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)</li>
35 <li><a href="#StringRef">The <tt>StringRef</tt> class</a> </li>
36 <li><a href="#Twine">The <tt>Twine</tt> class</a> </li>
38 <li><a href="#DEBUG">The <tt>DEBUG()</tt> macro and <tt>-debug</tt>
41 <li><a href="#DEBUG_TYPE">Fine grained debug info with <tt>DEBUG_TYPE</tt>
42 and the <tt>-debug-only</tt> option</a> </li>
45 <li><a href="#Statistic">The <tt>Statistic</tt> class & <tt>-stats</tt>
48 <li>The <tt>InstVisitor</tt> template
49 <li>The general graph API
51 <li><a href="#ViewGraph">Viewing graphs while debugging code</a></li>
54 <li><a href="#datastructure">Picking the Right Data Structure for a Task</a>
56 <li><a href="#ds_sequential">Sequential Containers (std::vector, std::list, etc)</a>
58 <li><a href="#dss_fixedarrays">Fixed Size Arrays</a></li>
59 <li><a href="#dss_heaparrays">Heap Allocated Arrays</a></li>
60 <li><a href="#dss_smallvector">"llvm/ADT/SmallVector.h"</a></li>
61 <li><a href="#dss_vector"><vector></a></li>
62 <li><a href="#dss_deque"><deque></a></li>
63 <li><a href="#dss_list"><list></a></li>
64 <li><a href="#dss_ilist">llvm/ADT/ilist.h</a></li>
65 <li><a href="#dss_other">Other Sequential Container Options</a></li>
67 <li><a href="#ds_set">Set-Like Containers (std::set, SmallSet, SetVector, etc)</a>
69 <li><a href="#dss_sortedvectorset">A sorted 'vector'</a></li>
70 <li><a href="#dss_smallset">"llvm/ADT/SmallSet.h"</a></li>
71 <li><a href="#dss_smallptrset">"llvm/ADT/SmallPtrSet.h"</a></li>
72 <li><a href="#dss_denseset">"llvm/ADT/DenseSet.h"</a></li>
73 <li><a href="#dss_FoldingSet">"llvm/ADT/FoldingSet.h"</a></li>
74 <li><a href="#dss_set"><set></a></li>
75 <li><a href="#dss_setvector">"llvm/ADT/SetVector.h"</a></li>
76 <li><a href="#dss_uniquevector">"llvm/ADT/UniqueVector.h"</a></li>
77 <li><a href="#dss_otherset">Other Set-Like ContainerOptions</a></li>
79 <li><a href="#ds_map">Map-Like Containers (std::map, DenseMap, etc)</a>
81 <li><a href="#dss_sortedvectormap">A sorted 'vector'</a></li>
82 <li><a href="#dss_stringmap">"llvm/ADT/StringMap.h"</a></li>
83 <li><a href="#dss_indexedmap">"llvm/ADT/IndexedMap.h"</a></li>
84 <li><a href="#dss_densemap">"llvm/ADT/DenseMap.h"</a></li>
85 <li><a href="#dss_map"><map></a></li>
86 <li><a href="#dss_othermap">Other Map-Like Container Options</a></li>
88 <li><a href="#ds_string">String-like containers</a>
92 <li><a href="#ds_bit">BitVector-like containers</a>
94 <li><a href="#dss_bitvector">A dense bitvector</a></li>
95 <li><a href="#dss_sparsebitvector">A sparse bitvector</a></li>
99 <li><a href="#common">Helpful Hints for Common Operations</a>
101 <li><a href="#inspection">Basic Inspection and Traversal Routines</a>
103 <li><a href="#iterate_function">Iterating over the <tt>BasicBlock</tt>s
104 in a <tt>Function</tt></a> </li>
105 <li><a href="#iterate_basicblock">Iterating over the <tt>Instruction</tt>s
106 in a <tt>BasicBlock</tt></a> </li>
107 <li><a href="#iterate_institer">Iterating over the <tt>Instruction</tt>s
108 in a <tt>Function</tt></a> </li>
109 <li><a href="#iterate_convert">Turning an iterator into a
110 class pointer</a> </li>
111 <li><a href="#iterate_complex">Finding call sites: a more
112 complex example</a> </li>
113 <li><a href="#calls_and_invokes">Treating calls and invokes
114 the same way</a> </li>
115 <li><a href="#iterate_chains">Iterating over def-use &
116 use-def chains</a> </li>
117 <li><a href="#iterate_preds">Iterating over predecessors &
118 successors of blocks</a></li>
121 <li><a href="#simplechanges">Making simple changes</a>
123 <li><a href="#schanges_creating">Creating and inserting new
124 <tt>Instruction</tt>s</a> </li>
125 <li><a href="#schanges_deleting">Deleting <tt>Instruction</tt>s</a> </li>
126 <li><a href="#schanges_replacing">Replacing an <tt>Instruction</tt>
127 with another <tt>Value</tt></a> </li>
128 <li><a href="#schanges_deletingGV">Deleting <tt>GlobalVariable</tt>s</a> </li>
131 <li><a href="#create_types">How to Create Types</a></li>
133 <li>Working with the Control Flow Graph
135 <li>Accessing predecessors and successors of a <tt>BasicBlock</tt>
143 <li><a href="#threading">Threads and LLVM</a>
145 <li><a href="#startmultithreaded">Entering and Exiting Multithreaded Mode
147 <li><a href="#shutdown">Ending execution with <tt>llvm_shutdown()</tt></a></li>
148 <li><a href="#managedstatic">Lazy initialization with <tt>ManagedStatic</tt></a></li>
152 <li><a href="#advanced">Advanced Topics</a>
154 <li><a href="#TypeResolve">LLVM Type Resolution</a>
156 <li><a href="#BuildRecType">Basic Recursive Type Construction</a></li>
157 <li><a href="#refineAbstractTypeTo">The <tt>refineAbstractTypeTo</tt> method</a></li>
158 <li><a href="#PATypeHolder">The PATypeHolder Class</a></li>
159 <li><a href="#AbstractTypeUser">The AbstractTypeUser Class</a></li>
162 <li><a href="#SymbolTable">The <tt>ValueSymbolTable</tt> and <tt>TypeSymbolTable</tt> classes</a></li>
163 <li><a href="#UserLayout">The <tt>User</tt> and owned <tt>Use</tt> classes' memory layout</a></li>
166 <li><a href="#coreclasses">The Core LLVM Class Hierarchy Reference</a>
168 <li><a href="#Type">The <tt>Type</tt> class</a> </li>
169 <li><a href="#Module">The <tt>Module</tt> class</a></li>
170 <li><a href="#Value">The <tt>Value</tt> class</a>
172 <li><a href="#User">The <tt>User</tt> class</a>
174 <li><a href="#Instruction">The <tt>Instruction</tt> class</a></li>
175 <li><a href="#Constant">The <tt>Constant</tt> class</a>
177 <li><a href="#GlobalValue">The <tt>GlobalValue</tt> class</a>
179 <li><a href="#Function">The <tt>Function</tt> class</a></li>
