</a> </li>
132 <li><a href="#create_types">How to Create Types</a></li>
134 <li>Working with the Control Flow Graph
136 <li>Accessing predecessors and successors of a <tt>BasicBlock</tt>
144 <li><a href="#threading">Threads and LLVM</a>
146 <li><a href="#startmultithreaded">Entering and Exiting Multithreaded Mode
148 <li><a href="#shutdown">Ending execution with <tt>llvm_shutdown()</tt></a></li>
149 <li><a href="#managedstatic">Lazy initialization with <tt>ManagedStatic</tt></a></li>
153 <li><a href="#advanced">Advanced Topics</a>
155 <li><a href="#TypeResolve">LLVM Type Resolution</a>
157 <li><a href="#BuildRecType">Basic Recursive Type Construction</a></li>
158 <li><a href="#refineAbstractTypeTo">The <tt>refineAbstractTypeTo</tt> method</a></li>
159 <li><a href="#PATypeHolder">The PATypeHolder Class</a></li>
160 <li><a href="#AbstractTypeUser">The AbstractTypeUser Class</a></li>
163 <li><a href="#SymbolTable">The <tt>ValueSymbolTable</tt> and <tt>TypeSymbolTable</tt> classes</a></li>
164 <li><a href="#UserLayout">The <tt>User</tt> and owned <tt>Use</tt> classes' memory layout</a></li>
167 <li><a href="#coreclasses">The Core LLVM Class Hierarchy Reference</a>
169 <li><a href="#Type">The <tt>Type</tt> class</a> </li>
170 <li><a href="#Module">The <tt>Module</tt> class</a></li>
171 <li><a href="#Value">The <tt>Value</tt> class</a>
173 <li><a href="#User">The <tt>User</tt> class</a>
175 <li><a href="#Instruction">The <tt>Instruction</tt> class</a></li>
176 <li><a href="#Constant">The <tt>Constant</tt> class</a>
178 <li><a href="#GlobalValue">The <tt>GlobalValue</tt> class</a>
180 <li><a href="#Function">The <tt>Function</tt> class</a></li>
181 <li><a href="#GlobalVariable">The <tt>GlobalVariable</tt> class</a></li>
188 <li><a href="#BasicBlock">The <tt>BasicBlock</tt> class</a></li>
189 <li><a href="#Argument">The <tt>Argument</tt> class</a></li>
196 <div class="doc_author">
197 <p>Written by <a href="mailto:sabre@nondot.org">Chris Lattner</a>,
198 <a href="mailto:dhurjati@cs.uiuc.edu">Dinakar Dhurjati</a>,
199 <a href="mailto:ggreif@gmail.com">Gabor Greif</a>,
200 <a href="mailto:jstanley@cs.uiuc.edu">Joel Stanley</a>,
201 <a href="mailto:rspencer@x10sys.com">Reid Spencer</a> and
202 <a href="mailto:owen@apple.com">Owen Anderson</a></p>
205 <!-- *********************************************************************** -->
206 <div class="doc_section">
207 <a name="introduction">Introduction </a>
209 <!-- *********************************************************************** -->
211 <div class="doc_text">
213 <p>This document is meant to highlight some of the important classes and
214 interfaces available in the LLVM source-base. This manual is not
215 intended to explain what LLVM is, how it works, and what LLVM code looks
216 like. It assumes that you know the basics of LLVM and are interested
217 in writing transformations or otherwise analyzing or manipulating the
220 <p>This document should get you oriented so that you can find your
221 way in the continuously growing source code that makes up the LLVM
222 infrastructure. Note that this manual is not intended to serve as a
223 replacement for reading the source code, so if you think there should be
224 a method in one of these classes to do something, but it's not listed,
225 check the source. Links to the <a href="/doxygen/">doxygen</a> sources
226 are provided to make this as easy as possible.</p>
228 <p>The first section of this document describes general information that is
229 useful to know when working in the LLVM infrastructure, and the second describes
230 the Core LLVM classes. In the future this manual will be extended with
231 information describing how to use extension libraries, such as dominator
232 information, CFG traversal routines, and useful utilities like the <tt><a
233 href="/doxygen/InstVisitor_8h-source.html">InstVisitor</a></tt> template.</p>
237 <!-- *********************************************************************** -->
238 <div class="doc_section">
239 <a name="general">General Information</a>
241 <!-- *********************************************************************** -->
243 <div class="doc_text">
245 <p>This section contains general information that is useful if you are working
246 in the LLVM source-base, but that isn't specific to any particular API.</p>
250 <!-- ======================================================================= -->
251 <div class="doc_subsection">
252 <a name="stl">The C++ Standard Template Library</a>
255 <div class="doc_text">
257 <p>LLVM makes heavy use of the C++ Standard Template Library (STL),
258 perhaps much more than you are used to, or have seen before. Because of
259 this, you might want to do a little background reading in the
260 techniques used and capabilities of the library. There are many good
261 pages that discuss the STL, and several books on the subject that you
262 can get, so it will not be discussed in this document.</p>
264 <p>Here are some useful links:</p>
268 <li><a href="http://www.dinkumware.com/refxcpp.html">Dinkumware C++ Library
269 reference</a> - an excellent reference for the STL and other parts of the
270 standard C++ library.</li>
272 <li><a href="http://www.tempest-sw.com/cpp/">C++ In a Nutshell</a> - This is an
273 O'Reilly book in the making. It has a decent Standard Library
274 Reference that rivals Dinkumware's, and is unfortunately no longer free since the
275 book has been published.</li>
277 <li><a href="http://www.parashift.com/c++-faq-lite/">C++ Frequently Asked
280 <li><a href="http://www.sgi.com/tech/stl/">SGI's STL Programmer's Guide</a> -
282 href="http://www.sgi.com/tech/stl/stl_introduction.html">Introduction to the
285 <li><a href="http://www.research.att.com/%7Ebs/C++.html">Bjarne Stroustrup's C++
288 <li><a href="http://64.78.49.204/">
289 Bruce Eckel's Thinking in C++, 2nd ed. Volume 2 Revision 4.0 (even better, get
294 <p>You are also encouraged to take a look at the <a
295 href="CodingStandards.html">LLVM Coding Standards</a> guide which focuses on how
296 to write maintainable code more than where to put your curly braces.</p>
300 <!-- ======================================================================= -->
301 <div class="doc_subsection">
302 <a name="stl">Other useful references</a>
305 <div class="doc_text">
308 <li><a href="http://www.psc.edu/%7Esemke/cvs_branches.html">CVS
309 Branch and Tag Primer</a></li>
310 <li><a href="http://www.fortran-2000.com/ArnaudRecipes/sharedlib.html">Using
311 static and shared libraries across platforms</a></li>
316 <!-- *********************************************************************** -->
317 <div class="doc_section">
318 <a name="apis">Important and useful LLVM APIs</a>
320 <!-- *********************************************************************** -->
322 <div class="doc_text">
324 <p>Here we highlight some LLVM APIs that are generally useful and good to
325 know about when writing transformations.</p>
329 <!-- ======================================================================= -->
330 <div class="doc_subsection">
331 <a name="isa">The <tt>isa<></tt>, <tt>cast<></tt> and
332 <tt>dyn_cast<></tt> templates</a>
335 <div class="doc_text">
337 <p>The LLVM source-base makes extensive use of a custom form of RTTI.
338 These templates have many similarities to the C++ <tt>dynamic_cast<></tt>
339 operator, but they don't have some drawbacks (primarily stemming from
340 the fact that <tt>dynamic_cast<></tt> only works on classes that
341 have a v-table). Because they are used so often, you must know what they
342 do and how they work. All of these templates are defined in the <a
343 href="/doxygen/Casting_8h-source.html"><tt>llvm/Support/Casting.h</tt></a>
344 file (note that you very rarely have to include this file directly).</p>
347 <dt><tt>isa<></tt>: </dt>
349 <dd><p>The <tt>isa<></tt> operator works exactly like the Java
350 "<tt>instanceof</tt>" operator. It returns true or false depending on whether
351 a reference or pointer points to an instance of the specified class. This can
352 be very useful for constraint checking of various sorts (example below).</p>
355 <dt><tt>cast<></tt>: </dt>
357 <dd><p>The <tt>cast<></tt> operator is a "checked cast" operation. It
358 converts a pointer or reference from a base class to a derived class, causing
359 an assertion failure if it is not really an instance of the right type. This
360 should be used in cases where you have some information that makes you believe
361 that something is of the right type. An example of the <tt>isa<></tt>
362 and <tt>cast<></tt> template is:</p>
364 <div class="doc_code">
366 static bool isLoopInvariant(const <a href="#Value">Value</a> *V, const Loop *L) {
367 if (isa<<a href="#Constant">Constant</a>>(V) || isa<<a href="#Argument">Argument</a>>(V) || isa<<a href="#GlobalValue">GlobalValue</a>>(V))
370 // <i>Otherwise, it must be an instruction...</i>
371 return !L->contains(cast<<a href="#Instruction">Instruction</a>>(V)->getParent());
376 <p>Note that you should <b>not</b> use an <tt>isa<></tt> test followed
377 by a <tt>cast<></tt>, for that use the <tt>dyn_cast<></tt>
382 <dt><tt>dyn_cast<></tt>:</dt>
384 <dd><p>The <tt>dyn_cast<></tt> operator is a "checking cast" operation.
385 It checks to see if the operand is of the specified type, and if so, returns a
386 pointer to it (this operator does not work with references). If the operand is
387 not of the correct type, a null pointer is returned. Thus, this works very
388 much like the <tt>dynamic_cast<></tt> operator in C++, and should be
389 used in the same circumstances. Typically, the <tt>dyn_cast<></tt>
390 operator is used in an <tt>if</tt> statement or some other flow control
391 statement like this:</p>
393 <div class="doc_code">
395 if (<a href="#AllocationInst">AllocationInst</a> *AI = dyn_cast<<a href="#AllocationInst">AllocationInst</a>>(Val)) {
401 <p>This form of the <tt>if</tt> statement effectively combines together a call
402 to <tt>isa<></tt> and a call to <tt>cast<></tt> into one
403 statement, which is very convenient.</p>
405 <p>Note that the <tt>dyn_cast<></tt> operator, like C++'s
406 <tt>dynamic_cast<></tt> or Java's <tt>instanceof</tt> operator, can be
407 abused. In particular, you should not use big chained <tt>if/then/else</tt>
408 blocks to check for lots of different variants of classes. If you find
409 yourself wanting to do this, it is much cleaner and more efficient to use the
410 <tt>InstVisitor</tt> class to dispatch over the instruction type directly.</p>
414 <dt><tt>cast_or_null<></tt>: </dt>
416 <dd><p>The <tt>cast_or_null<></tt> operator works just like the
417 <tt>cast<></tt> operator, except that it allows for a null pointer as an
418 argument (which it then propagates). This can sometimes be useful, allowing
419 you to combine several null checks into one.</p></dd>
421 <dt><tt>dyn_cast_or_null<></tt>: </dt>
423 <dd><p>The <tt>dyn_cast_or_null<></tt> operator works just like the
424 <tt>dyn_cast<></tt> operator, except that it allows for a null pointer
425 as an argument (which it then propagates). This can sometimes be useful,
426 allowing you to combine several null checks into one.</p></dd>
430 <p>These five templates can be used with any classes, whether they have a
431 v-table or not. To add support for these templates, you simply need to add
432 <tt>classof</tt> static methods to the class you are interested casting
433 to. Describing this is currently outside the scope of this document, but there
434 are lots of examples in the LLVM source base.</p>
439 <!-- ======================================================================= -->
440 <div class="doc_subsection">
441 <a name="string_apis">Passing strings (the <tt>StringRef</tt>
442 and <tt>Twine</tt> classes)</a>
445 <div class="doc_text">
447 <p>Although LLVM generally does not do much string manipulation, we do have
448 several important APIs which take strings. Two important examples are the
449 Value class -- which has names for instructions, functions, etc. -- and the
450 StringMap class which is used extensively in LLVM and Clang.</p>
452 <p>These are generic classes, and they need to be able to accept strings which
453 may have embedded null characters. Therefore, they cannot simply take
454 a <tt>const char *</tt>, and taking a <tt>const std::string&</tt> requires
455 clients to perform a heap allocation which is usually unnecessary. Instead,
456 many LLVM APIs use a <tt>const StringRef&</tt> or a <tt>const
457 Twine&</tt> for passing strings efficiently.</p>
461 <!-- _______________________________________________________________________ -->
462 <div class="doc_subsubsection">
463 <a name="StringRef">The <tt>StringRef</tt> class</a>
466 <div class="doc_text">
468 <p>The <tt>StringRef</tt> data type represents a reference to a constant string
469 (a character array and a length) and supports the common operations available
470 on <tt>std:string</tt>, but does not require heap allocation.</p>
472 <p>It can be implicitly constructed using a C style null-terminated string,
473 an <tt>std::string</tt>, or explicitly with a character pointer and length.