180 <li><a href="#GlobalVariable">The <tt>GlobalVariable</tt> class</a></li>
187 <li><a href="#BasicBlock">The <tt>BasicBlock</tt> class</a></li>
188 <li><a href="#Argument">The <tt>Argument</tt> class</a></li>
195 <div class="doc_author">
196 <p>Written by <a href="mailto:sabre@nondot.org">Chris Lattner</a>,
197 <a href="mailto:dhurjati@cs.uiuc.edu">Dinakar Dhurjati</a>,
198 <a href="mailto:ggreif@gmail.com">Gabor Greif</a>,
199 <a href="mailto:jstanley@cs.uiuc.edu">Joel Stanley</a>,
200 <a href="mailto:rspencer@x10sys.com">Reid Spencer</a> and
201 <a href="mailto:owen@apple.com">Owen Anderson</a></p>
204 <!-- *********************************************************************** -->
205 <div class="doc_section">
206 <a name="introduction">Introduction </a>
208 <!-- *********************************************************************** -->
210 <div class="doc_text">
212 <p>This document is meant to highlight some of the important classes and
213 interfaces available in the LLVM source-base. This manual is not
214 intended to explain what LLVM is, how it works, and what LLVM code looks
215 like. It assumes that you know the basics of LLVM and are interested
216 in writing transformations or otherwise analyzing or manipulating the
219 <p>This document should get you oriented so that you can find your
220 way in the continuously growing source code that makes up the LLVM
221 infrastructure. Note that this manual is not intended to serve as a
222 replacement for reading the source code, so if you think there should be
223 a method in one of these classes to do something, but it's not listed,
224 check the source. Links to the <a href="/doxygen/">doxygen</a> sources
225 are provided to make this as easy as possible.</p>
227 <p>The first section of this document describes general information that is
228 useful to know when working in the LLVM infrastructure, and the second describes
229 the Core LLVM classes. In the future this manual will be extended with
230 information describing how to use extension libraries, such as dominator
231 information, CFG traversal routines, and useful utilities like the <tt><a
232 href="/doxygen/InstVisitor_8h-source.html">InstVisitor</a></tt> template.</p>
236 <!-- *********************************************************************** -->
237 <div class="doc_section">
238 <a name="general">General Information</a>
240 <!-- *********************************************************************** -->
242 <div class="doc_text">
244 <p>This section contains general information that is useful if you are working
245 in the LLVM source-base, but that isn't specific to any particular API.</p>
249 <!-- ======================================================================= -->
250 <div class="doc_subsection">
251 <a name="stl">The C++ Standard Template Library</a>
254 <div class="doc_text">
256 <p>LLVM makes heavy use of the C++ Standard Template Library (STL),
257 perhaps much more than you are used to, or have seen before. Because of
258 this, you might want to do a little background reading in the
259 techniques used and capabilities of the library. There are many good
260 pages that discuss the STL, and several books on the subject that you
261 can get, so it will not be discussed in this document.</p>
263 <p>Here are some useful links:</p>
267 <li><a href="http://www.dinkumware.com/refxcpp.html">Dinkumware C++ Library
268 reference</a> - an excellent reference for the STL and other parts of the
269 standard C++ library.</li>
271 <li><a href="http://www.tempest-sw.com/cpp/">C++ In a Nutshell</a> - This is an
272 O'Reilly book in the making. It has a decent Standard Library
273 Reference that rivals Dinkumware's, and is unfortunately no longer free since the
274 book has been published.</li>
276 <li><a href="http://www.parashift.com/c++-faq-lite/">C++ Frequently Asked
279 <li><a href="http://www.sgi.com/tech/stl/">SGI's STL Programmer's Guide</a> -
281 href="http://www.sgi.com/tech/stl/stl_introduction.html">Introduction to the
284 <li><a href="http://www.research.att.com/%7Ebs/C++.html">Bjarne Stroustrup's C++
287 <li><a href="http://64.78.49.204/">
288 Bruce Eckel's Thinking in C++, 2nd ed. Volume 2 Revision 4.0 (even better, get
293 <p>You are also encouraged to take a look at the <a
294 href="CodingStandards.html">LLVM Coding Standards</a> guide which focuses on how
295 to write maintainable code more than where to put your curly braces.</p>
299 <!-- ======================================================================= -->
300 <div class="doc_subsection">
301 <a name="stl">Other useful references</a>
304 <div class="doc_text">
307 <li><a href="http://www.psc.edu/%7Esemke/cvs_branches.html">CVS
308 Branch and Tag Primer</a></li>
309 <li><a href="http://www.fortran-2000.com/ArnaudRecipes/sharedlib.html">Using
310 static and shared libraries across platforms</a></li>
315 <!-- *********************************************************************** -->
316 <div class="doc_section">
317 <a name="apis">Important and useful LLVM APIs</a>
319 <!-- *********************************************************************** -->
321 <div class="doc_text">
323 <p>Here we highlight some LLVM APIs that are generally useful and good to
324 know about when writing transformations.</p>
328 <!-- ======================================================================= -->
329 <div class="doc_subsection">
330 <a name="isa">The <tt>isa<></tt>, <tt>cast<></tt> and
331 <tt>dyn_cast<></tt> templates</a>
334 <div class="doc_text">
336 <p>The LLVM source-base makes extensive use of a custom form of RTTI.
337 These templates have many similarities to the C++ <tt>dynamic_cast<></tt>
338 operator, but they don't have some drawbacks (primarily stemming from
339 the fact that <tt>dynamic_cast<></tt> only works on classes that
340 have a v-table). Because they are used so often, you must know what they
341 do and how they work. All of these templates are defined in the <a
342 href="/doxygen/Casting_8h-source.html"><tt>llvm/Support/Casting.h</tt></a>
343 file (note that you very rarely have to include this file directly).</p>
346 <dt><tt>isa<></tt>: </dt>
348 <dd><p>The <tt>isa<></tt> operator works exactly like the Java
349 "<tt>instanceof</tt>" operator. It returns true or false depending on whether
350 a reference or pointer points to an instance of the specified class. This can
351 be very useful for constraint checking of various sorts (example below).</p>
354 <dt><tt>cast<></tt>: </dt>
356 <dd><p>The <tt>cast<></tt> operator is a "checked cast" operation. It
357 converts a pointer or reference from a base class to a derived class, causing
358 an assertion failure if it is not really an instance of the right type. This
359 should be used in cases where you have some information that makes you believe
360 that something is of the right type. An example of the <tt>isa<></tt>
361 and <tt>cast<></tt> template is:</p>
363 <div class="doc_code">
365 static bool isLoopInvariant(const <a href="#Value">Value</a> *V, const Loop *L) {
366 if (isa<<a href="#Constant">Constant</a>>(V) || isa<<a href="#Argument">Argument</a>>(V) || isa<<a href="#GlobalValue">GlobalValue</a>>(V))
369 // <i>Otherwise, it must be an instruction...</i>
370 return !L->contains(cast<<a href="#Instruction">Instruction</a>>(V)->getParent());
375 <p>Note that you should <b>not</b> use an <tt>isa<></tt> test followed
376 by a <tt>cast<></tt>, for that use the <tt>dyn_cast<></tt>
381 <dt><tt>dyn_cast<></tt>:</dt>
383 <dd><p>The <tt>dyn_cast<></tt> operator is a "checking cast" operation.
384 It checks to see if the operand is of the specified type, and if so, returns a
385 pointer to it (this operator does not work with references). If the operand is
386 not of the correct type, a null pointer is returned. Thus, this works very
387 much like the <tt>dynamic_cast<></tt> operator in C++, and should be
388 used in the same circumstances. Typically, the <tt>dyn_cast<></tt>
389 operator is used in an <tt>if</tt> statement or some other flow control
390 statement like this:</p>
392 <div class="doc_code">
394 if (<a href="#AllocationInst">AllocationInst</a> *AI = dyn_cast<<a href="#AllocationInst">AllocationInst</a>>(Val)) {
400 <p>This form of the <tt>if</tt> statement effectively combines together a call
401 to <tt>isa<></tt> and a call to <tt>cast<></tt> into one
402 statement, which is very convenient.</p>
404 <p>Note that the <tt>dyn_cast<></tt> operator, like C++'s
405 <tt>dynamic_cast<></tt> or Java's <tt>instanceof</tt> operator, can be
406 abused. In particular, you should not use big chained <tt>if/then/else</tt>
407 blocks to check for lots of different variants of classes. If you find
408 yourself wanting to do this, it is much cleaner and more efficient to use the
409 <tt>InstVisitor</tt> class to dispatch over the instruction type directly.</p>
413 <dt><tt>cast_or_null<></tt>: </dt>
415 <dd><p>The <tt>cast_or_null<></tt> operator works just like the
416 <tt>cast<></tt> operator, except that it allows for a null pointer as an
417 argument (which it then propagates). This can sometimes be useful, allowing
418 you to combine several null checks into one.</p></dd>
420 <dt><tt>dyn_cast_or_null<></tt>: </dt>
422 <dd><p>The <tt>dyn_cast_or_null<></tt> operator works just like the
423 <tt>dyn_cast<></tt> operator, except that it allows for a null pointer
424 as an argument (which it then propagates). This can sometimes be useful,
425 allowing you to combine several null checks into one.</p></dd>
429 <p>These five templates can be used with any classes, whether they have a
430 v-table or not. To add support for these templates, you simply need to add
431 <tt>classof</tt> static methods to the class you are interested casting
432 to. Describing this is currently outside the scope of this document, but there
433 are lots of examples in the LLVM source base.</p>
438 <!-- ======================================================================= -->
439 <div class="doc_subsection">
440 <a name="string_apis">Passing strings (the <tt>StringRef</tt>
441 and <tt>Twine</tt> classes)</a>
444 <div class="doc_text">
446 <p>Although LLVM generally does not do much string manipulation, we do have
447 several important APIs which take strings. Two important examples are the
448 Value class -- which has names for instructions, functions, etc. -- and the
449 StringMap class which is used extensively in LLVM and Clang.</p>
451 <p>These are generic classes, and they need to be able to accept strings which
452 may have embedded null characters. Therefore, they cannot simply take
453 a <tt>const char *</tt>, and taking a <tt>const std::string&</tt> requires
454 clients to perform a heap allocation which is usually unnecessary. Instead,
455 many LLVM APIs use a <tt>const StringRef&</tt> or a <tt>const
456 Twine&</tt> for passing strings efficiently.</p>