474 For example, the <tt>StringRef</tt> find function is declared as:</p>
476 <div class="doc_code">
477 iterator find(const StringRef &Key);
480 <p>and clients can call it using any one of:</p>
482 <div class="doc_code">
484 Map.find("foo"); <i>// Lookup "foo"</i>
485 Map.find(std::string("bar")); <i>// Lookup "bar"</i>
486 Map.find(StringRef("\0baz", 4)); <i>// Lookup "\0baz"</i>
490 <p>Similarly, APIs which need to return a string may return a <tt>StringRef</tt>
491 instance, which can be used directly or converted to an <tt>std::string</tt>
492 using the <tt>str</tt> member function. See
493 "<tt><a href="/doxygen/classllvm_1_1StringRef_8h-source.html">llvm/ADT/StringRef.h</a></tt>"
494 for more information.</p>
496 <p>You should rarely use the <tt>StringRef</tt> class directly, because it contains
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).</p>
502 <!-- _______________________________________________________________________ -->
503 <div class="doc_subsubsection">
504 <a name="Twine">The <tt>Twine</tt> class</a>
507 <div class="doc_text">
509 <p>The <tt>Twine</tt> class is an efficient way for APIs to accept concatenated
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>
513 <div class="doc_code">
515 New = CmpInst::Create(<i>...</i>, SO->getName() + ".cmp");
519 <p>The <tt>Twine</tt> class is effectively a
520 lightweight <a href="http://en.wikipedia.org/wiki/Rope_(computer_science)">rope</a>
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
523 strings, an <tt>std::string</tt>, or a <tt>StringRef</tt>). The twine delays the
524 actual concatentation 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
528 "<tt><a href="/doxygen/classllvm_1_1Twine_8h-source.html">llvm/ADT/Twine.h</a></tt>"
529 for more information.</p>
531 <p>As with a <tt>StringRef</tt>, <tt>Twine</tt> objects point to external memory
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>
539 <!-- ======================================================================= -->
540 <div class="doc_subsection">
541 <a name="DEBUG">The <tt>DEBUG()</tt> macro and <tt>-debug</tt> option</a>
544 <div class="doc_text">
546 <p>Often when working on your pass you will put a bunch of debugging printouts
547 and other code into your pass. After you get it working, you want to remove
548 it, but you may need it again in the future (to work out new bugs that you run
551 <p> Naturally, because of this, you don't want to delete the debug printouts,
552 but you don't want them to always be noisy. A standard compromise is to comment
553 them out, allowing you to enable them if you need them in the future.</p>
555 <p>The "<tt><a href="/doxygen/Debug_8h-source.html">llvm/Support/Debug.h</a></tt>"
556 file provides a macro named <tt>DEBUG()</tt> that is a much nicer solution to
557 this problem. Basically, you can put arbitrary code into the argument of the
558 <tt>DEBUG</tt> macro, and it is only executed if '<tt>opt</tt>' (or any other
559 tool) is run with the '<tt>-debug</tt>' command line argument:</p>
561 <div class="doc_code">
563 DEBUG(errs() << "I am here!\n");
567 <p>Then you can run your pass like this:</p>
569 <div class="doc_code">
571 $ opt < a.bc > /dev/null -mypass
572 <i><no output></i>
573 $ opt < a.bc > /dev/null -mypass -debug
578 <p>Using the <tt>DEBUG()</tt> macro instead of a home-brewed solution allows you
579 to not have to create "yet another" command line option for the debug output for
580 your pass. Note that <tt>DEBUG()</tt> macros are disabled for optimized builds,
581 so they do not cause a performance impact at all (for the same reason, they
582 should also not contain side-effects!).</p>
584 <p>One additional nice thing about the <tt>DEBUG()</tt> macro is that you can
585 enable or disable it directly in gdb. Just use "<tt>set DebugFlag=0</tt>" or
586 "<tt>set DebugFlag=1</tt>" from the gdb if the program is running. If the
587 program hasn't been started yet, you can always just run it with
592 <!-- _______________________________________________________________________ -->
593 <div class="doc_subsubsection">
594 <a name="DEBUG_TYPE">Fine grained debug info with <tt>DEBUG_TYPE</tt> and
595 the <tt>-debug-only</tt> option</a>
598 <div class="doc_text">
600 <p>Sometimes you may find yourself in a situation where enabling <tt>-debug</tt>
601 just turns on <b>too much</b> information (such as when working on the code
602 generator). If you want to enable debug information with more fine-grained
603 control, you define the <tt>DEBUG_TYPE</tt> macro and the <tt>-debug</tt> only
604 option as follows:</p>
606 <div class="doc_code">
609 DEBUG(errs() << "No debug type\n");
610 #define DEBUG_TYPE "foo"
611 DEBUG(errs() << "'foo' debug type\n");
613 #define DEBUG_TYPE "bar"
614 DEBUG(errs() << "'bar' debug type\n"));
616 #define DEBUG_TYPE ""
617 DEBUG(errs() << "No debug type (2)\n");
621 <p>Then you can run your pass like this:</p>
623 <div class="doc_code">
625 $ opt < a.bc > /dev/null -mypass
626 <i><no output></i>
627 $ opt < a.bc > /dev/null -mypass -debug
632 $ opt < a.bc > /dev/null -mypass -debug-only=foo
634 $ opt < a.bc > /dev/null -mypass -debug-only=bar
639 <p>Of course, in practice, you should only set <tt>DEBUG_TYPE</tt> at the top of
640 a file, to specify the debug type for the entire module (if you do this before
641 you <tt>#include "llvm/Support/Debug.h"</tt>, you don't have to insert the ugly
642 <tt>#undef</tt>'s). Also, you should use names more meaningful than "foo" and
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
646 for instruction scheduling to be enabled with <tt>-debug-type=InstrSched</tt>,
647 even if the source lives in multiple files.</p>
649 <p>The <tt>DEBUG_WITH_TYPE</tt> macro is also available for situations where you
650 would like to set <tt>DEBUG_TYPE</tt>, but only for one specific <tt>DEBUG</tt>
651 statement. It takes an additional first parameter, which is the type to use. For
652 example, the preceeding example could be written as:</p>
655 <div class="doc_code">
657 DEBUG_WITH_TYPE("", errs() << "No debug type\n");
658 DEBUG_WITH_TYPE("foo", errs() << "'foo' debug type\n");
659 DEBUG_WITH_TYPE("bar", errs() << "'bar' debug type\n"));
660 DEBUG_WITH_TYPE("", errs() << "No debug type (2)\n");
666 <!-- ======================================================================= -->
667 <div class="doc_subsection">
668 <a name="Statistic">The <tt>Statistic</tt> class & <tt>-stats</tt>
672 <div class="doc_text">
675 href="/doxygen/Statistic_8h-source.html">llvm/ADT/Statistic.h</a></tt>" file
676 provides a class named <tt>Statistic</tt> that is used as a unified way to
677 keep track of what the LLVM compiler is doing and how effective various
678 optimizations are. It is useful to see what optimizations are contributing to
679 making a particular program run faster.</p>
681 <p>Often you may run your pass on some big program, and you're interested to see
682 how many times it makes a certain transformation. Although you can do this with
683 hand inspection, or some ad-hoc method, this is a real pain and not very useful
684 for big programs. Using the <tt>Statistic</tt> class makes it very easy to
685 keep track of this information, and the calculated information is presented in a
686 uniform manner with the rest of the passes being executed.</p>
688 <p>There are many examples of <tt>Statistic</tt> uses, but the basics of using
689 it are as follows:</p>
692 <li><p>Define your statistic like this:</p>
694 <div class="doc_code">
696 #define <a href="#DEBUG_TYPE">DEBUG_TYPE</a> "mypassname" <i>// This goes before any #includes.</i>
697 STATISTIC(NumXForms, "The # of times I did stuff");
701 <p>The <tt>STATISTIC</tt> macro defines a static variable, whose name is
702 specified by the first argument. The pass name is taken from the DEBUG_TYPE
703 macro, and the description is taken from the second argument. The variable
704 defined ("NumXForms" in this case) acts like an unsigned integer.</p></li>
706 <li><p>Whenever you make a transformation, bump the counter:</p>
708 <div class="doc_code">
710 ++NumXForms; // <i>I did stuff!</i>
717 <p>That's all you have to do. To get '<tt>opt</tt>' to print out the
718 statistics gathered, use the '<tt>-stats</tt>' option:</p>
720 <div class="doc_code">
722 $ opt -stats -mypassname < program.bc > /dev/null
723 <i>... statistics output ...</i>
727 <p> When running <tt>opt</tt> on a C file from the SPEC benchmark
728 suite, it gives a report that looks like this:</p>
730 <div class="doc_code">
732 7646 bitcodewriter - Number of normal instructions
733 725 bitcodewriter - Number of oversized instructions
734 129996 bitcodewriter - Number of bitcode bytes written
735 2817 raise - Number of insts DCEd or constprop'd
736 3213 raise - Number of cast-of-self removed
737 5046 raise - Number of expression trees converted
738 75 raise - Number of other getelementptr's formed
739 138 raise - Number of load/store peepholes
740 42 deadtypeelim - Number of unused typenames removed from symtab
741 392 funcresolve - Number of varargs functions resolved
742 27 globaldce - Number of global variables removed
743 2 adce - Number of basic blocks removed
744 134 cee - Number of branches revectored
745 49 cee - Number of setcc instruction eliminated
746 532 gcse - Number of loads removed
747 2919 gcse - Number of instructions removed
748 86 indvars - Number of canonical indvars added
749 87 indvars - Number of aux indvars removed
750 25 instcombine - Number of dead inst eliminate
751 434 instcombine - Number of insts combined
752 248 licm - Number of load insts hoisted
753 1298 licm - Number of insts hoisted to a loop pre-header
754 3 licm - Number of insts hoisted to multiple loop preds (bad, no loop pre-header)
755 75 mem2reg - Number of alloca's promoted
756 1444 cfgsimplify - Number of blocks simplified
760 <p>Obviously, with so many optimizations, having a unified framework for this
761 stuff is very nice. Making your pass fit well into the framework makes it more
762 maintainable and useful.</p>
766 <!-- ======================================================================= -->
767 <div class="doc_subsection">
768 <a name="ViewGraph">Viewing graphs while debugging code</a>
771 <div class="doc_text">
773 <p>Several of the important data structures in LLVM are graphs: for example
774 CFGs made out of LLVM <a href="#BasicBlock">BasicBlock</a>s, CFGs made out of
775 LLVM <a href="CodeGenerator.html#machinebasicblock">MachineBasicBlock</a>s, and
776 <a href="CodeGenerator.html#selectiondag_intro">Instruction Selection
777 DAGs</a>. In many cases, while debugging various parts of the compiler, it is
778 nice to instantly visualize these graphs.</p>
780 <p>LLVM provides several callbacks that are available in a debug build to do
781 exactly that. If you call the <tt>Function::viewCFG()</tt> method, for example,
782 the current LLVM tool will pop up a window containing the CFG for the function
783 where each basic block is a node in the graph, and each node contains the
784 instructions in the block. Similarly, there also exists
785 <tt>Function::viewCFGOnly()</tt> (does not include the instructions), the
786 <tt>MachineFunction::viewCFG()</tt> and <tt>MachineFunction::viewCFGOnly()</tt>,
787 and the <tt>SelectionDAG::viewGraph()</tt> methods. Within GDB, for example,
788 you can usually use something like <tt>call DAG.viewGraph()</tt> to pop
789 up a window. Alternatively, you can sprinkle calls to these functions in your
790 code in places you want to debug.</p>
792 <p>Getting this to work requires a small amount of configuration. On Unix
793 systems with X11, install the <a href="http://www.graphviz.org">graphviz</a>
794 toolkit, and make sure 'dot' and 'gv' are in your path. If you are running on
795 Mac OS/X, download and install the Mac OS/X <a
796 href="http://www.pixelglow.com/graphviz/">Graphviz program</a>, and add
797 <tt>/Applications/Graphviz.app/Contents/MacOS/</tt> (or wherever you install
798 it) to your path. Once in your system and path are set up, rerun the LLVM
799 configure script and rebuild LLVM to enable this functionality.</p>
801 <p><tt>SelectionDAG</tt> has been extended to make it easier to locate
802 <i>interesting</i> nodes in large complex graphs. From gdb, if you
803 <tt>call DAG.setGraphColor(<i>node</i>, "<i>color</i>")</tt>, then the
804 next <tt>call DAG.viewGraph()</tt> would highlight the node in the
805 specified color (choices of colors can be found at <a
806 href="http://www.graphviz.org/doc/info/colors.html">colors</a>.) More
807 complex node attributes can be provided with <tt>call
808 DAG.setGraphAttrs(<i>node</i>, "<i>attributes</i>")</tt> (choices can be
809 found at <a href="http://www.graphviz.org/doc/info/attrs.html">Graph
810 Attributes</a>.) If you want to restart and clear all the current graph
811 attributes, then you can <tt>call DAG.clearGraphAttrs()</tt>. </p>
815 <!-- *********************************************************************** -->
816 <div class="doc_section">
817 <a name="datastructure">Picking the Right Data Structure for a Task</a>
819 <!-- *********************************************************************** -->
821 <div class="doc_text">
823 <p>LLVM has a plethora of data structures in the <tt>llvm/ADT/</tt> directory,
824 and we commonly use STL data structures. This section describes the trade-offs
825 you should consider when you pick one.</p>
828 The first step is a choose your own adventure: do you want a sequential
829 container, a set-like container, or a map-like container? The most important
830 thing when choosing a container is the algorithmic properties of how you plan to
831 access the container. Based on that, you should use:</p>
834 <li>a <a href="#ds_map">map-like</a> container if you need efficient look-up
835 of an value based on another value. Map-like containers also support
836 efficient queries for containment (whether a key is in the map). Map-like
837 containers generally do not support efficient reverse mapping (values to
838 keys). If you need that, use two maps. Some map-like containers also
839 support efficient iteration through the keys in sorted order. Map-like
840 containers are the most expensive sort, only use them if you need one of
841 these capabilities.</li>
843 <li>a <a href="#ds_set">set-like</a> container if you need to put a bunch of
844 stuff into a container that automatically eliminates duplicates. Some
845 set-like containers support efficient iteration through the elements in
846 sorted order. Set-like containers are more expensive than sequential
850 <li>a <a href="#ds_sequential">sequential</a> container provides
851 the most efficient way to add elements and keeps track of the order they are
852 added to the collection. They permit duplicates and support efficient
853 iteration, but do not support efficient look-up based on a key.
856 <li>a <a href="#ds_string">string</a> container is a specialized sequential
857 container or reference structure that is used for character or byte
860 <li>a <a href="#ds_bit">bit</a> container provides an efficient way to store and
861 perform set operations on sets of numeric id's, while automatically
862 eliminating duplicates. Bit containers require a maximum of 1 bit for each
863 identifier you want to store.
868 Once the proper category of container is determined, you can fine tune the
869 memory use, constant factors, and cache behaviors of access by intelligently
870 picking a member of the category. Note that constant factors and cache behavior
871 can be a big deal. If you have a vector that usually only contains a few
872 elements (but could contain many), for example, it's much better to use
873 <a href="#dss_smallvector">SmallVector</a> than <a href="#dss_vector">vector</a>
874 . Doing so avoids (relatively) expensive malloc/free calls, which dwarf the
875 cost of adding the elements to the container. </p>
879 <!-- ======================================================================= -->
880 <div class="doc_subsection">
881 <a name="ds_sequential">Sequential Containers (std::vector, std::list, etc)</a>
884 <div class="doc_text">
885 There are a variety of sequential containers available for you, based on your
886 needs. Pick the first in this section that will do what you want.
889 <!-- _______________________________________________________________________ -->
890 <div class="doc_subsubsection">
891 <a name="dss_fixedarrays">Fixed Size Arrays</a>
894 <div class="doc_text">
895 <p>Fixed size arrays are very simple and very fast. They are good if you know
896 exactly how many elements you have, or you have a (low) upper bound on how many
900 <!-- _______________________________________________________________________ -->
901 <div class="doc_subsubsection">
902 <a name="dss_heaparrays">Heap Allocated Arrays</a>
905 <div class="doc_text">
906 <p>Heap allocated arrays (new[] + delete[]) are also simple. They are good if
907 the number of elements is variable, if you know how many elements you will need
908 before the array is allocated, and if the array is usually large (if not,
909 consider a <a href="#dss_smallvector">SmallVector</a>). The cost of a heap
910 allocated array is the cost of the new/delete (aka malloc/free). Also note that
911 if you are allocating an array of a type with a constructor, the constructor and
912 destructors will be run for every element in the array (re-sizable vectors only
913 construct those elements actually used).</p>
916 <!-- _______________________________________________________________________ -->
917 <div class="doc_subsubsection">
918 <a name="dss_smallvector">"llvm/ADT/SmallVector.h"</a>
921 <div class="doc_text">
922 <p><tt>SmallVector<Type, N></tt> is a simple class that looks and smells
923 just like <tt>vector<Type></tt>:
924 it supports efficient iteration, lays out elements in memory order (so you can
925 do pointer arithmetic between elements), supports efficient push_back/pop_back
926 operations, supports efficient random access to its elements, etc.</p>
928 <p>The advantage of SmallVector is that it allocates space for
929 some number of elements (N) <b>in the object itself</b>. Because of this, if
930 the SmallVector is dynamically smaller than N, no malloc is performed. This can
931 be a big win in cases where the malloc/free call is far more expensive than the
932 code that fiddles around with the elements.</p>
934 <p>This is good for vectors that are "usually small" (e.g. the number of
935 predecessors/successors of a block is usually less than 8). On the other hand,
936 this makes the size of the SmallVector itself large, so you don't want to
937 allocate lots of them (doing so will waste a lot of space). As such,
938 SmallVectors are most useful when on the stack.</p>
940 <p>SmallVector also provides a nice portable and efficient replacement for
945 <!-- _______________________________________________________________________ -->
946 <div class="doc_subsubsection">
947 <a name="dss_vector"><vector></a>
950 <div class="doc_text">
952 std::vector is well loved and respected. It is useful when SmallVector isn't:
953 when the size of the vector is often large (thus the small optimization will
954 rarely be a benefit) or if you will be allocating many instances of the vector
955 itself (which would waste space for elements that aren't in the container).
956 vector is also useful when interfacing with code that expects vectors :).
959 <p>One worthwhile note about std::vector: avoid code like this:</p>
961 <div class="doc_code">
964 std::vector<foo> V;
970 <p>Instead, write this as:</p>
972 <div class="doc_code">
974 std::vector<foo> V;
982 <p>Doing so will save (at least) one heap allocation and free per iteration of
987 <!-- _______________________________________________________________________ -->
988 <div class="doc_subsubsection">
989 <a name="dss_deque"><deque></a>
992 <div class="doc_text">
993 <p>std::deque is, in some senses, a generalized version of std::vector. Like
994 std::vector, it provides constant time random access and other similar
995 properties, but it also provides efficient access to the front of the list. It
996 does not guarantee continuity of elements within memory.</p>
998 <p>In exchange for this extra flexibility, std::deque has significantly higher
999 constant factor costs than std::vector. If possible, use std::vector or
1000 something cheaper.</p>
1003 <!-- _______________________________________________________________________ -->
1004 <div class="doc_subsubsection">
1005 <a name="dss_list"><list></a>
1008 <div class="doc_text">
1009 <p>std::list is an extremely inefficient class that is rarely useful.
1010 It performs a heap allocation for every element inserted into it, thus having an
1011 extremely high constant factor, particularly for small data types. std::list
1012 also only supports bidirectional iteration, not random access iteration.</p>
1014 <p>In exchange for this high cost, std::list supports efficient access to both
1015 ends of the list (like std::deque, but unlike std::vector or SmallVector). In
1016 addition, the iterator invalidation characteristics of std::list are stronger
1017 than that of a vector class: inserting or removing an element into the list does
1018 not invalidate iterator or pointers to other elements in the list.</p>
1021 <!-- _______________________________________________________________________ -->
1022 <div class="doc_subsubsection">
1023 <a name="dss_ilist">llvm/ADT/ilist.h</a>
1026 <div class="doc_text">
1027 <p><tt>ilist<T></tt> implements an 'intrusive' doubly-linked list. It is
1028 intrusive, because it requires the element to store and provide access to the
1029 prev/next pointers for the list.</p>
1031 <p><tt>ilist</tt> has the same drawbacks as <tt>std::list</tt>, and additionally
1032 requires an <tt>ilist_traits</tt> implementation for the element type, but it
1033 provides some novel characteristics. In particular, it can efficiently store
1034 polymorphic objects, the traits class is informed when an element is inserted or
1035 removed from the list, and <tt>ilist</tt>s are guaranteed to support a
1036 constant-time splice operation.</p>
1038 <p>These properties are exactly what we want for things like
1039 <tt>Instruction</tt>s and basic blocks, which is why these are implemented with
1040 <tt>ilist</tt>s.</p>
1042 Related classes of interest are explained in the following subsections:
1044 <li><a href="#dss_ilist_traits">ilist_traits</a></li>
1045 <li><a href="#dss_iplist">iplist</a></li>
1046 <li><a href="#dss_ilist_node">llvm/ADT/ilist_node.h</a></li>
1047 <li><a href="#dss_ilist_sentinel">Sentinels</a></li>
1051 <!-- _______________________________________________________________________ -->
1052 <div class="doc_subsubsection">
1053 <a name="dss_ilist_traits">ilist_traits</a>
1056 <div class="doc_text">
1057 <p><tt>ilist_traits<T></tt> is <tt>ilist<T></tt>'s customization
1058 mechanism. <tt>iplist<T></tt> (and consequently <tt>ilist<T></tt>)
1059 publicly derive from this traits class.</p>
1062 <!-- _______________________________________________________________________ -->
1063 <div class="doc_subsubsection">
1064 <a name="dss_iplist">iplist</a>
1067 <div class="doc_text">
1068 <p><tt>iplist<T></tt> is <tt>ilist<T></tt>'s base and as such
1069 supports a slightly narrower interface. Notably, inserters from
1070 <tt>T&</tt> are absent.</p>
1072 <p><tt>ilist_traits<T></tt> is a public base of this class and can be
1073 used for a wide variety of customizations.</p>
1076 <!-- _______________________________________________________________________ -->
1077 <div class="doc_subsubsection">
1078 <a name="dss_ilist_node">llvm/ADT/ilist_node.h</a>
1081 <div class="doc_text">
1082 <p><tt>ilist_node<T></tt> implements a the forward and backward links
1083 that are expected by the <tt>ilist<T></tt> (and analogous containers)
1084 in the default manner.</p>
1086 <p><tt>ilist_node<T></tt>s are meant to be embedded in the node type
1087 <tt>T</tt>, usually <tt>T</tt> publicly derives from
1088 <tt>ilist_node<T></tt>.</p>
1091 <!-- _______________________________________________________________________ -->
1092 <div class="doc_subsubsection">
1093 <a name="dss_ilist_sentinel">Sentinels</a>
1096 <div class="doc_text">
1097 <p><tt>ilist</tt>s have another speciality that must be considered. To be a good
1098 citizen in the C++ ecosystem, it needs to support the standard container
1099 operations, such as <tt>begin</tt> and <tt>end</tt> iterators, etc. Also, the
1100 <tt>operator--</tt> must work correctly on the <tt>end</tt> iterator in the
1101 case of non-empty <tt>ilist</tt>s.</p>
1103 <p>The only sensible solution to this problem is to allocate a so-called
1104 <i>sentinel</i> along with the intrusive list, which serves as the <tt>end</tt>
1105 iterator, providing the back-link to the last element. However conforming to the
1106 C++ convention it is illegal to <tt>operator++</tt> beyond the sentinel and it
1107 also must not be dereferenced.</p>
1109 <p>These constraints allow for some implementation freedom to the <tt>ilist</tt>
1110 how to allocate and store the sentinel. The corresponding policy is dictated
1111 by <tt>ilist_traits<T></tt>. By default a <tt>T</tt> gets heap-allocated
1112 whenever the need for a sentinel arises.</p>
1114 <p>While the default policy is sufficient in most cases, it may break down when
1115 <tt>T</tt> does not provide a default constructor. Also, in the case of many
1116 instances of <tt>ilist</tt>s, the memory overhead of the associated sentinels
1117 is wasted. To alleviate the situation with numerous and voluminous
1118 <tt>T</tt>-sentinels, sometimes a trick is employed, leading to <i>ghostly
1121 <p>Ghostly sentinels are obtained by specially-crafted <tt>ilist_traits<T></tt>
1122 which superpose the sentinel with the <tt>ilist</tt> instance in memory. Pointer
1123 arithmetic is used to obtain the sentinel, which is relative to the
1124 <tt>ilist</tt>'s <tt>this</tt> pointer. The <tt>ilist</tt> is augmented by an
1125 extra pointer, which serves as the back-link of the sentinel. This is the only
1126 field in the ghostly sentinel which can be legally accessed.</p>
1129 <!-- _______________________________________________________________________ -->
1130 <div class="doc_subsubsection">
1131 <a name="dss_other">Other Sequential Container options</a>
1134 <div class="doc_text">
1135 <p>Other STL containers are available, such as std::string.</p>
1137 <p>There are also various STL adapter classes such as std::queue,
1138 std::priority_queue, std::stack, etc. These provide simplified access to an
1139 underlying container but don't affect the cost of the container itself.</p>
1144 <!-- ======================================================================= -->
1145 <div class="doc_subsection">
1146 <a name="ds_set">Set-Like Containers (std::set, SmallSet, SetVector, etc)</a>
1149 <div class="doc_text">
1151 <p>Set-like containers are useful when you need to canonicalize multiple values
1152 into a single representation. There are several different choices for how to do
1153 this, providing various trade-offs.</p>
1158 <!-- _______________________________________________________________________ -->
1159 <div class="doc_subsubsection">
1160 <a name="dss_sortedvectorset">A sorted 'vector'</a>
1163 <div class="doc_text">
1165 <p>If you intend to insert a lot of elements, then do a lot of queries, a
1166 great approach is to use a vector (or other sequential container) with
1167 std::sort+std::unique to remove duplicates. This approach works really well if
1168 your usage pattern has these two distinct phases (insert then query), and can be
1169 coupled with a good choice of <a href="#ds_sequential">sequential container</a>.