460 <!-- _______________________________________________________________________ -->
461 <div class="doc_subsubsection">
462 <a name="StringRef">The <tt>StringRef</tt> class</a>
465 <div class="doc_text">
467 <p>The <tt>StringRef</tt> data type represents a reference to a constant string
468 (a character array and a length) and supports the common operations available
469 on <tt>std:string</tt>, but does not require heap allocation.</p>
471 <p>It can be implicitly constructed using a C style null-terminated string,
472 an <tt>std::string</tt>, or explicitly with a character pointer and length.
473 For example, the <tt>StringRef</tt> find function is declared as:</p>
475 <div class="doc_code">
476 iterator find(const StringRef &Key);
479 <p>and clients can call it using any one of:</p>
481 <div class="doc_code">
483 Map.find("foo"); <i>// Lookup "foo"</i>
484 Map.find(std::string("bar")); <i>// Lookup "bar"</i>
485 Map.find(StringRef("\0baz", 4)); <i>// Lookup "\0baz"</i>
489 <p>Similarly, APIs which need to return a string may return a <tt>StringRef</tt>
490 instance, which can be used directly or converted to an <tt>std::string</tt>
491 using the <tt>str</tt> member function. See
492 "<tt><a href="/doxygen/classllvm_1_1StringRef_8h-source.html">llvm/ADT/StringRef.h</a></tt>"
493 for more information.</p>
495 <p>You should rarely use the <tt>StringRef</tt> class directly, because it contains
496 pointers to external memory it is not generally safe to store an instance of the
497 class (unless you know that the external storage will not be freed).</p>
501 <!-- _______________________________________________________________________ -->
502 <div class="doc_subsubsection">
503 <a name="Twine">The <tt>Twine</tt> class</a>
506 <div class="doc_text">
508 <p>The <tt>Twine</tt> class is an efficient way for APIs to accept concatenated
509 strings. For example, a common LLVM paradigm is to name one instruction based on
510 the name of another instruction with a suffix, for example:</p>
512 <div class="doc_code">
514 New = CmpInst::Create(<i>...</i>, SO->getName() + ".cmp");
518 <p>The <tt>Twine</tt> class is effectively a
519 lightweight <a href="http://en.wikipedia.org/wiki/Rope_(computer_science)">rope</a>
520 which points to temporary (stack allocated) objects. Twines can be implicitly
521 constructed as the result of the plus operator applied to strings (i.e., a C
522 strings, an <tt>std::string</tt>, or a <tt>StringRef</tt>). The twine delays the
523 actual concatentation of strings until it is actually required, at which point
524 it can be efficiently rendered directly into a character array. This avoids
525 unnecessary heap allocation involved in constructing the temporary results of
526 string concatenation. See
527 "<tt><a href="/doxygen/classllvm_1_1Twine_8h-source.html">llvm/ADT/Twine.h</a></tt>"
528 for more information.</p></tt>
530 <p>As with a <tt>StringRef</tt>, <tt>Twine</tt> objects point to external memory
531 and should almost never be stored or mentioned directly. They are intended
532 solely for use when defining a function which should be able to efficiently
533 accept concatenated strings.</p>
538 <!-- ======================================================================= -->
539 <div class="doc_subsection">
540 <a name="DEBUG">The <tt>DEBUG()</tt> macro and <tt>-debug</tt> option</a>
543 <div class="doc_text">
545 <p>Often when working on your pass you will put a bunch of debugging printouts
546 and other code into your pass. After you get it working, you want to remove
547 it, but you may need it again in the future (to work out new bugs that you run
550 <p> Naturally, because of this, you don't want to delete the debug printouts,
551 but you don't want them to always be noisy. A standard compromise is to comment
552 them out, allowing you to enable them if you need them in the future.</p>
554 <p>The "<tt><a href="/doxygen/Debug_8h-source.html">llvm/Support/Debug.h</a></tt>"
555 file provides a macro named <tt>DEBUG()</tt> that is a much nicer solution to
556 this problem. Basically, you can put arbitrary code into the argument of the
557 <tt>DEBUG</tt> macro, and it is only executed if '<tt>opt</tt>' (or any other
558 tool) is run with the '<tt>-debug</tt>' command line argument:</p>
560 <div class="doc_code">
562 DEBUG(errs() << "I am here!\n");
566 <p>Then you can run your pass like this:</p>
568 <div class="doc_code">
570 $ opt < a.bc > /dev/null -mypass
571 <i><no output></i>
572 $ opt < a.bc > /dev/null -mypass -debug
577 <p>Using the <tt>DEBUG()</tt> macro instead of a home-brewed solution allows you
578 to not have to create "yet another" command line option for the debug output for
579 your pass. Note that <tt>DEBUG()</tt> macros are disabled for optimized builds,
580 so they do not cause a performance impact at all (for the same reason, they
581 should also not contain side-effects!).</p>
583 <p>One additional nice thing about the <tt>DEBUG()</tt> macro is that you can
584 enable or disable it directly in gdb. Just use "<tt>set DebugFlag=0</tt>" or
585 "<tt>set DebugFlag=1</tt>" from the gdb if the program is running. If the
586 program hasn't been started yet, you can always just run it with