1173 This combination provides the several nice properties: the result data is
1174 contiguous in memory (good for cache locality), has few allocations, is easy to
1175 address (iterators in the final vector are just indices or pointers), and can be
1176 efficiently queried with a standard binary or radix search.</p>
1180 <!-- _______________________________________________________________________ -->
1181 <div class="doc_subsubsection">
1182 <a name="dss_smallset">"llvm/ADT/SmallSet.h"</a>
1185 <div class="doc_text">
1187 <p>If you have a set-like data structure that is usually small and whose elements
1188 are reasonably small, a <tt>SmallSet<Type, N></tt> is a good choice. This set
1189 has space for N elements in place (thus, if the set is dynamically smaller than
1190 N, no malloc traffic is required) and accesses them with a simple linear search.
1191 When the set grows beyond 'N' elements, it allocates a more expensive representation that
1192 guarantees efficient access (for most types, it falls back to std::set, but for
1193 pointers it uses something far better, <a
1194 href="#dss_smallptrset">SmallPtrSet</a>).</p>
1196 <p>The magic of this class is that it handles small sets extremely efficiently,
1197 but gracefully handles extremely large sets without loss of efficiency. The
1198 drawback is that the interface is quite small: it supports insertion, queries
1199 and erasing, but does not support iteration.</p>
1203 <!-- _______________________________________________________________________ -->
1204 <div class="doc_subsubsection">
1205 <a name="dss_smallptrset">"llvm/ADT/SmallPtrSet.h"</a>
1208 <div class="doc_text">
1210 <p>SmallPtrSet has all the advantages of SmallSet (and a SmallSet of pointers is
1211 transparently implemented with a SmallPtrSet), but also supports iterators. If
1212 more than 'N' insertions are performed, a single quadratically
1213 probed hash table is allocated and grows as needed, providing extremely
1214 efficient access (constant time insertion/deleting/queries with low constant
1215 factors) and is very stingy with malloc traffic.</p>
1217 <p>Note that, unlike std::set, the iterators of SmallPtrSet are invalidated
1218 whenever an insertion occurs. Also, the values visited by the iterators are not
1219 visited in sorted order.</p>
1223 <!-- _______________________________________________________________________ -->
1224 <div class="doc_subsubsection">
1225 <a name="dss_denseset">"llvm/ADT/DenseSet.h"</a>
1228 <div class="doc_text">
1231 DenseSet is a simple quadratically probed hash table. It excels at supporting
1232 small values: it uses a single allocation to hold all of the pairs that
1233 are currently inserted in the set. DenseSet is a great way to unique small
1234 values that are not simple pointers (use <a
1235 href="#dss_smallptrset">SmallPtrSet</a> for pointers). Note that DenseSet has
1236 the same requirements for the value type that <a
1237 href="#dss_densemap">DenseMap</a> has.
1242 <!-- _______________________________________________________________________ -->
1243 <div class="doc_subsubsection">
1244 <a name="dss_FoldingSet">"llvm/ADT/FoldingSet.h"</a>
1247 <div class="doc_text">
1250 FoldingSet is an aggregate class that is really good at uniquing
1251 expensive-to-create or polymorphic objects. It is a combination of a chained
1252 hash table with intrusive links (uniqued objects are required to inherit from
1253 FoldingSetNode) that uses <a href="#dss_smallvector">SmallVector</a> as part of
1256 <p>Consider a case where you want to implement a "getOrCreateFoo" method for
1257 a complex object (for example, a node in the code generator). The client has a
1258 description of *what* it wants to generate (it knows the opcode and all the
1259 operands), but we don't want to 'new' a node, then try inserting it into a set
1260 only to find out it already exists, at which point we would have to delete it
1261 and return the node that already exists.
1264 <p>To support this style of client, FoldingSet perform a query with a
1265 FoldingSetNodeID (which wraps SmallVector) that can be used to describe the
1266 element that we want to query for. The query either returns the element
1267 matching the ID or it returns an opaque ID that indicates where insertion should
1268 take place. Construction of the ID usually does not require heap traffic.</p>
1270 <p>Because FoldingSet uses intrusive links, it can support polymorphic objects
1271 in the set (for example, you can have SDNode instances mixed with LoadSDNodes).
1272 Because the elements are individually allocated, pointers to the elements are
1273 stable: inserting or removing elements does not invalidate any pointers to other
1279 <!-- _______________________________________________________________________ -->
1280 <div class="doc_subsubsection">
1281 <a name="dss_set"><set></a>
1284 <div class="doc_text">
1286 <p><tt>std::set</tt> is a reasonable all-around set class, which is decent at
1287 many things but great at nothing. std::set allocates memory for each element
1288 inserted (thus it is very malloc intensive) and typically stores three pointers
1289 per element in the set (thus adding a large amount of per-element space
1290 overhead). It offers guaranteed log(n) performance, which is not particularly
1291 fast from a complexity standpoint (particularly if the elements of the set are
1292 expensive to compare, like strings), and has extremely high constant factors for
1293 lookup, insertion and removal.</p>
1295 <p>The advantages of std::set are that its iterators are stable (deleting or
1296 inserting an element from the set does not affect iterators or pointers to other
1297 elements) and that iteration over the set is guaranteed to be in sorted order.
1298 If the elements in the set are large, then the relative overhead of the pointers
1299 and malloc traffic is not a big deal, but if the elements of the set are small,
1300 std::set is almost never a good choice.</p>
1304 <!-- _______________________________________________________________________ -->
1305 <div class="doc_subsubsection">
1306 <a name="dss_setvector">"llvm/ADT/SetVector.h"</a>
1309 <div class="doc_text">
1310 <p>LLVM's SetVector<Type> is an adapter class that combines your choice of
1311 a set-like container along with a <a href="#ds_sequential">Sequential
1312 Container</a>. The important property
1313 that this provides is efficient insertion with uniquing (duplicate elements are
1314 ignored) with iteration support. It implements this by inserting elements into
1315 both a set-like container and the sequential container, using the set-like
1316 container for uniquing and the sequential container for iteration.
1319 <p>The difference between SetVector and other sets is that the order of
1320 iteration is guaranteed to match the order of insertion into the SetVector.
1321 This property is really important for things like sets of pointers. Because
1322 pointer values are non-deterministic (e.g. vary across runs of the program on
1323 different machines), iterating over the pointers in the set will
1324 not be in a well-defined order.</p>
1327 The drawback of SetVector is that it requires twice as much space as a normal
1328 set and has the sum of constant factors from the set-like container and the
1329 sequential container that it uses. Use it *only* if you need to iterate over
1330 the elements in a deterministic order. SetVector is also expensive to delete
1331 elements out of (linear time), unless you use it's "pop_back" method, which is
1335 <p>SetVector is an adapter class that defaults to using std::vector and std::set
1336 for the underlying containers, so it is quite expensive. However,
1337 <tt>"llvm/ADT/SetVector.h"</tt> also provides a SmallSetVector class, which
1338 defaults to using a SmallVector and SmallSet of a specified size. If you use
1339 this, and if your sets are dynamically smaller than N, you will save a lot of
1344 <!-- _______________________________________________________________________ -->
1345 <div class="doc_subsubsection">
1346 <a name="dss_uniquevector">"llvm/ADT/UniqueVector.h"</a>
1349 <div class="doc_text">
1352 UniqueVector is similar to <a href="#dss_setvector">SetVector</a>, but it
1353 retains a unique ID for each element inserted into the set. It internally
1354 contains a map and a vector, and it assigns a unique ID for each value inserted
1357 <p>UniqueVector is very expensive: its cost is the sum of the cost of
1358 maintaining both the map and vector, it has high complexity, high constant
1359 factors, and produces a lot of malloc traffic. It should be avoided.</p>
1364 <!-- _______________________________________________________________________ -->
1365 <div class="doc_subsubsection">
1366 <a name="dss_otherset">Other Set-Like Container Options</a>
1369 <div class="doc_text">
1372 The STL provides several other options, such as std::multiset and the various
1373 "hash_set" like containers (whether from C++ TR1 or from the SGI library). We
1374 never use hash_set and unordered_set because they are generally very expensive
1375 (each insertion requires a malloc) and very non-portable.
1378 <p>std::multiset is useful if you're not interested in elimination of
1379 duplicates, but has all the drawbacks of std::set. A sorted vector (where you
1380 don't delete duplicate entries) or some other approach is almost always
1385 <!-- ======================================================================= -->
1386 <div class="doc_subsection">
1387 <a name="ds_map">Map-Like Containers (std::map, DenseMap, etc)</a>
1390 <div class="doc_text">
1391 Map-like containers are useful when you want to associate data to a key. As
1392 usual, there are a lot of different ways to do this. :)
1395 <!-- _______________________________________________________________________ -->
1396 <div class="doc_subsubsection">
1397 <a name="dss_sortedvectormap">A sorted 'vector'</a>
1400 <div class="doc_text">
1403 If your usage pattern follows a strict insert-then-query approach, you can
1404 trivially use the same approach as <a href="#dss_sortedvectorset">sorted vectors
1405 for set-like containers</a>. The only difference is that your query function
1406 (which uses std::lower_bound to get efficient log(n) lookup) should only compare
1407 the key, not both the key and value. This yields the same advantages as sorted
1412 <!-- _______________________________________________________________________ -->
1413 <div class="doc_subsubsection">
1414 <a name="dss_stringmap">"llvm/ADT/StringMap.h"</a>
1417 <div class="doc_text">
1420 Strings are commonly used as keys in maps, and they are difficult to support
1421 efficiently: they are variable length, inefficient to hash and compare when
1422 long, expensive to copy, etc. StringMap is a specialized container designed to
1423 cope with these issues. It supports mapping an arbitrary range of bytes to an
1424 arbitrary other object.</p>
1426 <p>The StringMap implementation uses a quadratically-probed hash table, where
1427 the buckets store a pointer to the heap allocated entries (and some other
1428 stuff). The entries in the map must be heap allocated because the strings are
1429 variable length. The string data (key) and the element object (value) are
1430 stored in the same allocation with the string data immediately after the element
1431 object. This container guarantees the "<tt>(char*)(&Value+1)</tt>" points
1432 to the key string for a value.</p>
1434 <p>The StringMap is very fast for several reasons: quadratic probing is very
1435 cache efficient for lookups, the hash value of strings in buckets is not
1436 recomputed when lookup up an element, StringMap rarely has to touch the
1437 memory for unrelated objects when looking up a value (even when hash collisions
1438 happen), hash table growth does not recompute the hash values for strings
1439 already in the table, and each pair in the map is store in a single allocation
1440 (the string data is stored in the same allocation as the Value of a pair).</p>
1442 <p>StringMap also provides query methods that take byte ranges, so it only ever
1443 copies a string if a value is inserted into the table.</p>
1446 <!-- _______________________________________________________________________ -->
1447 <div class="doc_subsubsection">
1448 <a name="dss_indexedmap">"llvm/ADT/IndexedMap.h"</a>
1451 <div class="doc_text">
1453 IndexedMap is a specialized container for mapping small dense integers (or
1454 values that can be mapped to small dense integers) to some other type. It is
1455 internally implemented as a vector with a mapping function that maps the keys to
1456 the dense integer range.
1460 This is useful for cases like virtual registers in the LLVM code generator: they
1461 have a dense mapping that is offset by a compile-time constant (the first
1462 virtual register ID).</p>
1466 <!-- _______________________________________________________________________ -->
1467 <div class="doc_subsubsection">
1468 <a name="dss_densemap">"llvm/ADT/DenseMap.h"</a>
1471 <div class="doc_text">
1474 DenseMap is a simple quadratically probed hash table. It excels at supporting
1475 small keys and values: it uses a single allocation to hold all of the pairs that
1476 are currently inserted in the map. DenseMap is a great way to map pointers to
1477 pointers, or map other small types to each other.
1481 There are several aspects of DenseMap that you should be aware of, however. The
1482 iterators in a densemap are invalidated whenever an insertion occurs, unlike
1483 map. Also, because DenseMap allocates space for a large number of key/value
1484 pairs (it starts with 64 by default), it will waste a lot of space if your keys
1485 or values are large. Finally, you must implement a partial specialization of
1486 DenseMapInfo for the key that you want, if it isn't already supported. This
1487 is required to tell DenseMap about two special marker values (which can never be
1488 inserted into the map) that it needs internally.</p>
1492 <!-- _______________________________________________________________________ -->
1493 <div class="doc_subsubsection">
1494 <a name="dss_map"><map></a>
1497 <div class="doc_text">
1500 std::map has similar characteristics to <a href="#dss_set">std::set</a>: it uses
1501 a single allocation per pair inserted into the map, it offers log(n) lookup with
1502 an extremely large constant factor, imposes a space penalty of 3 pointers per
1503 pair in the map, etc.</p>
1505 <p>std::map is most useful when your keys or values are very large, if you need
1506 to iterate over the collection in sorted order, or if you need stable iterators
1507 into the map (i.e. they don't get invalidated if an insertion or deletion of
1508 another element takes place).</p>
1512 <!-- _______________________________________________________________________ -->
1513 <div class="doc_subsubsection">
1514 <a name="dss_othermap">Other Map-Like Container Options</a>
1517 <div class="doc_text">
1520 The STL provides several other options, such as std::multimap and the various
1521 "hash_map" like containers (whether from C++ TR1 or from the SGI library). We
1522 never use hash_set and unordered_set because they are generally very expensive
1523 (each insertion requires a malloc) and very non-portable.</p>
1525 <p>std::multimap is useful if you want to map a key to multiple values, but has
1526 all the drawbacks of std::map. A sorted vector or some other approach is almost
1531 <!-- ======================================================================= -->
1532 <div class="doc_subsection">
1533 <a name="ds_string">String-like containers</a>
1536 <div class="doc_text">
1539 TODO: const char* vs stringref vs smallstring vs std::string. Describe twine,
1540 xref to #string_apis.
1545 <!-- ======================================================================= -->
1546 <div class="doc_subsection">
1547 <a name="ds_bit">Bit storage containers (BitVector, SparseBitVector)</a>
1550 <div class="doc_text">
1551 <p>Unlike the other containers, there are only two bit storage containers, and
1552 choosing when to use each is relatively straightforward.</p>
1554 <p>One additional option is
1555 <tt>std::vector<bool></tt>: we discourage its use for two reasons 1) the
1556 implementation in many common compilers (e.g. commonly available versions of
1557 GCC) is extremely inefficient and 2) the C++ standards committee is likely to
1558 deprecate this container and/or change it significantly somehow. In any case,
1559 please don't use it.</p>
1562 <!-- _______________________________________________________________________ -->
1563 <div class="doc_subsubsection">
1564 <a name="dss_bitvector">BitVector</a>
1567 <div class="doc_text">
1568 <p> The BitVector container provides a fixed size set of bits for manipulation.