591 <!-- _______________________________________________________________________ -->
592 <div class="doc_subsubsection">
593 <a name="DEBUG_TYPE">Fine grained debug info with <tt>DEBUG_TYPE</tt> and
594 the <tt>-debug-only</tt> option</a>
597 <div class="doc_text">
599 <p>Sometimes you may find yourself in a situation where enabling <tt>-debug</tt>
600 just turns on <b>too much</b> information (such as when working on the code
601 generator). If you want to enable debug information with more fine-grained
602 control, you define the <tt>DEBUG_TYPE</tt> macro and the <tt>-debug</tt> only
603 option as follows:</p>
605 <div class="doc_code">
608 DEBUG(errs() << "No debug type\n");
609 #define DEBUG_TYPE "foo"
610 DEBUG(errs() << "'foo' debug type\n");
612 #define DEBUG_TYPE "bar"
613 DEBUG(errs() << "'bar' debug type\n"));
615 #define DEBUG_TYPE ""
616 DEBUG(errs() << "No debug type (2)\n");
620 <p>Then you can run your pass like this:</p>
622 <div class="doc_code">
624 $ opt < a.bc > /dev/null -mypass
625 <i><no output></i>
626 $ opt < a.bc > /dev/null -mypass -debug
631 $ opt < a.bc > /dev/null -mypass -debug-only=foo
633 $ opt < a.bc > /dev/null -mypass -debug-only=bar
638 <p>Of course, in practice, you should only set <tt>DEBUG_TYPE</tt> at the top of
639 a file, to specify the debug type for the entire module (if you do this before
640 you <tt>#include "llvm/Support/Debug.h"</tt>, you don't have to insert the ugly
641 <tt>#undef</tt>'s). Also, you should use names more meaningful than "foo" and
642 "bar", because there is no system in place to ensure that names do not
643 conflict. If two different modules use the same string, they will all be turned
644 on when the name is specified. This allows, for example, all debug information
645 for instruction scheduling to be enabled with <tt>-debug-type=InstrSched</tt>,
646 even if the source lives in multiple files.</p>
650 <!-- ======================================================================= -->
651 <div class="doc_subsection">
652 <a name="Statistic">The <tt>Statistic</tt> class & <tt>-stats</tt>
656 <div class="doc_text">
659 href="/doxygen/Statistic_8h-source.html">llvm/ADT/Statistic.h</a></tt>" file
660 provides a class named <tt>Statistic</tt> that is used as a unified way to
661 keep track of what the LLVM compiler is doing and how effective various
662 optimizations are. It is useful to see what optimizations are contributing to
663 making a particular program run faster.</p>
665 <p>Often you may run your pass on some big program, and you're interested to see
666 how many times it makes a certain transformation. Although you can do this with
667 hand inspection, or some ad-hoc method, this is a real pain and not very useful
668 for big programs. Using the <tt>Statistic</tt> class makes it very easy to
669 keep track of this information, and the calculated information is presented in a
670 uniform manner with the rest of the passes being executed.</p>
672 <p>There are many examples of <tt>Statistic</tt> uses, but the basics of using
673 it are as follows:</p>
676 <li><p>Define your statistic like this:</p>
678 <div class="doc_code">
680 #define <a href="#DEBUG_TYPE">DEBUG_TYPE</a> "mypassname" <i>// This goes before any #includes.</i>
681 STATISTIC(NumXForms, "The # of times I did stuff");
685 <p>The <tt>STATISTIC</tt> macro defines a static variable, whose name is
686 specified by the first argument. The pass name is taken from the DEBUG_TYPE
687 macro, and the description is taken from the second argument. The variable
688 defined ("NumXForms" in this case) acts like an unsigned integer.</p></li>
690 <li><p>Whenever you make a transformation, bump the counter:</p>
692 <div class="doc_code">
694 ++NumXForms; // <i>I did stuff!</i>
701 <p>That's all you have to do. To get '<tt>opt</tt>' to print out the
702 statistics gathered, use the '<tt>-stats</tt>' option:</p>
704 <div class="doc_code">
706 $ opt -stats -mypassname < program.bc > /dev/null
707 <i>... statistics output ...</i>
711 <p> When running <tt>opt</tt> on a C file from the SPEC benchmark
712 suite, it gives a report that looks like this:</p>
714 <div class="doc_code">
716 7646 bitcodewriter - Number of normal instructions
717 725 bitcodewriter - Number of oversized instructions
718 129996 bitcodewriter - Number of bitcode bytes written
719 2817 raise - Number of insts DCEd or constprop'd
720 3213 raise - Number of cast-of-self removed
721 5046 raise - Number of expression trees converted
722 75 raise - Number of other getelementptr's formed
723 138 raise - Number of load/store peepholes
724 42 deadtypeelim - Number of unused typenames removed from symtab
725 392 funcresolve - Number of varargs functions resolved
726 27 globaldce - Number of global variables removed
727 2 adce - Number of basic blocks removed
728 134 cee - Number of branches revectored
729 49 cee - Number of setcc instruction eliminated
730 532 gcse - Number of loads removed
731 2919 gcse - Number of instructions removed
732 86 indvars - Number of canonical indvars added
733 87 indvars - Number of aux indvars removed
734 25 instcombine - Number of dead inst eliminate
735 434 instcombine - Number of insts combined
736 248 licm - Number of load insts hoisted