1569 It supports individual bit setting/testing, as well as set operations. The set
1570 operations take time O(size of bitvector), but operations are performed one word
1571 at a time, instead of one bit at a time. This makes the BitVector very fast for
1572 set operations compared to other containers. Use the BitVector when you expect
1573 the number of set bits to be high (IE a dense set).
1577 <!-- _______________________________________________________________________ -->
1578 <div class="doc_subsubsection">
1579 <a name="dss_sparsebitvector">SparseBitVector</a>
1582 <div class="doc_text">
1583 <p> The SparseBitVector container is much like BitVector, with one major
1584 difference: Only the bits that are set, are stored. This makes the
1585 SparseBitVector much more space efficient than BitVector when the set is sparse,
1586 as well as making set operations O(number of set bits) instead of O(size of
1587 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
1588 (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).
1592 <!-- *********************************************************************** -->
1593 <div class="doc_section">
1594 <a name="common">Helpful Hints for Common Operations</a>
1596 <!-- *********************************************************************** -->
1598 <div class="doc_text">
1600 <p>This section describes how to perform some very simple transformations of
1601 LLVM code. This is meant to give examples of common idioms used, showing the
1602 practical side of LLVM transformations. <p> Because this is a "how-to" section,
1603 you should also read about the main classes that you will be working with. The
1604 <a href="#coreclasses">Core LLVM Class Hierarchy Reference</a> contains details
1605 and descriptions of the main classes that you should know about.</p>
1609 <!-- NOTE: this section should be heavy on example code -->
1610 <!-- ======================================================================= -->
1611 <div class="doc_subsection">
1612 <a name="inspection">Basic Inspection and Traversal Routines</a>
1615 <div class="doc_text">
1617 <p>The LLVM compiler infrastructure have many different data structures that may
1618 be traversed. Following the example of the C++ standard template library, the
1619 techniques used to traverse these various data structures are all basically the
1620 same. For a enumerable sequence of values, the <tt>XXXbegin()</tt> function (or
1621 method) returns an iterator to the start of the sequence, the <tt>XXXend()</tt>
1622 function returns an iterator pointing to one past the last valid element of the
1623 sequence, and there is some <tt>XXXiterator</tt> data type that is common
1624 between the two operations.</p>
1626 <p>Because the pattern for iteration is common across many different aspects of
1627 the program representation, the standard template library algorithms may be used
1628 on them, and it is easier to remember how to iterate. First we show a few common
1629 examples of the data structures that need to be traversed. Other data
1630 structures are traversed in very similar ways.</p>
1634 <!-- _______________________________________________________________________ -->
1635 <div class="doc_subsubsection">
1636 <a name="iterate_function">Iterating over the </a><a
1637 href="#BasicBlock"><tt>BasicBlock</tt></a>s in a <a
1638 href="#Function"><tt>Function</tt></a>
1641 <div class="doc_text">
1643 <p>It's quite common to have a <tt>Function</tt> instance that you'd like to
1644 transform in some way; in particular, you'd like to manipulate its
1645 <tt>BasicBlock</tt>s. To facilitate this, you'll need to iterate over all of
1646 the <tt>BasicBlock</tt>s that constitute the <tt>Function</tt>. The following is
1647 an example that prints the name of a <tt>BasicBlock</tt> and the number of
1648 <tt>Instruction</tt>s it contains:</p>
1650 <div class="doc_code">
1652 // <i>func is a pointer to a Function instance</i>
1653 for (Function::iterator i = func->begin(), e = func->end(); i != e; ++i)
1654 // <i>Print out the name of the basic block if it has one, and then the</i>
1655 // <i>number of instructions that it contains</i>
1656 llvm::cerr << "Basic block (name=" << i->getName() << ") has "
1657 << i->size() << " instructions.\n";
1661 <p>Note that i can be used as if it were a pointer for the purposes of
1662 invoking member functions of the <tt>Instruction</tt> class. This is
1663 because the indirection operator is overloaded for the iterator
1664 classes. In the above code, the expression <tt>i->size()</tt> is
1665 exactly equivalent to <tt>(*i).size()</tt> just like you'd expect.</p>
1669 <!-- _______________________________________________________________________ -->
1670 <div class="doc_subsubsection">
1671 <a name="iterate_basicblock">Iterating over the </a><a
1672 href="#Instruction"><tt>Instruction</tt></a>s in a <a
1673 href="#BasicBlock"><tt>BasicBlock</tt></a>
1676 <div class="doc_text">
1678 <p>Just like when dealing with <tt>BasicBlock</tt>s in <tt>Function</tt>s, it's
1679 easy to iterate over the individual instructions that make up
1680 <tt>BasicBlock</tt>s. Here's a code snippet that prints out each instruction in
1681 a <tt>BasicBlock</tt>:</p>
1683 <div class="doc_code">
1685 // <i>blk is a pointer to a BasicBlock instance</i>
1686 for (BasicBlock::iterator i = blk->begin(), e = blk->end(); i != e; ++i)
1687 // <i>The next statement works since operator<<(ostream&,...)</i>
1688 // <i>is overloaded for Instruction&</i>
1689 llvm::cerr << *i << "\n";
1693 <p>However, this isn't really the best way to print out the contents of a
1694 <tt>BasicBlock</tt>! Since the ostream operators are overloaded for virtually
1695 anything you'll care about, you could have just invoked the print routine on the
1696 basic block itself: <tt>llvm::cerr << *blk << "\n";</tt>.</p>
1700 <!-- _______________________________________________________________________ -->
1701 <div class="doc_subsubsection">
1702 <a name="iterate_institer">Iterating over the </a><a
1703 href="#Instruction"><tt>Instruction</tt></a>s in a <a
1704 href="#Function"><tt>Function</tt></a>
1707 <div class="doc_text">
1709 <p>If you're finding that you commonly iterate over a <tt>Function</tt>'s
1710 <tt>BasicBlock</tt>s and then that <tt>BasicBlock</tt>'s <tt>Instruction</tt>s,
1711 <tt>InstIterator</tt> should be used instead. You'll need to include <a
1712 href="/doxygen/InstIterator_8h-source.html"><tt>llvm/Support/InstIterator.h</tt></a>,
1713 and then instantiate <tt>InstIterator</tt>s explicitly in your code. Here's a
1714 small example that shows how to dump all instructions in a function to the standard error stream:<p>
1716 <div class="doc_code">
1718 #include "<a href="/doxygen/InstIterator_8h-source.html">llvm/Support/InstIterator.h</a>"
1720 // <i>F is a pointer to a Function instance</i>
1721 for (inst_iterator I = inst_begin(F), E = inst_end(F); I != E; ++I)
1722 llvm::cerr << *I << "\n";
1726 <p>Easy, isn't it? You can also use <tt>InstIterator</tt>s to fill a
1727 work list with its initial contents. For example, if you wanted to
1728 initialize a work list to contain all instructions in a <tt>Function</tt>
1729 F, all you would need to do is something like:</p>
1731 <div class="doc_code">
1733 std::set<Instruction*> worklist;
1734 // or better yet, SmallPtrSet<Instruction*, 64> worklist;
1736 for (inst_iterator I = inst_begin(F), E = inst_end(F); I != E; ++I)
1737 worklist.insert(&*I);
1741 <p>The STL set <tt>worklist</tt> would now contain all instructions in the
1742 <tt>Function</tt> pointed to by F.</p>
1746 <!-- _______________________________________________________________________ -->
1747 <div class="doc_subsubsection">
1748 <a name="iterate_convert">Turning an iterator into a class pointer (and
1752 <div class="doc_text">
1754 <p>Sometimes, it'll be useful to grab a reference (or pointer) to a class
1755 instance when all you've got at hand is an iterator. Well, extracting
1756 a reference or a pointer from an iterator is very straight-forward.
1757 Assuming that <tt>i</tt> is a <tt>BasicBlock::iterator</tt> and <tt>j</tt>
1758 is a <tt>BasicBlock::const_iterator</tt>:</p>
1760 <div class="doc_code">
1762 Instruction& inst = *i; // <i>Grab reference to instruction reference</i>
1763 Instruction* pinst = &*i; // <i>Grab pointer to instruction reference</i>
1764 const Instruction& inst = *j;
1768 <p>However, the iterators you'll be working with in the LLVM framework are
1769 special: they will automatically convert to a ptr-to-instance type whenever they
1770 need to. Instead of dereferencing the iterator and then taking the address of
1771 the result, you can simply assign the iterator to the proper pointer type and
1772 you get the dereference and address-of operation as a result of the assignment
1773 (behind the scenes, this is a result of overloading casting mechanisms). Thus
1774 the last line of the last example,</p>
1776 <div class="doc_code">
1778 Instruction *pinst = &*i;
1782 <p>is semantically equivalent to</p>
1784 <div class="doc_code">
1786 Instruction *pinst = i;
1790 <p>It's also possible to turn a class pointer into the corresponding iterator,
1791 and this is a constant time operation (very efficient). The following code
1792 snippet illustrates use of the conversion constructors provided by LLVM
1793 iterators. By using these, you can explicitly grab the iterator of something
1794 without actually obtaining it via iteration over some structure:</p>
1796 <div class="doc_code">
1798 void printNextInstruction(Instruction* inst) {
1799 BasicBlock::iterator it(inst);
1800 ++it; // <i>After this line, it refers to the instruction after *inst</i>
1801 if (it != inst->getParent()->end()) llvm::cerr << *it << "\n";
1808 <!--_______________________________________________________________________-->
1809 <div class="doc_subsubsection">
1810 <a name="iterate_complex">Finding call sites: a slightly more complex
1814 <div class="doc_text">
1816 <p>Say that you're writing a FunctionPass and would like to count all the
1817 locations in the entire module (that is, across every <tt>Function</tt>) where a
1818 certain function (i.e., some <tt>Function</tt>*) is already in scope. As you'll
1819 learn later, you may want to use an <tt>InstVisitor</tt> to accomplish this in a
1820 much more straight-forward manner, but this example will allow us to explore how
1821 you'd do it if you didn't have <tt>InstVisitor</tt> around. In pseudo-code, this
1822 is what we want to do:</p>
1824 <div class="doc_code">
1826 initialize callCounter to zero
1827 for each Function f in the Module
1828 for each BasicBlock b in f
1829 for each Instruction i in b
1830 if (i is a CallInst and calls the given function)
1831 increment callCounter
1835 <p>And the actual code is (remember, because we're writing a
1836 <tt>FunctionPass</tt>, our <tt>FunctionPass</tt>-derived class simply has to
1837 override the <tt>runOnFunction</tt> method):</p>
1839 <div class="doc_code">
1841 Function* targetFunc = ...;
1843 class OurFunctionPass : public FunctionPass {
1845 OurFunctionPass(): callCounter(0) { }
1847 virtual runOnFunction(Function& F) {
1848 for (Function::iterator b = F.begin(), be = F.end(); b != be; ++b) {
1849 for (BasicBlock::iterator i = b->begin(), ie = b->end(); i != ie; ++i) {
1850 if (<a href="#CallInst">CallInst</a>* callInst = <a href="#isa">dyn_cast</a><<a
1851 href="#CallInst">CallInst</a>>(&*i)) {
1852 // <i>We know we've encountered a call instruction, so we</i>
1853 // <i>need to determine if it's a call to the</i>
1854 // <i>function pointed to by m_func or not.</i>
1855 if (callInst->getCalledFunction() == targetFunc)
1863 unsigned callCounter;
1870 <!--_______________________________________________________________________-->
1871 <div class="doc_subsubsection">
1872 <a name="calls_and_invokes">Treating calls and invokes the same way</a>
1875 <div class="doc_text">
1877 <p>You may have noticed that the previous example was a bit oversimplified in
1878 that it did not deal with call sites generated by 'invoke' instructions. In
1879 this, and in other situations, you may find that you want to treat
1880 <tt>CallInst</tt>s and <tt>InvokeInst</tt>s the same way, even though their
1881 most-specific common base class is <tt>Instruction</tt>, which includes lots of
1882 less closely-related things. For these cases, LLVM provides a handy wrapper
1884 href="http://llvm.org/doxygen/classllvm_1_1CallSite.html"><tt>CallSite</tt></a>.
1885 It is essentially a wrapper around an <tt>Instruction</tt> pointer, with some
1886 methods that provide functionality common to <tt>CallInst</tt>s and
1887 <tt>InvokeInst</tt>s.</p>
1889 <p>This class has "value semantics": it should be passed by value, not by
1890 reference and it should not be dynamically allocated or deallocated using
1891 <tt>operator new</tt> or <tt>operator delete</tt>. It is efficiently copyable,
1892 assignable and constructable, with costs equivalents to that of a bare pointer.
1893 If you look at its definition, it has only a single pointer member.</p>
1897 <!--_______________________________________________________________________-->
1898 <div class="doc_subsubsection">
1899 <a name="iterate_chains">Iterating over def-use & use-def chains</a>
1902 <div class="doc_text">
1904 <p>Frequently, we might have an instance of the <a
1905 href="/doxygen/classllvm_1_1Value.html">Value Class</a> and we want to
1906 determine which <tt>User</tt>s use the <tt>Value</tt>. The list of all
1907 <tt>User</tt>s of a particular <tt>Value</tt> is called a <i>def-use</i> chain.