737 1298 licm - Number of insts hoisted to a loop pre-header
738 3 licm - Number of insts hoisted to multiple loop preds (bad, no loop pre-header)
739 75 mem2reg - Number of alloca's promoted
740 1444 cfgsimplify - Number of blocks simplified
744 <p>Obviously, with so many optimizations, having a unified framework for this
745 stuff is very nice. Making your pass fit well into the framework makes it more
746 maintainable and useful.</p>
750 <!-- ======================================================================= -->
751 <div class="doc_subsection">
752 <a name="ViewGraph">Viewing graphs while debugging code</a>
755 <div class="doc_text">
757 <p>Several of the important data structures in LLVM are graphs: for example
758 CFGs made out of LLVM <a href="#BasicBlock">BasicBlock</a>s, CFGs made out of
759 LLVM <a href="CodeGenerator.html#machinebasicblock">MachineBasicBlock</a>s, and
760 <a href="CodeGenerator.html#selectiondag_intro">Instruction Selection
761 DAGs</a>. In many cases, while debugging various parts of the compiler, it is
762 nice to instantly visualize these graphs.</p>
764 <p>LLVM provides several callbacks that are available in a debug build to do
765 exactly that. If you call the <tt>Function::viewCFG()</tt> method, for example,
766 the current LLVM tool will pop up a window containing the CFG for the function
767 where each basic block is a node in the graph, and each node contains the
768 instructions in the block. Similarly, there also exists
769 <tt>Function::viewCFGOnly()</tt> (does not include the instructions), the
770 <tt>MachineFunction::viewCFG()</tt> and <tt>MachineFunction::viewCFGOnly()</tt>,
771 and the <tt>SelectionDAG::viewGraph()</tt> methods. Within GDB, for example,
772 you can usually use something like <tt>call DAG.viewGraph()</tt> to pop
773 up a window. Alternatively, you can sprinkle calls to these functions in your
774 code in places you want to debug.</p>
776 <p>Getting this to work requires a small amount of configuration. On Unix
777 systems with X11, install the <a href="http://www.graphviz.org">graphviz</a>
778 toolkit, and make sure 'dot' and 'gv' are in your path. If you are running on
779 Mac OS/X, download and install the Mac OS/X <a
780 href="http://www.pixelglow.com/graphviz/">Graphviz program</a>, and add
781 <tt>/Applications/Graphviz.app/Contents/MacOS/</tt> (or wherever you install
782 it) to your path. Once in your system and path are set up, rerun the LLVM
783 configure script and rebuild LLVM to enable this functionality.</p>
785 <p><tt>SelectionDAG</tt> has been extended to make it easier to locate
786 <i>interesting</i> nodes in large complex graphs. From gdb, if you
787 <tt>call DAG.setGraphColor(<i>node</i>, "<i>color</i>")</tt>, then the
788 next <tt>call DAG.viewGraph()</tt> would highlight the node in the
789 specified color (choices of colors can be found at <a
790 href="http://www.graphviz.org/doc/info/colors.html">colors</a>.) More
791 complex node attributes can be provided with <tt>call
792 DAG.setGraphAttrs(<i>node</i>, "<i>attributes</i>")</tt> (choices can be
793 found at <a href="http://www.graphviz.org/doc/info/attrs.html">Graph
794 Attributes</a>.) If you want to restart and clear all the current graph
795 attributes, then you can <tt>call DAG.clearGraphAttrs()</tt>. </p>
799 <!-- *********************************************************************** -->
800 <div class="doc_section">
801 <a name="datastructure">Picking the Right Data Structure for a Task</a>
803 <!-- *********************************************************************** -->
805 <div class="doc_text">
807 <p>LLVM has a plethora of data structures in the <tt>llvm/ADT/</tt> directory,
808 and we commonly use STL data structures. This section describes the trade-offs
809 you should consider when you pick one.</p>
812 The first step is a choose your own adventure: do you want a sequential
813 container, a set-like container, or a map-like container? The most important
814 thing when choosing a container is the algorithmic properties of how you plan to
815 access the container. Based on that, you should use:</p>
818 <li>a <a href="#ds_map">map-like</a> container if you need efficient look-up
819 of an value based on another value. Map-like containers also support
820 efficient queries for containment (whether a key is in the map). Map-like
821 containers generally do not support efficient reverse mapping (values to
822 keys). If you need that, use two maps. Some map-like containers also
823 support efficient iteration through the keys in sorted order. Map-like
824 containers are the most expensive sort, only use them if you need one of
825 these capabilities.</li>
827 <li>a <a href="#ds_set">set-like</a> container if you need to put a bunch of
828 stuff into a container that automatically eliminates duplicates. Some
829 set-like containers support efficient iteration through the elements in
830 sorted order. Set-like containers are more expensive than sequential
834 <li>a <a href="#ds_sequential">sequential</a> container provides
835 the most efficient way to add elements and keeps track of the order they are
836 added to the collection. They permit duplicates and support efficient
837 iteration, but do not support efficient look-up based on a key.