1908 For example, let's say we have a <tt>Function*</tt> named <tt>F</tt> to a
1909 particular function <tt>foo</tt>. Finding all of the instructions that
1910 <i>use</i> <tt>foo</tt> is as simple as iterating over the <i>def-use</i> chain
1913 <div class="doc_code">
1917 for (Value::use_iterator i = F->use_begin(), e = F->use_end(); i != e; ++i)
1918 if (Instruction *Inst = dyn_cast<Instruction>(*i)) {
1919 llvm::cerr << "F is used in instruction:\n";
1920 llvm::cerr << *Inst << "\n";
1925 <p>Alternately, it's common to have an instance of the <a
1926 href="/doxygen/classllvm_1_1User.html">User Class</a> and need to know what
1927 <tt>Value</tt>s are used by it. The list of all <tt>Value</tt>s used by a
1928 <tt>User</tt> is known as a <i>use-def</i> chain. Instances of class
1929 <tt>Instruction</tt> are common <tt>User</tt>s, so we might want to iterate over
1930 all of the values that a particular instruction uses (that is, the operands of
1931 the particular <tt>Instruction</tt>):</p>
1933 <div class="doc_code">
1935 Instruction *pi = ...;
1937 for (User::op_iterator i = pi->op_begin(), e = pi->op_end(); i != e; ++i) {
1945 def-use chains ("finding all users of"): Value::use_begin/use_end
1946 use-def chains ("finding all values used"): User::op_begin/op_end [op=operand]
1951 <!--_______________________________________________________________________-->
1952 <div class="doc_subsubsection">
1953 <a name="iterate_preds">Iterating over predecessors &
1954 successors of blocks</a>
1957 <div class="doc_text">
1959 <p>Iterating over the predecessors and successors of a block is quite easy
1960 with the routines defined in <tt>"llvm/Support/CFG.h"</tt>. Just use code like
1961 this to iterate over all predecessors of BB:</p>
1963 <div class="doc_code">
1965 #include "llvm/Support/CFG.h"
1966 BasicBlock *BB = ...;
1968 for (pred_iterator PI = pred_begin(BB), E = pred_end(BB); PI != E; ++PI) {
1969 BasicBlock *Pred = *PI;
1975 <p>Similarly, to iterate over successors use
1976 succ_iterator/succ_begin/succ_end.</p>
1981 <!-- ======================================================================= -->
1982 <div class="doc_subsection">
1983 <a name="simplechanges">Making simple changes</a>
1986 <div class="doc_text">
1988 <p>There are some primitive transformation operations present in the LLVM
1989 infrastructure that are worth knowing about. When performing
1990 transformations, it's fairly common to manipulate the contents of basic
1991 blocks. This section describes some of the common methods for doing so
1992 and gives example code.</p>
1996 <!--_______________________________________________________________________-->
1997 <div class="doc_subsubsection">
1998 <a name="schanges_creating">Creating and inserting new
1999 <tt>Instruction</tt>s</a>
2002 <div class="doc_text">
2004 <p><i>Instantiating Instructions</i></p>
2006 <p>Creation of <tt>Instruction</tt>s is straight-forward: simply call the
2007 constructor for the kind of instruction to instantiate and provide the necessary
2008 parameters. For example, an <tt>AllocaInst</tt> only <i>requires</i> a
2009 (const-ptr-to) <tt>Type</tt>. Thus:</p>
2011 <div class="doc_code">
2013 AllocaInst* ai = new AllocaInst(Type::Int32Ty);
2017 <p>will create an <tt>AllocaInst</tt> instance that represents the allocation of
2018 one integer in the current stack frame, at run time. Each <tt>Instruction</tt>
2019 subclass is likely to have varying default parameters which change the semantics
2020 of the instruction, so refer to the <a
2021 href="/doxygen/classllvm_1_1Instruction.html">doxygen documentation for the subclass of
2022 Instruction</a> that you're interested in instantiating.</p>
2024 <p><i>Naming values</i></p>
2026 <p>It is very useful to name the values of instructions when you're able to, as
2027 this facilitates the debugging of your transformations. If you end up looking
2028 at generated LLVM machine code, you definitely want to have logical names
2029 associated with the results of instructions! By supplying a value for the
2030 <tt>Name</tt> (default) parameter of the <tt>Instruction</tt> constructor, you
2031 associate a logical name with the result of the instruction's execution at
2032 run time. For example, say that I'm writing a transformation that dynamically
2033 allocates space for an integer on the stack, and that integer is going to be
2034 used as some kind of index by some other code. To accomplish this, I place an
2035 <tt>AllocaInst</tt> at the first point in the first <tt>BasicBlock</tt> of some
2036 <tt>Function</tt>, and I'm intending to use it within the same
2037 <tt>Function</tt>. I might do:</p>
2039 <div class="doc_code">
2041 AllocaInst* pa = new AllocaInst(Type::Int32Ty, 0, "indexLoc");
2045 <p>where <tt>indexLoc</tt> is now the logical name of the instruction's
2046 execution value, which is a pointer to an integer on the run time stack.</p>
2048 <p><i>Inserting instructions</i></p>
2050 <p>There are essentially two ways to insert an <tt>Instruction</tt>
2051 into an existing sequence of instructions that form a <tt>BasicBlock</tt>:</p>
2054 <li>Insertion into an explicit instruction list
2056 <p>Given a <tt>BasicBlock* pb</tt>, an <tt>Instruction* pi</tt> within that
2057 <tt>BasicBlock</tt>, and a newly-created instruction we wish to insert
2058 before <tt>*pi</tt>, we do the following: </p>
2060 <div class="doc_code">
2062 BasicBlock *pb = ...;
2063 Instruction *pi = ...;
2064 Instruction *newInst = new Instruction(...);
2066 pb->getInstList().insert(pi, newInst); // <i>Inserts newInst before pi in pb</i>
2070 <p>Appending to the end of a <tt>BasicBlock</tt> is so common that
2071 the <tt>Instruction</tt> class and <tt>Instruction</tt>-derived
2072 classes provide constructors which take a pointer to a
2073 <tt>BasicBlock</tt> to be appended to. For example code that
2076 <div class="doc_code">
2078 BasicBlock *pb = ...;
2079 Instruction *newInst = new Instruction(...);
2081 pb->getInstList().push_back(newInst); // <i>Appends newInst to pb</i>
2087 <div class="doc_code">
2089 BasicBlock *pb = ...;
2090 Instruction *newInst = new Instruction(..., pb);
2094 <p>which is much cleaner, especially if you are creating
2095 long instruction streams.</p></li>
2097 <li>Insertion into an implicit instruction list
2099 <p><tt>Instruction</tt> instances that are already in <tt>BasicBlock</tt>s
2100 are implicitly associated with an existing instruction list: the instruction
2101 list of the enclosing basic block. Thus, we could have accomplished the same
2102 thing as the above code without being given a <tt>BasicBlock</tt> by doing:
2105 <div class="doc_code">
2107 Instruction *pi = ...;
2108 Instruction *newInst = new Instruction(...);
2110 pi->getParent()->getInstList().insert(pi, newInst);
2114 <p>In fact, this sequence of steps occurs so frequently that the
2115 <tt>Instruction</tt> class and <tt>Instruction</tt>-derived classes provide
2116 constructors which take (as a default parameter) a pointer to an
2117 <tt>Instruction</tt> which the newly-created <tt>Instruction</tt> should
2118 precede. That is, <tt>Instruction</tt> constructors are capable of
2119 inserting the newly-created instance into the <tt>BasicBlock</tt> of a
2120 provided instruction, immediately before that instruction. Using an
2121 <tt>Instruction</tt> constructor with a <tt>insertBefore</tt> (default)
2122 parameter, the above code becomes:</p>
2124 <div class="doc_code">
2126 Instruction* pi = ...;
2127 Instruction* newInst = new Instruction(..., pi);
2131 <p>which is much cleaner, especially if you're creating a lot of
2132 instructions and adding them to <tt>BasicBlock</tt>s.</p></li>
2137 <!--_______________________________________________________________________-->
2138 <div class="doc_subsubsection">
2139 <a name="schanges_deleting">Deleting <tt>Instruction</tt>s</a>
2142 <div class="doc_text">
2144 <p>Deleting an instruction from an existing sequence of instructions that form a
2145 <a href="#BasicBlock"><tt>BasicBlock</tt></a> is very straight-forward. First,
2146 you must have a pointer to the instruction that you wish to delete. Second, you
2147 need to obtain the pointer to that instruction's basic block. You use the
2148 pointer to the basic block to get its list of instructions and then use the
2149 erase function to remove your instruction. For example:</p>
2151 <div class="doc_code">
2153 <a href="#Instruction">Instruction</a> *I = .. ;
2154 I->eraseFromParent();
2160 <!--_______________________________________________________________________-->
2161 <div class="doc_subsubsection">
2162 <a name="schanges_replacing">Replacing an <tt>Instruction</tt> with another
2166 <div class="doc_text">
2168 <p><i>Replacing individual instructions</i></p>
2170 <p>Including "<a href="/doxygen/BasicBlockUtils_8h-source.html">llvm/Transforms/Utils/BasicBlockUtils.h</a>"
2171 permits use of two very useful replace functions: <tt>ReplaceInstWithValue</tt>
2172 and <tt>ReplaceInstWithInst</tt>.</p>
2174 <h4><a name="schanges_deleting">Deleting <tt>Instruction</tt>s</a></h4>
2177 <li><tt>ReplaceInstWithValue</tt>
2179 <p>This function replaces all uses of a given instruction with a value,
2180 and then removes the original instruction. The following example
2181 illustrates the replacement of the result of a particular
2182 <tt>AllocaInst</tt> that allocates memory for a single integer with a null
2183 pointer to an integer.</p>
2185 <div class="doc_code">
2187 AllocaInst* instToReplace = ...;
2188 BasicBlock::iterator ii(instToReplace);
2190 ReplaceInstWithValue(instToReplace->getParent()->getInstList(), ii,
2191 Constant::getNullValue(PointerType::getUnqual(Type::Int32Ty)));
2194 <li><tt>ReplaceInstWithInst</tt>
2196 <p>This function replaces a particular instruction with another
2197 instruction, inserting the new instruction into the basic block at the
2198 location where the old instruction was, and replacing any uses of the old
2199 instruction with the new instruction. The following example illustrates
2200 the replacement of one <tt>AllocaInst</tt> with another.</p>
2202 <div class="doc_code">
2204 AllocaInst* instToReplace = ...;
2205 BasicBlock::iterator ii(instToReplace);
2207 ReplaceInstWithInst(instToReplace->getParent()->getInstList(), ii,
2208 new AllocaInst(Type::Int32Ty, 0, "ptrToReplacedInt"));
2212 <p><i>Replacing multiple uses of <tt>User</tt>s and <tt>Value</tt>s</i></p>
2214 <p>You can use <tt>Value::replaceAllUsesWith</tt> and
2215 <tt>User::replaceUsesOfWith</tt> to change more than one use at a time. See the
2216 doxygen documentation for the <a href="/doxygen/classllvm_1_1Value.html">Value Class</a>
2217 and <a href="/doxygen/classllvm_1_1User.html">User Class</a>, respectively, for more
2220 <!-- Value::replaceAllUsesWith User::replaceUsesOfWith Point out:
2221 include/llvm/Transforms/Utils/ especially BasicBlockUtils.h with:
2222 ReplaceInstWithValue, ReplaceInstWithInst -->
2226 <!--_______________________________________________________________________-->
2227 <div class="doc_subsubsection">
2228 <a name="schanges_deletingGV">Deleting <tt>GlobalVariable</tt>s</a>
2231 <div class="doc_text">
2233 <p>Deleting a global variable from a module is just as easy as deleting an
2234 Instruction. First, you must have a pointer to the global variable that you wish
2235 to delete. You use this pointer to erase it from its parent, the module.
2238 <div class="doc_code">
2240 <a href="#GlobalVariable">GlobalVariable</a> *GV = .. ;
2242 GV->eraseFromParent();
2248 <!-- ======================================================================= -->
2249 <div class="doc_subsection">
2250 <a name="create_types">How to Create Types</a>
2253 <div class="doc_text">
2255 <p>In generating IR, you may need some complex types. If you know these types
2256 statically, you can use <tt>TypeBuilder<...>::get()</tt>, defined
2257 in <tt>llvm/Support/TypeBuilder.h</tt>, to retrieve them. <tt>TypeBuilder</tt>
2258 has two forms depending on whether you're building types for cross-compilation
2259 or native library use. <tt>TypeBuilder<T, true></tt> requires
2260 that <tt>T</tt> be independent of the host environment, meaning that it's built
2262 the <a href="/doxygen/namespacellvm_1_1types.html"><tt>llvm::types</tt></a>
2263 namespace and pointers, functions, arrays, etc. built of
2264 those. <tt>TypeBuilder<T, false></tt> additionally allows native C types
2265 whose size may depend on the host compiler. For example,</p>
2267 <div class="doc_code">
2269 FunctionType *ft = TypeBuilder<types::i<8>(types::i<32>*), true>::get();
2273 <p>is easier to read and write than the equivalent</p>
2275 <div class="doc_code">
2277 std::vector<const Type*> params;
2278 params.push_back(PointerType::getUnqual(Type::Int32Ty));
2279 FunctionType *ft = FunctionType::get(Type::Int8Ty, params, false);
2283 <p>See the <a href="/doxygen/TypeBuilder_8h-source.html#l00001">class
2284 comment</a> for more details.</p>
2288 <!-- *********************************************************************** -->
2289 <div class="doc_section">
2290 <a name="threading">Threads and LLVM</a>
2292 <!-- *********************************************************************** -->
2294 <div class="doc_text">
2296 This section describes the interaction of the LLVM APIs with multithreading,
2297 both on the part of client applications, and in the JIT, in the hosted
2302 Note that LLVM's support for multithreading is still relatively young. Up
2303 through version 2.5, the execution of threaded hosted applications was
2304 supported, but not threaded client access to the APIs. While this use case is
2305 now supported, clients <em>must</em> adhere to the guidelines specified below to
2306 ensure proper operation in multithreaded mode.
2310 Note that, on Unix-like platforms, LLVM requires the presence of GCC's atomic
2311 intrinsics in order to support threaded operation. If you need a
2312 multhreading-capable LLVM on a platform without a suitably modern system
2313 compiler, consider compiling LLVM and LLVM-GCC in single-threaded mode, and
2314 using the resultant compiler to build a copy of LLVM with multithreading
2319 <!-- ======================================================================= -->
2320 <div class="doc_subsection">
2321 <a name="startmultithreaded">Entering and Exiting Multithreaded Mode</a>
2324 <div class="doc_text">
2327 In order to properly protect its internal data structures while avoiding
2328 excessive locking overhead in the single-threaded case, the LLVM must intialize
2329 certain data structures necessary to provide guards around its internals. To do
2330 so, the client program must invoke <tt>llvm_start_multithreaded()</tt> before
2331 making any concurrent LLVM API calls. To subsequently tear down these
2332 structures, use the <tt>llvm_stop_multithreaded()</tt> call. You can also use
2333 the <tt>llvm_is_multithreaded()</tt> call to check the status of multithreaded
2338 Note that both of these calls must be made <em>in isolation</em>. That is to
2339 say that no other LLVM API calls may be executing at any time during the
2340 execution of <tt>llvm_start_multithreaded()</tt> or <tt>llvm_stop_multithreaded
2341 </tt>. It's is the client's responsibility to enforce this isolation.
2345 The return value of <tt>llvm_start_multithreaded()</tt> indicates the success or
2346 failure of the initialization. Failure typically indicates that your copy of
2347 LLVM was built without multithreading support, typically because GCC atomic
2348 intrinsics were not found in your system compiler. In this case, the LLVM API
2349 will not be safe for concurrent calls. However, it <em>will</em> be safe for
2350 hosting threaded applications in the JIT, though care must be taken to ensure
2351 that side exits and the like do not accidentally result in concurrent LLVM API
2356 <!-- ======================================================================= -->
2357 <div class="doc_subsection">
2358 <a name="shutdown">Ending Execution with <tt>llvm_shutdown()</tt></a>
2361 <div class="doc_text">
2363 When you are done using the LLVM APIs, you should call <tt>llvm_shutdown()</tt>
2364 to deallocate memory used for internal structures. This will also invoke
2365 <tt>llvm_stop_multithreaded()</tt> if LLVM is operating in multithreaded mode.
2366 As such, <tt>llvm_shutdown()</tt> requires the same isolation guarantees as
2367 <tt>llvm_stop_multithreaded()</tt>.
2371 Note that, if you use scope-based shutdown, you can use the
2372 <tt>llvm_shutdown_obj</tt> class, which calls <tt>llvm_shutdown()</tt> in its
2376 <!-- ======================================================================= -->
2377 <div class="doc_subsection">
2378 <a name="managedstatic">Lazy Initialization with <tt>ManagedStatic</tt></a>
2381 <div class="doc_text">
2383 <tt>ManagedStatic</tt> is a utility class in LLVM used to implement static
2384 initialization of static resources, such as the global type tables. Before the
2385 invocation of <tt>llvm_shutdown()</tt>, it implements a simple lazy
2386 initialization scheme. Once <tt>llvm_start_multithreaded()</tt> returns,
2387 however, it uses double-checked locking to implement thread-safe lazy
2392 Note that, because no other threads are allowed to issue LLVM API calls before
2393 <tt>llvm_start_multithreaded()</tt> returns, it is possible to have
2394 <tt>ManagedStatic</tt>s of <tt>llvm::sys::Mutex</tt>s.
2398 The <tt>llvm_acquire_global_lock()</tt> and <tt>llvm_release_global_lock</tt>
2399 APIs provide access to the global lock used to implement the double-checked
2400 locking for lazy initialization. These should only be used internally to LLVM,
2401 and only if you know what you're doing!
2405 <!-- *********************************************************************** -->
2406 <div class="doc_section">
2407 <a name="advanced">Advanced Topics</a>
2409 <!-- *********************************************************************** -->
2411 <div class="doc_text">
2413 This section describes some of the advanced or obscure API's that most clients
2414 do not need to be aware of. These API's tend manage the inner workings of the
2415 LLVM system, and only need to be accessed in unusual circumstances.
2419 <!-- ======================================================================= -->
2420 <div class="doc_subsection">
2421 <a name="TypeResolve">LLVM Type Resolution</a>
2424 <div class="doc_text">
2427 The LLVM type system has a very simple goal: allow clients to compare types for
2428 structural equality with a simple pointer comparison (aka a shallow compare).
2429 This goal makes clients much simpler and faster, and is used throughout the LLVM
2434 Unfortunately achieving this goal is not a simple matter. In particular,
2435 recursive types and late resolution of opaque types makes the situation very
2436 difficult to handle. Fortunately, for the most part, our implementation makes
2437 most clients able to be completely unaware of the nasty internal details. The
2438 primary case where clients are exposed to the inner workings of it are when
2439 building a recursive type. In addition to this case, the LLVM bitcode reader,
2440 assembly parser, and linker also have to be aware of the inner workings of this
2445 For our purposes below, we need three concepts. First, an "Opaque Type" is
2446 exactly as defined in the <a href="LangRef.html#t_opaque">language
2447 reference</a>. Second an "Abstract Type" is any type which includes an
2448 opaque type as part of its type graph (for example "<tt>{ opaque, i32 }</tt>").
2449 Third, a concrete type is a type that is not an abstract type (e.g. "<tt>{ i32,
2455 <!-- ______________________________________________________________________ -->
2456 <div class="doc_subsubsection">
2457 <a name="BuildRecType">Basic Recursive Type Construction</a>
2460 <div class="doc_text">
2463 Because the most common question is "how do I build a recursive type with LLVM",
2464 we answer it now and explain it as we go. Here we include enough to cause this
2465 to be emitted to an output .ll file:
2468 <div class="doc_code">
2470 %mylist = type { %mylist*, i32 }
2475 To build this, use the following LLVM APIs:
2478 <div class="doc_code">
2480 // <i>Create the initial outer struct</i>
2481 <a href="#PATypeHolder">PATypeHolder</a> StructTy = OpaqueType::get();
2482 std::vector<const Type*> Elts;
2483 Elts.push_back(PointerType::getUnqual(StructTy));
2484 Elts.push_back(Type::Int32Ty);
2485 StructType *NewSTy = StructType::get(Elts);
2487 // <i>At this point, NewSTy = "{ opaque*, i32 }". Tell VMCore that</i>
2488 // <i>the struct and the opaque type are actually the same.</i>
2489 cast<OpaqueType>(StructTy.get())-><a href="#refineAbstractTypeTo">refineAbstractTypeTo</a>(NewSTy);
2491 // <i>NewSTy is potentially invalidated, but StructTy (a <a href="#PATypeHolder">PATypeHolder</a>) is</i>
2492 // <i>kept up-to-date</i>
2493 NewSTy = cast<StructType>(StructTy.get());
2495 // <i>Add a name for the type to the module symbol table (optional)</i>
2496 MyModule->addTypeName("mylist", NewSTy);
2501 This code shows the basic approach used to build recursive types: build a
2502 non-recursive type using 'opaque', then use type unification to close the cycle.
2503 The type unification step is performed by the <tt><a
2504 href="#refineAbstractTypeTo">refineAbstractTypeTo</a></tt> method, which is
2505 described next. After that, we describe the <a
2506 href="#PATypeHolder">PATypeHolder class</a>.
2511 <!-- ______________________________________________________________________ -->
2512 <div class="doc_subsubsection">
2513 <a name="refineAbstractTypeTo">The <tt>refineAbstractTypeTo</tt> method</a>
2516 <div class="doc_text">
2518 The <tt>refineAbstractTypeTo</tt> method starts the type unification process.
2519 While this method is actually a member of the DerivedType class, it is most
2520 often used on OpaqueType instances. Type unification is actually a recursive
2521 process. After unification, types can become structurally isomorphic to
2522 existing types, and all duplicates are deleted (to preserve pointer equality).
2526 In the example above, the OpaqueType object is definitely deleted.
2527 Additionally, if there is an "{ \2*, i32}" type already created in the system,
2528 the pointer and struct type created are <b>also</b> deleted. Obviously whenever
2529 a type is deleted, any "Type*" pointers in the program are invalidated. As
2530 such, it is safest to avoid having <i>any</i> "Type*" pointers to abstract types
2531 live across a call to <tt>refineAbstractTypeTo</tt> (note that non-abstract
2532 types can never move or be deleted). To deal with this, the <a
2533 href="#PATypeHolder">PATypeHolder</a> class is used to maintain a stable
2534 reference to a possibly refined type, and the <a
2535 href="#AbstractTypeUser">AbstractTypeUser</a> class is used to update more
2536 complex datastructures.