840 <li>a <a href="#ds_string">string</a> container is a specialized sequential
841 container or reference structure that is used for character or byte
844 <li>a <a href="#ds_bit">bit</a> container provides an efficient way to store and
845 perform set operations on sets of numeric id's, while automatically
846 eliminating duplicates. Bit containers require a maximum of 1 bit for each
847 identifier you want to store.
852 Once the proper category of container is determined, you can fine tune the
853 memory use, constant factors, and cache behaviors of access by intelligently
854 picking a member of the category. Note that constant factors and cache behavior
855 can be a big deal. If you have a vector that usually only contains a few
856 elements (but could contain many), for example, it's much better to use
857 <a href="#dss_smallvector">SmallVector</a> than <a href="#dss_vector">vector</a>
858 . Doing so avoids (relatively) expensive malloc/free calls, which dwarf the
859 cost of adding the elements to the container. </p>
863 <!-- ======================================================================= -->
864 <div class="doc_subsection">
865 <a name="ds_sequential">Sequential Containers (std::vector, std::list, etc)</a>
868 <div class="doc_text">
869 There are a variety of sequential containers available for you, based on your
870 needs. Pick the first in this section that will do what you want.
873 <!-- _______________________________________________________________________ -->
874 <div class="doc_subsubsection">
875 <a name="dss_fixedarrays">Fixed Size Arrays</a>
878 <div class="doc_text">
879 <p>Fixed size arrays are very simple and very fast. They are good if you know
880 exactly how many elements you have, or you have a (low) upper bound on how many
884 <!-- _______________________________________________________________________ -->
885 <div class="doc_subsubsection">
886 <a name="dss_heaparrays">Heap Allocated Arrays</a>
889 <div class="doc_text">
890 <p>Heap allocated arrays (new[] + delete[]) are also simple. They are good if
891 the number of elements is variable, if you know how many elements you will need
892 before the array is allocated, and if the array is usually large (if not,
893 consider a <a href="#dss_smallvector">SmallVector</a>). The cost of a heap
894 allocated array is the cost of the new/delete (aka malloc/free). Also note that
895 if you are allocating an array of a type with a constructor, the constructor and
896 destructors will be run for every element in the array (re-sizable vectors only
897 construct those elements actually used).</p>
900 <!-- _______________________________________________________________________ -->
901 <div class="doc_subsubsection">
902 <a name="dss_smallvector">"llvm/ADT/SmallVector.h"</a>
905 <div class="doc_text">
906 <p><tt>SmallVector<Type, N></tt> is a simple class that looks and smells
907 just like <tt>vector<Type></tt>:
908 it supports efficient iteration, lays out elements in memory order (so you can
909 do pointer arithmetic between elements), supports efficient push_back/pop_back
910 operations, supports efficient random access to its elements, etc.</p>
912 <p>The advantage of SmallVector is that it allocates space for
913 some number of elements (N) <b>in the object itself</b>. Because of this, if
914 the SmallVector is dynamically smaller than N, no malloc is performed. This can
915 be a big win in cases where the malloc/free call is far more expensive than the
916 code that fiddles around with the elements.</p>
918 <p>This is good for vectors that are "usually small" (e.g. the number of
919 predecessors/successors of a block is usually less than 8). On the other hand,
920 this makes the size of the SmallVector itself large, so you don't want to
921 allocate lots of them (doing so will waste a lot of space). As such,
922 SmallVectors are most useful when on the stack.</p>
924 <p>SmallVector also provides a nice portable and efficient replacement for
929 <!-- _______________________________________________________________________ -->
930 <div class="doc_subsubsection">
931 <a name="dss_vector"><vector></a>
934 <div class="doc_text">
936 std::vector is well loved and respected. It is useful when SmallVector isn't:
937 when the size of the vector is often large (thus the small optimization will
938 rarely be a benefit) or if you will be allocating many instances of the vector
939 itself (which would waste space for elements that aren't in the container).
940 vector is also useful when interfacing with code that expects vectors :).
943 <p>One worthwhile note about std::vector: avoid code like this:</p>
945 <div class="doc_code">
948 std::vector<foo> V;
954 <p>Instead, write this as:</p>
956 <div class="doc_code">
958 std::vector<foo> V;
966 <p>Doing so will save (at least) one heap allocation and free per iteration of
971 <!-- _______________________________________________________________________ -->
972 <div class="doc_subsubsection">
973 <a name="dss_deque"><deque></a>
976 <div class="doc_text">
977 <p>std::deque is, in some senses, a generalized version of std::vector. Like
978 std::vector, it provides constant time random access and other similar
979 properties, but it also provides efficient access to the front of the list. It
980 does not guarantee continuity of elements within memory.</p>
982 <p>In exchange for this extra flexibility, std::deque has significantly higher
983 constant factor costs than std::vector. If possible, use std::vector or
984 something cheaper.</p>
987 <!-- _______________________________________________________________________ -->
988 <div class="doc_subsubsection">
989 <a name="dss_list"><list></a>
992 <div class="doc_text">
993 <p>std::list is an extremely inefficient class that is rarely useful.