2541 <!-- ______________________________________________________________________ -->
2542 <div class="doc_subsubsection">
2543 <a name="PATypeHolder">The PATypeHolder Class</a>
2546 <div class="doc_text">
2548 PATypeHolder is a form of a "smart pointer" for Type objects. When VMCore
2549 happily goes about nuking types that become isomorphic to existing types, it
2550 automatically updates all PATypeHolder objects to point to the new type. In the
2551 example above, this allows the code to maintain a pointer to the resultant
2552 resolved recursive type, even though the Type*'s are potentially invalidated.
2556 PATypeHolder is an extremely light-weight object that uses a lazy union-find
2557 implementation to update pointers. For example the pointer from a Value to its
2558 Type is maintained by PATypeHolder objects.
2563 <!-- ______________________________________________________________________ -->
2564 <div class="doc_subsubsection">
2565 <a name="AbstractTypeUser">The AbstractTypeUser Class</a>
2568 <div class="doc_text">
2571 Some data structures need more to perform more complex updates when types get
2572 resolved. To support this, a class can derive from the AbstractTypeUser class.
2574 allows it to get callbacks when certain types are resolved. To register to get
2575 callbacks for a particular type, the DerivedType::{add/remove}AbstractTypeUser
2576 methods can be called on a type. Note that these methods only work for <i>
2577 abstract</i> types. Concrete types (those that do not include any opaque
2578 objects) can never be refined.
2583 <!-- ======================================================================= -->
2584 <div class="doc_subsection">
2585 <a name="SymbolTable">The <tt>ValueSymbolTable</tt> and
2586 <tt>TypeSymbolTable</tt> classes</a>
2589 <div class="doc_text">
2590 <p>The <tt><a href="http://llvm.org/doxygen/classllvm_1_1ValueSymbolTable.html">
2591 ValueSymbolTable</a></tt> class provides a symbol table that the <a
2592 href="#Function"><tt>Function</tt></a> and <a href="#Module">
2593 <tt>Module</tt></a> classes use for naming value definitions. The symbol table
2594 can provide a name for any <a href="#Value"><tt>Value</tt></a>.
2595 The <tt><a href="http://llvm.org/doxygen/classllvm_1_1TypeSymbolTable.html">
2596 TypeSymbolTable</a></tt> class is used by the <tt>Module</tt> class to store
2597 names for types.</p>
2599 <p>Note that the <tt>SymbolTable</tt> class should not be directly accessed
2600 by most clients. It should only be used when iteration over the symbol table
2601 names themselves are required, which is very special purpose. Note that not
2603 <tt><a href="#Value">Value</a></tt>s have names, and those without names (i.e. they have
2604 an empty name) do not exist in the symbol table.
2607 <p>These symbol tables support iteration over the values/types in the symbol
2608 table with <tt>begin/end/iterator</tt> and supports querying to see if a
2609 specific name is in the symbol table (with <tt>lookup</tt>). The
2610 <tt>ValueSymbolTable</tt> class exposes no public mutator methods, instead,
2611 simply call <tt>setName</tt> on a value, which will autoinsert it into the
2612 appropriate symbol table. For types, use the Module::addTypeName method to
2613 insert entries into the symbol table.</p>
2619 <!-- ======================================================================= -->
2620 <div class="doc_subsection">
2621 <a name="UserLayout">The <tt>User</tt> and owned <tt>Use</tt> classes' memory layout</a>
2624 <div class="doc_text">
2625 <p>The <tt><a href="http://llvm.org/doxygen/classllvm_1_1User.html">
2626 User</a></tt> class provides a basis for expressing the ownership of <tt>User</tt>
2627 towards other <tt><a href="http://llvm.org/doxygen/classllvm_1_1Value.html">
2628 Value</a></tt>s. The <tt><a href="http://llvm.org/doxygen/classllvm_1_1Use.html">
2629 Use</a></tt> helper class is employed to do the bookkeeping and to facilitate <i>O(1)</i>
2630 addition and removal.</p>
2632 <!-- ______________________________________________________________________ -->
2633 <div class="doc_subsubsection">
2634 <a name="Use2User">Interaction and relationship between <tt>User</tt> and <tt>Use</tt> objects</a>
2637 <div class="doc_text">
2639 A subclass of <tt>User</tt> can choose between incorporating its <tt>Use</tt> objects
2640 or refer to them out-of-line by means of a pointer. A mixed variant
2641 (some <tt>Use</tt>s inline others hung off) is impractical and breaks the invariant
2642 that the <tt>Use</tt> objects belonging to the same <tt>User</tt> form a contiguous array.
2647 We have 2 different layouts in the <tt>User</tt> (sub)classes:
2650 The <tt>Use</tt> object(s) are inside (resp. at fixed offset) of the <tt>User</tt>
2651 object and there are a fixed number of them.</p>
2654 The <tt>Use</tt> object(s) are referenced by a pointer to an
2655 array from the <tt>User</tt> object and there may be a variable
2659 As of v2.4 each layout still possesses a direct pointer to the
2660 start of the array of <tt>Use</tt>s. Though not mandatory for layout a),
2661 we stick to this redundancy for the sake of simplicity.
2662 The <tt>User</tt> object also stores the number of <tt>Use</tt> objects it
2663 has. (Theoretically this information can also be calculated
2664 given the scheme presented below.)</p>
2666 Special forms of allocation operators (<tt>operator new</tt>)
2667 enforce the following memory layouts:</p>
2670 <li><p>Layout a) is modelled by prepending the <tt>User</tt> object by the <tt>Use[]</tt> array.</p>
2673 ...---.---.---.---.-------...
2674 | P | P | P | P | User
2675 '''---'---'---'---'-------'''
2678 <li><p>Layout b) is modelled by pointing at the <tt>Use[]</tt> array.</p>
2690 <i>(In the above figures '<tt>P</tt>' stands for the <tt>Use**</tt> that
2691 is stored in each <tt>Use</tt> object in the member <tt>Use::Prev</tt>)</i>
2693 <!-- ______________________________________________________________________ -->
2694 <div class="doc_subsubsection">
2695 <a name="Waymarking">The waymarking algorithm</a>
2698 <div class="doc_text">
2700 Since the <tt>Use</tt> objects are deprived of the direct (back)pointer to
2701 their <tt>User</tt> objects, there must be a fast and exact method to
2702 recover it. This is accomplished by the following scheme:</p>
2705 A bit-encoding in the 2 LSBits (least significant bits) of the <tt>Use::Prev</tt> allows to find the
2706 start of the <tt>User</tt> object:
2708 <li><tt>00</tt> —> binary digit 0</li>
2709 <li><tt>01</tt> —> binary digit 1</li>
2710 <li><tt>10</tt> —> stop and calculate (<tt>s</tt>)</li>
2711 <li><tt>11</tt> —> full stop (<tt>S</tt>)</li>
2714 Given a <tt>Use*</tt>, all we have to do is to walk till we get
2715 a stop and we either have a <tt>User</tt> immediately behind or
2716 we have to walk to the next stop picking up digits
2717 and calculating the offset:</p>
2719 .---.---.---.---.---.---.---.---.---.---.---.---.---.---.---.---.----------------
2720 | 1 | s | 1 | 0 | 1 | 0 | s | 1 | 1 | 0 | s | 1 | 1 | s | 1 | S | User (or User*)
2721 '---'---'---'---'---'---'---'---'---'---'---'---'---'---'---'---'----------------
2722 |+15 |+10 |+6 |+3 |+1
2725 | | |______________________>
2726 | |______________________________________>
2727 |__________________________________________________________>
2730 Only the significant number of bits need to be stored between the
2731 stops, so that the <i>worst case is 20 memory accesses</i> when there are
2732 1000 <tt>Use</tt> objects associated with a <tt>User</tt>.</p>
2734 <!-- ______________________________________________________________________ -->
2735 <div class="doc_subsubsection">
2736 <a name="ReferenceImpl">Reference implementation</a>
2739 <div class="doc_text">
2741 The following literate Haskell fragment demonstrates the concept:</p>
2744 <div class="doc_code">
2746 > import Test.QuickCheck
2748 > digits :: Int -> [Char] -> [Char]
2749 > digits 0 acc = '0' : acc
2750 > digits 1 acc = '1' : acc
2751 > digits n acc = digits (n `div` 2) $ digits (n `mod` 2) acc
2753 > dist :: Int -> [Char] -> [Char]
2756 > dist 1 acc = let r = dist 0 acc in 's' : digits (length r) r
2757 > dist n acc = dist (n - 1) $ dist 1 acc
2759 > takeLast n ss = reverse $ take n $ reverse ss
2761 > test = takeLast 40 $ dist 20 []
2766 Printing <test> gives: <tt>"1s100000s11010s10100s1111s1010s110s11s1S"</tt></p>
2768 The reverse algorithm computes the length of the string just by examining
2769 a certain prefix:</p>
2771 <div class="doc_code">
2773 > pref :: [Char] -> Int
2775 > pref ('s':'1':rest) = decode 2 1 rest
2776 > pref (_:rest) = 1 + pref rest
2778 > decode walk acc ('0':rest) = decode (walk + 1) (acc * 2) rest
2779 > decode walk acc ('1':rest) = decode (walk + 1) (acc * 2 + 1) rest
2780 > decode walk acc _ = walk + acc
2785 Now, as expected, printing <pref test> gives <tt>40</tt>.</p>
2787 We can <i>quickCheck</i> this with following property:</p>
2789 <div class="doc_code">
2791 > testcase = dist 2000 []
2792 > testcaseLength = length testcase
2794 > identityProp n = n > 0 && n <= testcaseLength ==> length arr == pref arr
2795 > where arr = takeLast n testcase
2800 As expected <quickCheck identityProp> gives:</p>
2803 *Main> quickCheck identityProp
2804 OK, passed 100 tests.
2807 Let's be a bit more exhaustive:</p>
2809 <div class="doc_code">
2812 > deepCheck p = check (defaultConfig { configMaxTest = 500 }) p
2817 And here is the result of <deepCheck identityProp>:</p>
2820 *Main> deepCheck identityProp
2821 OK, passed 500 tests.
2824 <!-- ______________________________________________________________________ -->
2825 <div class="doc_subsubsection">
2826 <a name="Tagging">Tagging considerations</a>
2830 To maintain the invariant that the 2 LSBits of each <tt>Use**</tt> in <tt>Use</tt>
2831 never change after being set up, setters of <tt>Use::Prev</tt> must re-tag the
2832 new <tt>Use**</tt> on every modification. Accordingly getters must strip the
2835 For layout b) instead of the <tt>User</tt> we find a pointer (<tt>User*</tt> with LSBit set).
2836 Following this pointer brings us to the <tt>User</tt>. A portable trick ensures
2837 that the first bytes of <tt>User</tt> (if interpreted as a pointer) never has
2838 the LSBit set. (Portability is relying on the fact that all known compilers place the
2839 <tt>vptr</tt> in the first word of the instances.)</p>
2843 <!-- *********************************************************************** -->
2844 <div class="doc_section">
2845 <a name="coreclasses">The Core LLVM Class Hierarchy Reference </a>
2847 <!-- *********************************************************************** -->
2849 <div class="doc_text">
2850 <p><tt>#include "<a href="/doxygen/Type_8h-source.html">llvm/Type.h</a>"</tt>
2851 <br>doxygen info: <a href="/doxygen/classllvm_1_1Type.html">Type Class</a></p>
2853 <p>The Core LLVM classes are the primary means of representing the program
2854 being inspected or transformed. The core LLVM classes are defined in
2855 header files in the <tt>include/llvm/</tt> directory, and implemented in
2856 the <tt>lib/VMCore</tt> directory.</p>
2860 <!-- ======================================================================= -->
2861 <div class="doc_subsection">
2862 <a name="Type">The <tt>Type</tt> class and Derived Types</a>
2865 <div class="doc_text">
2867 <p><tt>Type</tt> is a superclass of all type classes. Every <tt>Value</tt> has
2868 a <tt>Type</tt>. <tt>Type</tt> cannot be instantiated directly but only
2869 through its subclasses. Certain primitive types (<tt>VoidType</tt>,
2870 <tt>LabelType</tt>, <tt>FloatType</tt> and <tt>DoubleType</tt>) have hidden
2871 subclasses. They are hidden because they offer no useful functionality beyond
2872 what the <tt>Type</tt> class offers except to distinguish themselves from
2873 other subclasses of <tt>Type</tt>.</p>
2874 <p>All other types are subclasses of <tt>DerivedType</tt>. Types can be
2875 named, but this is not a requirement. There exists exactly
2876 one instance of a given shape at any one time. This allows type equality to
2877 be performed with address equality of the Type Instance. That is, given two
2878 <tt>Type*</tt> values, the types are identical if the pointers are identical.
2882 <!-- _______________________________________________________________________ -->
2883 <div class="doc_subsubsection">
2884 <a name="m_Type">Important Public Methods</a>
2887 <div class="doc_text">
2890 <li><tt>bool isInteger() const</tt>: Returns true for any integer type.</li>
2892 <li><tt>bool isFloatingPoint()</tt>: Return true if this is one of the two
2893 floating point types.</li>
2895 <li><tt>bool isAbstract()</tt>: Return true if the type is abstract (contains
2896 an OpaqueType anywhere in its definition).</li>
2898 <li><tt>bool isSized()</tt>: Return true if the type has known size. Things
2899 that don't have a size are abstract types, labels and void.</li>
2904 <!-- _______________________________________________________________________ -->
2905 <div class="doc_subsubsection">
2906 <a name="derivedtypes">Important Derived Types</a>
2908 <div class="doc_text">
2910 <dt><tt>IntegerType</tt></dt>
2911 <dd>Subclass of DerivedType that represents integer types of any bit width.
2912 Any bit width between <tt>IntegerType::MIN_INT_BITS</tt> (1) and
2913 <tt>IntegerType::MAX_INT_BITS</tt> (~8 million) can be represented.
2915 <li><tt>static const IntegerType* get(unsigned NumBits)</tt>: get an integer
2916 type of a specific bit width.</li>
2917 <li><tt>unsigned getBitWidth() const</tt>: Get the bit width of an integer
2921 <dt><tt>SequentialType</tt></dt>
2922 <dd>This is subclassed by ArrayType and PointerType
2924 <li><tt>const Type * getElementType() const</tt>: Returns the type of each
2925 of the elements in the sequential type. </li>
2928 <dt><tt>ArrayType</tt></dt>
2929 <dd>This is a subclass of SequentialType and defines the interface for array
2932 <li><tt>unsigned getNumElements() const</tt>: Returns the number of
2933 elements in the array. </li>
2936 <dt><tt>PointerType</tt></dt>
2937 <dd>Subclass of SequentialType for pointer types.</dd>
2938 <dt><tt>VectorType</tt></dt>
2939 <dd>Subclass of SequentialType for vector types. A
2940 vector type is similar to an ArrayType but is distinguished because it is
2941 a first class type wherease ArrayType is not. Vector types are used for
2942 vector operations and are usually small vectors of of an integer or floating
2944 <dt><tt>StructType</tt></dt>
2945 <dd>Subclass of DerivedTypes for struct types.</dd>
2946 <dt><tt><a name="FunctionType">FunctionType</a></tt></dt>
2947 <dd>Subclass of DerivedTypes for function types.
2949 <li><tt>bool isVarArg() const</tt>: Returns true if its a vararg
2951 <li><tt> const Type * getReturnType() const</tt>: Returns the
2952 return type of the function.</li>
2953 <li><tt>const Type * getParamType (unsigned i)</tt>: Returns
2954 the type of the ith parameter.</li>
2955 <li><tt> const unsigned getNumParams() const</tt>: Returns the
2956 number of formal parameters.</li>
2959 <dt><tt>OpaqueType</tt></dt>
2960 <dd>Sublcass of DerivedType for abstract types. This class
2961 defines no content and is used as a placeholder for some other type. Note
2962 that OpaqueType is used (temporarily) during type resolution for forward
2963 references of types. Once the referenced type is resolved, the OpaqueType
2964 is replaced with the actual type. OpaqueType can also be used for data
2965 abstraction. At link time opaque types can be resolved to actual types
2966 of the same name.</dd>
2972 <!-- ======================================================================= -->
2973 <div class="doc_subsection">
2974 <a name="Module">The <tt>Module</tt> class</a>
2977 <div class="doc_text">
2980 href="/doxygen/Module_8h-source.html">llvm/Module.h</a>"</tt><br> doxygen info:
2981 <a href="/doxygen/classllvm_1_1Module.html">Module Class</a></p>
2983 <p>The <tt>Module</tt> class represents the top level structure present in LLVM
2984 programs. An LLVM module is effectively either a translation unit of the
2985 original program or a combination of several translation units merged by the
2986 linker. The <tt>Module</tt> class keeps track of a list of <a
2987 href="#Function"><tt>Function</tt></a>s, a list of <a
2988 href="#GlobalVariable"><tt>GlobalVariable</tt></a>s, and a <a
2989 href="#SymbolTable"><tt>SymbolTable</tt></a>. Additionally, it contains a few
2990 helpful member functions that try to make common operations easy.</p>
2994 <!-- _______________________________________________________________________ -->
2995 <div class="doc_subsubsection">
2996 <a name="m_Module">Important Public Members of the <tt>Module</tt> class</a>
2999 <div class="doc_text">
3002 <li><tt>Module::Module(std::string name = "")</tt></li>
3005 <p>Constructing a <a href="#Module">Module</a> is easy. You can optionally
3006 provide a name for it (probably based on the name of the translation unit).</p>
3009 <li><tt>Module::iterator</tt> - Typedef for function list iterator<br>
3010 <tt>Module::const_iterator</tt> - Typedef for const_iterator.<br>
3012 <tt>begin()</tt>, <tt>end()</tt>
3013 <tt>size()</tt>, <tt>empty()</tt>
3015 <p>These are forwarding methods that make it easy to access the contents of
3016 a <tt>Module</tt> object's <a href="#Function"><tt>Function</tt></a>
3019 <li><tt>Module::FunctionListType &getFunctionList()</tt>
3021 <p> Returns the list of <a href="#Function"><tt>Function</tt></a>s. This is
3022 necessary to use when you need to update the list or perform a complex
3023 action that doesn't have a forwarding method.</p>
3025 <p><!-- Global Variable --></p></li>
3031 <li><tt>Module::global_iterator</tt> - Typedef for global variable list iterator<br>
3033 <tt>Module::const_global_iterator</tt> - Typedef for const_iterator.<br>
3035 <tt>global_begin()</tt>, <tt>global_end()</tt>
3036 <tt>global_size()</tt>, <tt>global_empty()</tt>
3038 <p> These are forwarding methods that make it easy to access the contents of
3039 a <tt>Module</tt> object's <a
3040 href="#GlobalVariable"><tt>GlobalVariable</tt></a> list.</p></li>
3042 <li><tt>Module::GlobalListType &getGlobalList()</tt>
3044 <p>Returns the list of <a
3045 href="#GlobalVariable"><tt>GlobalVariable</tt></a>s. This is necessary to
3046 use when you need to update the list or perform a complex action that
3047 doesn't have a forwarding method.</p>
3049 <p><!-- Symbol table stuff --> </p></li>
3055 <li><tt><a href="#SymbolTable">SymbolTable</a> *getSymbolTable()</tt>
3057 <p>Return a reference to the <a href="#SymbolTable"><tt>SymbolTable</tt></a>
3058 for this <tt>Module</tt>.</p>
3060 <p><!-- Convenience methods --></p></li>
3066 <li><tt><a href="#Function">Function</a> *getFunction(const std::string
3067 &Name, const <a href="#FunctionType">FunctionType</a> *Ty)</tt>
3069 <p>Look up the specified function in the <tt>Module</tt> <a
3070 href="#SymbolTable"><tt>SymbolTable</tt></a>. If it does not exist, return
3071 <tt>null</tt>.</p></li>
3073 <li><tt><a href="#Function">Function</a> *getOrInsertFunction(const
3074 std::string &Name, const <a href="#FunctionType">FunctionType</a> *T)</tt>
3076 <p>Look up the specified function in the <tt>Module</tt> <a
3077 href="#SymbolTable"><tt>SymbolTable</tt></a>. If it does not exist, add an
3078 external declaration for the function and return it.</p></li>
3080 <li><tt>std::string getTypeName(const <a href="#Type">Type</a> *Ty)</tt>
3082 <p>If there is at least one entry in the <a
3083 href="#SymbolTable"><tt>SymbolTable</tt></a> for the specified <a
3084 href="#Type"><tt>Type</tt></a>, return it. Otherwise return the empty
3087 <li><tt>bool addTypeName(const std::string &Name, const <a
3088 href="#Type">Type</a> *Ty)</tt>
3090 <p>Insert an entry in the <a href="#SymbolTable"><tt>SymbolTable</tt></a>
3091 mapping <tt>Name</tt> to <tt>Ty</tt>. If there is already an entry for this
3092 name, true is returned and the <a
3093 href="#SymbolTable"><tt>SymbolTable</tt></a> is not modified.</p></li>
3099 <!-- ======================================================================= -->
3100 <div class="doc_subsection">
3101 <a name="Value">The <tt>Value</tt> class</a>
3104 <div class="doc_text">
3106 <p><tt>#include "<a href="/doxygen/Value_8h-source.html">llvm/Value.h</a>"</tt>
3108 doxygen info: <a href="/doxygen/classllvm_1_1Value.html">Value Class</a></p>
3110 <p>The <tt>Value</tt> class is the most important class in the LLVM Source
3111 base. It represents a typed value that may be used (among other things) as an
3112 operand to an instruction. There are many different types of <tt>Value</tt>s,
3113 such as <a href="#Constant"><tt>Constant</tt></a>s,<a
3114 href="#Argument"><tt>Argument</tt></a>s. Even <a
3115 href="#Instruction"><tt>Instruction</tt></a>s and <a
3116 href="#Function"><tt>Function</tt></a>s are <tt>Value</tt>s.</p>
3118 <p>A particular <tt>Value</tt> may be used many times in the LLVM representation
3119 for a program. For example, an incoming argument to a function (represented
3120 with an instance of the <a href="#Argument">Argument</a> class) is "used" by
3121 every instruction in the function that references the argument. To keep track
3122 of this relationship, the <tt>Value</tt> class keeps a list of all of the <a
3123 href="#User"><tt>User</tt></a>s that is using it (the <a
3124 href="#User"><tt>User</tt></a> class is a base class for all nodes in the LLVM
3125 graph that can refer to <tt>Value</tt>s). This use list is how LLVM represents
3126 def-use information in the program, and is accessible through the <tt>use_</tt>*
3127 methods, shown below.</p>
3129 <p>Because LLVM is a typed representation, every LLVM <tt>Value</tt> is typed,
3130 and this <a href="#Type">Type</a> is available through the <tt>getType()</tt>
3131 method. In addition, all LLVM values can be named. The "name" of the
3132 <tt>Value</tt> is a symbolic string printed in the LLVM code:</p>
3134 <div class="doc_code">
3136 %<b>foo</b> = add i32 1, 2
3140 <p><a name="nameWarning">The name of this instruction is "foo".</a> <b>NOTE</b>
3141 that the name of any value may be missing (an empty string), so names should
3142 <b>ONLY</b> be used for debugging (making the source code easier to read,
3143 debugging printouts), they should not be used to keep track of values or map
3144 between them. For this purpose, use a <tt>std::map</tt> of pointers to the
3145 <tt>Value</tt> itself instead.</p>
3147 <p>One important aspect of LLVM is that there is no distinction between an SSA
3148 variable and the operation that produces it. Because of this, any reference to
3149 the value produced by an instruction (or the value available as an incoming
3150 argument, for example) is represented as a direct pointer to the instance of
3152 represents this value. Although this may take some getting used to, it
3153 simplifies the representation and makes it easier to manipulate.</p>
3157 <!-- _______________________________________________________________________ -->
3158 <div class="doc_subsubsection">
3159 <a name="m_Value">Important Public Members of the <tt>Value</tt> class</a>
3162 <div class="doc_text">
3165 <li><tt>Value::use_iterator</tt> - Typedef for iterator over the
3167 <tt>Value::use_const_iterator</tt> - Typedef for const_iterator over
3169 <tt>unsigned use_size()</tt> - Returns the number of users of the
3171 <tt>bool use_empty()</tt> - Returns true if there are no users.<br>
3172 <tt>use_iterator use_begin()</tt> - Get an iterator to the start of
3174 <tt>use_iterator use_end()</tt> - Get an iterator to the end of the
3176 <tt><a href="#User">User</a> *use_back()</tt> - Returns the last
3177 element in the list.
3178 <p> These methods are the interface to access the def-use
3179 information in LLVM. As with all other iterators in LLVM, the naming
3180 conventions follow the conventions defined by the <a href="#stl">STL</a>.</p>
3182 <li><tt><a href="#Type">Type</a> *getType() const</tt>
3183 <p>This method returns the Type of the Value.</p>
3185 <li><tt>bool hasName() const</tt><br>
3186 <tt>std::string getName() const</tt><br>
3187 <tt>void setName(const std::string &Name)</tt>
3188 <p> This family of methods is used to access and assign a name to a <tt>Value</tt>,
3189 be aware of the <a href="#nameWarning">precaution above</a>.</p>
3191 <li><tt>void replaceAllUsesWith(Value *V)</tt>
3193 <p>This method traverses the use list of a <tt>Value</tt> changing all <a
3194 href="#User"><tt>User</tt>s</a> of the current value to refer to
3195 "<tt>V</tt>" instead. For example, if you detect that an instruction always
3196 produces a constant value (for example through constant folding), you can
3197 replace all uses of the instruction with the constant like this:</p>
3199 <div class="doc_code">
3201 Inst->replaceAllUsesWith(ConstVal);
3209 <!-- ======================================================================= -->
3210 <div class="doc_subsection">
3211 <a name="User">The <tt>User</tt> class</a>
3214 <div class="doc_text">
3217 <tt>#include "<a href="/doxygen/User_8h-source.html">llvm/User.h</a>"</tt><br>
3218 doxygen info: <a href="/doxygen/classllvm_1_1User.html">User Class</a><br>
3219 Superclass: <a href="#Value"><tt>Value</tt></a></p>
3221 <p>The <tt>User</tt> class is the common base class of all LLVM nodes that may
3222 refer to <a href="#Value"><tt>Value</tt></a>s. It exposes a list of "Operands"
3223 that are all of the <a href="#Value"><tt>Value</tt></a>s that the User is
3224 referring to. The <tt>User</tt> class itself is a subclass of
3227 <p>The operands of a <tt>User</tt> point directly to the LLVM <a
3228 href="#Value"><tt>Value</tt></a> that it refers to. Because LLVM uses Static
3229 Single Assignment (SSA) form, there can only be one definition referred to,
3230 allowing this direct connection. This connection provides the use-def
3231 information in LLVM.</p>
3235 <!-- _______________________________________________________________________ -->
3236 <div class="doc_subsubsection">
3237 <a name="m_User">Important Public Members of the <tt>User</tt> class</a>
3240 <div class="doc_text">
3242 <p>The <tt>User</tt> class exposes the operand list in two ways: through
3243 an index access interface and through an iterator based interface.</p>
3246 <li><tt>Value *getOperand(unsigned i)</tt><br>
3247 <tt>unsigned getNumOperands()</tt>
3248 <p> These two methods expose the operands of the <tt>User</tt> in a
3249 convenient form for direct access.</p></li>
3251 <li><tt>User::op_iterator</tt> - Typedef for iterator over the operand
3253 <tt>op_iterator op_begin()</tt> - Get an iterator to the start of
3254 the operand list.<br>
3255 <tt>op_iterator op_end()</tt> - Get an iterator to the end of the
3257 <p> Together, these methods make up the iterator based interface to
3258 the operands of a <tt>User</tt>.</p></li>
3263 <!-- ======================================================================= -->
3264 <div class="doc_subsection">
3265 <a name="Instruction">The <tt>Instruction</tt> class</a>
3268 <div class="doc_text">
3270 <p><tt>#include "</tt><tt><a
3271 href="/doxygen/Instruction_8h-source.html">llvm/Instruction.h</a>"</tt><br>
3272 doxygen info: <a href="/doxygen/classllvm_1_1Instruction.html">Instruction Class</a><br>
3273 Superclasses: <a href="#User"><tt>User</tt></a>, <a
3274 href="#Value"><tt>Value</tt></a></p>
3276 <p>The <tt>Instruction</tt> class is the common base class for all LLVM
3277 instructions. It provides only a few methods, but is a very commonly used
3278 class. The primary data tracked by the <tt>Instruction</tt> class itself is the
3279 opcode (instruction type) and the parent <a
3280 href="#BasicBlock"><tt>BasicBlock</tt></a> the <tt>Instruction</tt> is embedded
3281 into. To represent a specific type of instruction, one of many subclasses of
3282 <tt>Instruction</tt> are used.</p>
3284 <p> Because the <tt>Instruction</tt> class subclasses the <a
3285 href="#User"><tt>User</tt></a> class, its operands can be accessed in the same
3286 way as for other <a href="#User"><tt>User</tt></a>s (with the
3287 <tt>getOperand()</tt>/<tt>getNumOperands()</tt> and
3288 <tt>op_begin()</tt>/<tt>op_end()</tt> methods).</p> <p> An important file for
3289 the <tt>Instruction</tt> class is the <tt>llvm/Instruction.def</tt> file. This
3290 file contains some meta-data about the various different types of instructions
3291 in LLVM. It describes the enum values that are used as opcodes (for example
3292 <tt>Instruction::Add</tt> and <tt>Instruction::ICmp</tt>), as well as the
3293 concrete sub-classes of <tt>Instruction</tt> that implement the instruction (for
3294 example <tt><a href="#BinaryOperator">BinaryOperator</a></tt> and <tt><a
3295 href="#CmpInst">CmpInst</a></tt>). Unfortunately, the use of macros in
3296 this file confuses doxygen, so these enum values don't show up correctly in the
3297 <a href="/doxygen/classllvm_1_1Instruction.html">doxygen output</a>.</p>
3301 <!-- _______________________________________________________________________ -->
3302 <div class="doc_subsubsection">
3303 <a name="s_Instruction">Important Subclasses of the <tt>Instruction</tt>
3306 <div class="doc_text">
3308 <li><tt><a name="BinaryOperator">BinaryOperator</a></tt>
3309 <p>This subclasses represents all two operand instructions whose operands
3310 must be the same type, except for the comparison instructions.</p></li>
3311 <li><tt><a name="CastInst">CastInst</a></tt>
3312 <p>This subclass is the parent of the 12 casting instructions. It provides
3313 common operations on cast instructions.</p>
3314 <li><tt><a name="CmpInst">CmpInst</a></tt>
3315 <p>This subclass respresents the two comparison instructions,
3316 <a href="LangRef.html#i_icmp">ICmpInst</a> (integer opreands), and
3317 <a href="LangRef.html#i_fcmp">FCmpInst</a> (floating point operands).</p>
3318 <li><tt><a name="TerminatorInst">TerminatorInst</a></tt>
3319 <p>This subclass is the parent of all terminator instructions (those which
3320 can terminate a block).</p>
3324 <!-- _______________________________________________________________________ -->
3325 <div class="doc_subsubsection">
3326 <a name="m_Instruction">Important Public Members of the <tt>Instruction</tt>
3330 <div class="doc_text">
3333 <li><tt><a href="#BasicBlock">BasicBlock</a> *getParent()</tt>
3334 <p>Returns the <a href="#BasicBlock"><tt>BasicBlock</tt></a> that
3335 this <tt>Instruction</tt> is embedded into.</p></li>
3336 <li><tt>bool mayWriteToMemory()</tt>
3337 <p>Returns true if the instruction writes to memory, i.e. it is a
3338 <tt>call</tt>,<tt>free</tt>,<tt>invoke</tt>, or <tt>store</tt>.</p></li>
3339 <li><tt>unsigned getOpcode()</tt>
3340 <p>Returns the opcode for the <tt>Instruction</tt>.</p></li>
3341 <li><tt><a href="#Instruction">Instruction</a> *clone() const</tt>
3342 <p>Returns another instance of the specified instruction, identical
3343 in all ways to the original except that the instruction has no parent
3344 (ie it's not embedded into a <a href="#BasicBlock"><tt>BasicBlock</tt></a>),
3345 and it has no name</p></li>
3350 <!-- ======================================================================= -->
3351 <div class="doc_subsection">
3352 <a name="Constant">The <tt>Constant</tt> class and subclasses</a>
3355 <div class="doc_text">
3357 <p>Constant represents a base class for different types of constants. It
3358 is subclassed by ConstantInt, ConstantArray, etc. for representing
3359 the various types of Constants. <a href="#GlobalValue">GlobalValue</a> is also
3360 a subclass, which represents the address of a global variable or function.
3365 <!-- _______________________________________________________________________ -->
3366 <div class="doc_subsubsection">Important Subclasses of Constant </div>
3367 <div class="doc_text">
3369 <li>ConstantInt : This subclass of Constant represents an integer constant of
3372 <li><tt>const APInt& getValue() const</tt>: Returns the underlying
3373 value of this constant, an APInt value.</li>
3374 <li><tt>int64_t getSExtValue() const</tt>: Converts the underlying APInt
3375 value to an int64_t via sign extension. If the value (not the bit width)
3376 of the APInt is too large to fit in an int64_t, an assertion will result.
3377 For this reason, use of this method is discouraged.</li>
3378 <li><tt>uint64_t getZExtValue() const</tt>: Converts the underlying APInt
3379 value to a uint64_t via zero extension. IF the value (not the bit width)
3380 of the APInt is too large to fit in a uint64_t, an assertion will result.
3381 For this reason, use of this method is discouraged.</li>
3382 <li><tt>static ConstantInt* get(const APInt& Val)</tt>: Returns the
3383 ConstantInt object that represents the value provided by <tt>Val</tt>.
3384 The type is implied as the IntegerType that corresponds to the bit width
3385 of <tt>Val</tt>.</li>
3386 <li><tt>static ConstantInt* get(const Type *Ty, uint64_t Val)</tt>:
3387 Returns the ConstantInt object that represents the value provided by
3388 <tt>Val</tt> for integer type <tt>Ty</tt>.</li>
3391 <li>ConstantFP : This class represents a floating point constant.
3393 <li><tt>double getValue() const</tt>: Returns the underlying value of
3394 this constant. </li>
3397 <li>ConstantArray : This represents a constant array.
3399 <li><tt>const std::vector<Use> &getValues() const</tt>: Returns
3400 a vector of component constants that makeup this array. </li>
3403 <li>ConstantStruct : This represents a constant struct.
3405 <li><tt>const std::vector<Use> &getValues() const</tt>: Returns
3406 a vector of component constants that makeup this array. </li>
3409 <li>GlobalValue : This represents either a global variable or a function. In
3410 either case, the value is a constant fixed address (after linking).
3416 <!-- ======================================================================= -->
3417 <div class="doc_subsection">
3418 <a name="GlobalValue">The <tt>GlobalValue</tt> class</a>
3421 <div class="doc_text">
3424 href="/doxygen/GlobalValue_8h-source.html">llvm/GlobalValue.h</a>"</tt><br>
3425 doxygen info: <a href="/doxygen/classllvm_1_1GlobalValue.html">GlobalValue
3427 Superclasses: <a href="#Constant"><tt>Constant</tt></a>,
3428 <a href="#User"><tt>User</tt></a>, <a href="#Value"><tt>Value</tt></a></p>
3430 <p>Global values (<a href="#GlobalVariable"><tt>GlobalVariable</tt></a>s or <a
3431 href="#Function"><tt>Function</tt></a>s) are the only LLVM values that are
3432 visible in the bodies of all <a href="#Function"><tt>Function</tt></a>s.
3433 Because they are visible at global scope, they are also subject to linking with
3434 other globals defined in different translation units. To control the linking
3435 process, <tt>GlobalValue</tt>s know their linkage rules. Specifically,
3436 <tt>GlobalValue</tt>s know whether they have internal or external linkage, as
3437 defined by the <tt>LinkageTypes</tt> enumeration.</p>
3439 <p>If a <tt>GlobalValue</tt> has internal linkage (equivalent to being
3440 <tt>static</tt> in C), it is not visible to code outside the current translation
3441 unit, and does not participate in linking. If it has external linkage, it is
3442 visible to external code, and does participate in linking. In addition to
3443 linkage information, <tt>GlobalValue</tt>s keep track of which <a
3444 href="#Module"><tt>Module</tt></a> they are currently part of.</p>
3446 <p>Because <tt>GlobalValue</tt>s are memory objects, they are always referred to
3447 by their <b>address</b>. As such, the <a href="#Type"><tt>Type</tt></a> of a
3448 global is always a pointer to its contents. It is important to remember this
3449 when using the <tt>GetElementPtrInst</tt> instruction because this pointer must
3450 be dereferenced first. For example, if you have a <tt>GlobalVariable</tt> (a
3451 subclass of <tt>GlobalValue)</tt> that is an array of 24 ints, type <tt>[24 x
3452 i32]</tt>, then the <tt>GlobalVariable</tt> is a pointer to that array. Although
3453 the address of the first element of this array and the value of the
3454 <tt>GlobalVariable</tt> are the same, they have different types. The
3455 <tt>GlobalVariable</tt>'s type is <tt>[24 x i32]</tt>. The first element's type
3456 is <tt>i32.</tt> Because of this, accessing a global value requires you to
3457 dereference the pointer with <tt>GetElementPtrInst</tt> first, then its elements
3458 can be accessed. This is explained in the <a href="LangRef.html#globalvars">LLVM
3459 Language Reference Manual</a>.</p>
3463 <!-- _______________________________________________________________________ -->
3464 <div class="doc_subsubsection">
3465 <a name="m_GlobalValue">Important Public Members of the <tt>GlobalValue</tt>
3469 <div class="doc_text">
3472 <li><tt>bool hasInternalLinkage() const</tt><br>
3473 <tt>bool hasExternalLinkage() const</tt><br>
3474 <tt>void setInternalLinkage(bool HasInternalLinkage)</tt>
3475 <p> These methods manipulate the linkage characteristics of the <tt>GlobalValue</tt>.</p>
3478 <li><tt><a href="#Module">Module</a> *getParent()</tt>
3479 <p> This returns the <a href="#Module"><tt>Module</tt></a> that the
3480 GlobalValue is currently embedded into.</p></li>
3485 <!-- ======================================================================= -->
3486 <div class="doc_subsection">
3487 <a name="Function">The <tt>Function</tt> class</a>
3490 <div class="doc_text">
3493 href="/doxygen/Function_8h-source.html">llvm/Function.h</a>"</tt><br> doxygen
3494 info: <a href="/doxygen/classllvm_1_1Function.html">Function Class</a><br>
3495 Superclasses: <a href="#GlobalValue"><tt>GlobalValue</tt></a>,
3496 <a href="#Constant"><tt>Constant</tt></a>,
3497 <a href="#User"><tt>User</tt></a>,
3498 <a href="#Value"><tt>Value</tt></a></p>
3500 <p>The <tt>Function</tt> class represents a single procedure in LLVM. It is
3501 actually one of the more complex classes in the LLVM heirarchy because it must
3502 keep track of a large amount of data. The <tt>Function</tt> class keeps track
3503 of a list of <a href="#BasicBlock"><tt>BasicBlock</tt></a>s, a list of formal
3504 <a href="#Argument"><tt>Argument</tt></a>s, and a
3505 <a href="#SymbolTable"><tt>SymbolTable</tt></a>.</p>
3507 <p>The list of <a href="#BasicBlock"><tt>BasicBlock</tt></a>s is the most
3508 commonly used part of <tt>Function</tt> objects. The list imposes an implicit
3509 ordering of the blocks in the function, which indicate how the code will be
3510 layed out by the backend. Additionally, the first <a
3511 href="#BasicBlock"><tt>BasicBlock</tt></a> is the implicit entry node for the
3512 <tt>Function</tt>. It is not legal in LLVM to explicitly branch to this initial
3513 block. There are no implicit exit nodes, and in fact there may be multiple exit
3514 nodes from a single <tt>Function</tt>. If the <a
3515 href="#BasicBlock"><tt>BasicBlock</tt></a> list is empty, this indicates that
3516 the <tt>Function</tt> is actually a function declaration: the actual body of the
3517 function hasn't been linked in yet.</p>
3519 <p>In addition to a list of <a href="#BasicBlock"><tt>BasicBlock</tt></a>s, the
3520 <tt>Function</tt> class also keeps track of the list of formal <a
3521 href="#Argument"><tt>Argument</tt></a>s that the function receives. This
3522 container manages the lifetime of the <a href="#Argument"><tt>Argument</tt></a>
3523 nodes, just like the <a href="#BasicBlock"><tt>BasicBlock</tt></a> list does for
3524 the <a href="#BasicBlock"><tt>BasicBlock</tt></a>s.</p>
3526 <p>The <a href="#SymbolTable"><tt>SymbolTable</tt></a> is a very rarely used
3527 LLVM feature that is only used when you have to look up a value by name. Aside
3528 from that, the <a href="#SymbolTable"><tt>SymbolTable</tt></a> is used
3529 internally to make sure that there are not conflicts between the names of <a
3530 href="#Instruction"><tt>Instruction</tt></a>s, <a
3531 href="#BasicBlock"><tt>BasicBlock</tt></a>s, or <a
3532 href="#Argument"><tt>Argument</tt></a>s in the function body.</p>
3534 <p>Note that <tt>Function</tt> is a <a href="#GlobalValue">GlobalValue</a>
3535 and therefore also a <a href="#Constant">Constant</a>. The value of the function
3536 is its address (after linking) which is guaranteed to be constant.</p>
3539 <!-- _______________________________________________________________________ -->
3540 <div class="doc_subsubsection">
3541 <a name="m_Function">Important Public Members of the <tt>Function</tt>
3545 <div class="doc_text">
3548 <li><tt>Function(const </tt><tt><a href="#FunctionType">FunctionType</a>
3549 *Ty, LinkageTypes Linkage, const std::string &N = "", Module* Parent = 0)</tt>
3551 <p>Constructor used when you need to create new <tt>Function</tt>s to add
3552 the the program. The constructor must specify the type of the function to
3553 create and what type of linkage the function should have. The <a
3554 href="#FunctionType"><tt>FunctionType</tt></a> argument
3555 specifies the formal arguments and return value for the function. The same
3556 <a href="#FunctionType"><tt>FunctionType</tt></a> value can be used to
3557 create multiple functions. The <tt>Parent</tt> argument specifies the Module
3558 in which the function is defined. If this argument is provided, the function
3559 will automatically be inserted into that module's list of
3562 <li><tt>bool isDeclaration()</tt>
3564 <p>Return whether or not the <tt>Function</tt> has a body defined. If the
3565 function is "external", it does not have a body, and thus must be resolved
3566 by linking with a function defined in a different translation unit.</p></li>
3568 <li><tt>Function::iterator</tt> - Typedef for basic block list iterator<br>
3569 <tt>Function::const_iterator</tt> - Typedef for const_iterator.<br>
3571 <tt>begin()</tt>, <tt>end()</tt>
3572 <tt>size()</tt>, <tt>empty()</tt>
3574 <p>These are forwarding methods that make it easy to access the contents of
3575 a <tt>Function</tt> object's <a href="#BasicBlock"><tt>BasicBlock</tt></a>
3578 <li><tt>Function::BasicBlockListType &getBasicBlockList()</tt>
3580 <p>Returns the list of <a href="#BasicBlock"><tt>BasicBlock</tt></a>s. This
3581 is necessary to use when you need to update the list or perform a complex
3582 action that doesn't have a forwarding method.</p></li>
3584 <li><tt>Function::arg_iterator</tt> - Typedef for the argument list
3586 <tt>Function::const_arg_iterator</tt> - Typedef for const_iterator.<br>
3588 <tt>arg_begin()</tt>, <tt>arg_end()</tt>
3589 <tt>arg_size()</tt>, <tt>arg_empty()</tt>
3591 <p>These are forwarding methods that make it easy to access the contents of
3592 a <tt>Function</tt> object's <a href="#Argument"><tt>Argument</tt></a>
3595 <li><tt>Function::ArgumentListType &getArgumentList()</tt>
3597 <p>Returns the list of <a href="#Argument"><tt>Argument</tt></a>s. This is
3598 necessary to use when you need to update the list or perform a complex
3599 action that doesn't have a forwarding method.</p></li>
3601 <li><tt><a href="#BasicBlock">BasicBlock</a> &getEntryBlock()</tt>
3603 <p>Returns the entry <a href="#BasicBlock"><tt>BasicBlock</tt></a> for the
3604 function. Because the entry block for the function is always the first
3605 block, this returns the first block of the <tt>Function</tt>.</p></li>
3607 <li><tt><a href="#Type">Type</a> *getReturnType()</tt><br>
3608 <tt><a href="#FunctionType">FunctionType</a> *getFunctionType()</tt>
3610 <p>This traverses the <a href="#Type"><tt>Type</tt></a> of the
3611 <tt>Function</tt> and returns the return type of the function, or the <a
3612 href="#FunctionType"><tt>FunctionType</tt></a> of the actual
3615 <li><tt><a href="#SymbolTable">SymbolTable</a> *getSymbolTable()</tt>
3617 <p> Return a pointer to the <a href="#SymbolTable"><tt>SymbolTable</tt></a>
3618 for this <tt>Function</tt>.</p></li>
3623 <!-- ======================================================================= -->
3624 <div class="doc_subsection">
3625 <a name="GlobalVariable">The <tt>GlobalVariable</tt> class</a>
3628 <div class="doc_text">
3631 href="/doxygen/GlobalVariable_8h-source.html">llvm/GlobalVariable.h</a>"</tt>
3633 doxygen info: <a href="/doxygen/classllvm_1_1GlobalVariable.html">GlobalVariable
3635 Superclasses: <a href="#GlobalValue"><tt>GlobalValue</tt></a>,
3636 <a href="#Constant"><tt>Constant</tt></a>,
3637 <a href="#User"><tt>User</tt></a>,
3638 <a href="#Value"><tt>Value</tt></a></p>
3640 <p>Global variables are represented with the (suprise suprise)
3641 <tt>GlobalVariable</tt> class. Like functions, <tt>GlobalVariable</tt>s are also
3642 subclasses of <a href="#GlobalValue"><tt>GlobalValue</tt></a>, and as such are
3643 always referenced by their address (global values must live in memory, so their
3644 "name" refers to their constant address). See
3645 <a href="#GlobalValue"><tt>GlobalValue</tt></a> for more on this. Global
3646 variables may have an initial value (which must be a
3647 <a href="#Constant"><tt>Constant</tt></a>), and if they have an initializer,
3648 they may be marked as "constant" themselves (indicating that their contents
3649 never change at runtime).</p>
3652 <!-- _______________________________________________________________________ -->
3653 <div class="doc_subsubsection">
3654 <a name="m_GlobalVariable">Important Public Members of the
3655 <tt>GlobalVariable</tt> class</a>
3658 <div class="doc_text">
3661 <li><tt>GlobalVariable(const </tt><tt><a href="#Type">Type</a> *Ty, bool
3662 isConstant, LinkageTypes& Linkage, <a href="#Constant">Constant</a>
3663 *Initializer = 0, const std::string &Name = "", Module* Parent = 0)</tt>
3665 <p>Create a new global variable of the specified type. If
3666 <tt>isConstant</tt> is true then the global variable will be marked as
3667 unchanging for the program. The Linkage parameter specifies the type of
3668 linkage (internal, external, weak, linkonce, appending) for the variable.
3669 If the linkage is InternalLinkage, WeakAnyLinkage, WeakODRLinkage,
3670 LinkOnceAnyLinkage or LinkOnceODRLinkage, then the resultant
3671 global variable will have internal linkage. AppendingLinkage concatenates
3672 together all instances (in different translation units) of the variable
3673 into a single variable but is only applicable to arrays. See
3674 the <a href="LangRef.html#modulestructure">LLVM Language Reference</a> for
3675 further details on linkage types. Optionally an initializer, a name, and the
3676 module to put the variable into may be specified for the global variable as
3679 <li><tt>bool isConstant() const</tt>
3681 <p>Returns true if this is a global variable that is known not to
3682 be modified at runtime.</p></li>
3684 <li><tt>bool hasInitializer()</tt>
3686 <p>Returns true if this <tt>GlobalVariable</tt> has an intializer.</p></li>
3688 <li><tt><a href="#Constant">Constant</a> *getInitializer()</tt>
3690 <p>Returns the intial value for a <tt>GlobalVariable</tt>. It is not legal
3691 to call this method if there is no initializer.</p></li>
3697 <!-- ======================================================================= -->
3698 <div class="doc_subsection">
3699 <a name="BasicBlock">The <tt>BasicBlock</tt> class</a>
3702 <div class="doc_text">
3705 href="/doxygen/BasicBlock_8h-source.html">llvm/BasicBlock.h</a>"</tt><br>
3706 doxygen info: <a href="/doxygen/classllvm_1_1BasicBlock.html">BasicBlock
3708 Superclass: <a href="#Value"><tt>Value</tt></a></p>
3710 <p>This class represents a single entry multiple exit section of the code,
3711 commonly known as a basic block by the compiler community. The
3712 <tt>BasicBlock</tt> class maintains a list of <a
3713 href="#Instruction"><tt>Instruction</tt></a>s, which form the body of the block.
3714 Matching the language definition, the last element of this list of instructions
3715 is always a terminator instruction (a subclass of the <a
3716 href="#TerminatorInst"><tt>TerminatorInst</tt></a> class).</p>
3718 <p>In addition to tracking the list of instructions that make up the block, the
3719 <tt>BasicBlock</tt> class also keeps track of the <a
3720 href="#Function"><tt>Function</tt></a> that it is embedded into.</p>
3722 <p>Note that <tt>BasicBlock</tt>s themselves are <a
3723 href="#Value"><tt>Value</tt></a>s, because they are referenced by instructions
3724 like branches and can go in the switch tables. <tt>BasicBlock</tt>s have type
3729 <!-- _______________________________________________________________________ -->
3730 <div class="doc_subsubsection">
3731 <a name="m_BasicBlock">Important Public Members of the <tt>BasicBlock</tt>
3735 <div class="doc_text">
3738 <li><tt>BasicBlock(const std::string &Name = "", </tt><tt><a
3739 href="#Function">Function</a> *Parent = 0)</tt>
3741 <p>The <tt>BasicBlock</tt> constructor is used to create new basic blocks for
3742 insertion into a function. The constructor optionally takes a name for the new
3743 block, and a <a href="#Function"><tt>Function</tt></a> to insert it into. If
3744 the <tt>Parent</tt> parameter is specified, the new <tt>BasicBlock</tt> is
3745 automatically inserted at the end of the specified <a
3746 href="#Function"><tt>Function</tt></a>, if not specified, the BasicBlock must be
3747 manually inserted into the <a href="#Function"><tt>Function</tt></a>.</p></li>
3749 <li><tt>BasicBlock::iterator</tt> - Typedef for instruction list iterator<br>
3750 <tt>BasicBlock::const_iterator</tt> - Typedef for const_iterator.<br>
3751 <tt>begin()</tt>, <tt>end()</tt>, <tt>front()</tt>, <tt>back()</tt>,
3752 <tt>size()</tt>, <tt>empty()</tt>
3753 STL-style functions for accessing the instruction list.
3755 <p>These methods and typedefs are forwarding functions that have the same
3756 semantics as the standard library methods of the same names. These methods
3757 expose the underlying instruction list of a basic block in a way that is easy to
3758 manipulate. To get the full complement of container operations (including
3759 operations to update the list), you must use the <tt>getInstList()</tt>
3762 <li><tt>BasicBlock::InstListType &getInstList()</tt>
3764 <p>This method is used to get access to the underlying container that actually
3765 holds the Instructions. This method must be used when there isn't a forwarding
3766 function in the <tt>BasicBlock</tt> class for the operation that you would like
3767 to perform. Because there are no forwarding functions for "updating"
3768 operations, you need to use this if you want to update the contents of a
3769 <tt>BasicBlock</tt>.</p></li>
3771 <li><tt><a href="#Function">Function</a> *getParent()</tt>
3773 <p> Returns a pointer to <a href="#Function"><tt>Function</tt></a> the block is
3774 embedded into, or a null pointer if it is homeless.</p></li>
3776 <li><tt><a href="#TerminatorInst">TerminatorInst</a> *getTerminator()</tt>
3778 <p> Returns a pointer to the terminator instruction that appears at the end of
3779 the <tt>BasicBlock</tt>. If there is no terminator instruction, or if the last
3780 instruction in the block is not a terminator, then a null pointer is
3788 <!-- ======================================================================= -->
3789 <div class="doc_subsection">
3790 <a name="Argument">The <tt>Argument</tt> class</a>
3793 <div class="doc_text">
3795 <p>This subclass of Value defines the interface for incoming formal
3796 arguments to a function. A Function maintains a list of its formal
3797 arguments. An argument has a pointer to the parent Function.</p>
3801 <!-- *********************************************************************** -->
3804 <a href="http://jigsaw.w3.org/css-validator/check/referer"><img
3805 src="http://jigsaw.w3.org/css-validator/images/vcss-blue" alt="Valid CSS"></a>
3806 <a href="http://validator.w3.org/check/referer"><img
3807 src="http://www.w3.org/Icons/valid-html401" alt="Valid HTML 4.01 Strict"></a>
3809 <a href="mailto:dhurjati@cs.uiuc.edu">Dinakar Dhurjati</a> and
3810 <a href="mailto:sabre@nondot.org">Chris Lattner</a><br>
3811 <a href="http://llvm.org">The LLVM Compiler Infrastructure</a><br>
3812 Last modified: $Date$