994 It performs a heap allocation for every element inserted into it, thus having an
995 extremely high constant factor, particularly for small data types. std::list
996 also only supports bidirectional iteration, not random access iteration.</p>
998 <p>In exchange for this high cost, std::list supports efficient access to both
999 ends of the list (like std::deque, but unlike std::vector or SmallVector). In
1000 addition, the iterator invalidation characteristics of std::list are stronger
1001 than that of a vector class: inserting or removing an element into the list does
1002 not invalidate iterator or pointers to other elements in the list.</p>
1005 <!-- _______________________________________________________________________ -->
1006 <div class="doc_subsubsection">
1007 <a name="dss_ilist">llvm/ADT/ilist.h</a>
1010 <div class="doc_text">
1011 <p><tt>ilist<T></tt> implements an 'intrusive' doubly-linked list. It is
1012 intrusive, because it requires the element to store and provide access to the
1013 prev/next pointers for the list.</p>
1015 <p><tt>ilist</tt> has the same drawbacks as <tt>std::list</tt>, and additionally
1016 requires an <tt>ilist_traits</tt> implementation for the element type, but it
1017 provides some novel characteristics. In particular, it can efficiently store
1018 polymorphic objects, the traits class is informed when an element is inserted or
1019 removed from the list, and <tt>ilist</tt>s are guaranteed to support a
1020 constant-time splice operation.</p>
1022 <p>These properties are exactly what we want for things like
1023 <tt>Instruction</tt>s and basic blocks, which is why these are implemented with
1024 <tt>ilist</tt>s.</p>
1026 Related classes of interest are explained in the following subsections:
1028 <li><a href="#dss_ilist_traits">ilist_traits</a></li>
1029 <li><a href="#dss_iplist">iplist</a></li>
1030 <li><a href="#dss_ilist_node">llvm/ADT/ilist_node.h</a></li>
1031 <li><a href="#dss_ilist_sentinel">Sentinels</a></li>
1035 <!-- _______________________________________________________________________ -->
1036 <div class="doc_subsubsection">
1037 <a name="dss_ilist_traits">ilist_traits</a>
1040 <div class="doc_text">
1041 <p><tt>ilist_traits<T></tt> is <tt>ilist<T></tt>'s customization
1042 mechanism. <tt>iplist<T></tt> (and consequently <tt>ilist<T></tt>)
1043 publicly derive from this traits class.</p>
1046 <!-- _______________________________________________________________________ -->
1047 <div class="doc_subsubsection">
1048 <a name="dss_iplist">iplist</a>
1051 <div class="doc_text">
1052 <p><tt>iplist<T></tt> is <tt>ilist<T></tt>'s base and as such
1053 supports a slightly narrower interface. Notably, inserters from
1054 <tt>T&</tt> are absent.</p>
1056 <p><tt>ilist_traits<T></tt> is a public base of this class and can be
1057 used for a wide variety of customizations.</p>
1060 <!-- _______________________________________________________________________ -->
1061 <div class="doc_subsubsection">
1062 <a name="dss_ilist_node">llvm/ADT/ilist_node.h</a>
1065 <div class="doc_text">
1066 <p><tt>ilist_node<T></tt> implements a the forward and backward links
1067 that are expected by the <tt>ilist<T></tt> (and analogous containers)
1068 in the default manner.</p>
1070 <p><tt>ilist_node<T></tt>s are meant to be embedded in the node type
1071 <tt>T</tt>, usually <tt>T</tt> publicly derives from
1072 <tt>ilist_node<T></tt>.</p>
1075 <!-- _______________________________________________________________________ -->
1076 <div class="doc_subsubsection">
1077 <a name="dss_ilist_sentinel">Sentinels</a>
1080 <div class="doc_text">
1081 <p><tt>ilist</tt>s have another speciality that must be considered. To be a good
1082 citizen in the C++ ecosystem, it needs to support the standard container
1083 operations, such as <tt>begin</tt> and <tt>end</tt> iterators, etc. Also, the
1084 <tt>operator--</tt> must work correctly on the <tt>end</tt> iterator in the
1085 case of non-empty <tt>ilist</tt>s.</p>
1087 <p>The only sensible solution to this problem is to allocate a so-called
1088 <i>sentinel</i> along with the intrusive list, which serves as the <tt>end</tt>
1089 iterator, providing the back-link to the last element. However conforming to the
1090 C++ convention it is illegal to <tt>operator++</tt> beyond the sentinel and it
1091 also must not be dereferenced.</p>
1093 <p>These constraints allow for some implementation freedom to the <tt>ilist</tt>
1094 how to allocate and store the sentinel. The corresponding policy is dictated
1095 by <tt>ilist_traits<T></tt>. By default a <tt>T</tt> gets heap-allocated
1096 whenever the need for a sentinel arises.</p>
1098 <p>While the default policy is sufficient in most cases, it may break down when
1099 <tt>T</tt> does not provide a default constructor. Also, in the case of many
1100 instances of <tt>ilist</tt>s, the memory overhead of the associated sentinels
1101 is wasted. To alleviate the situation with numerous and voluminous
1102 <tt>T</tt>-sentinels, sometimes a trick is employed, leading to <i>ghostly
1105 <p>Ghostly sentinels are obtained by specially-crafted <tt>ilist_traits<T></tt>
1106 which superpose the sentinel with the <tt>ilist</tt> instance in memory. Pointer
1107 arithmetic is used to obtain the sentinel, which is relative to the
1108 <tt>ilist</tt>'s <tt>this</tt> pointer. The <tt>ilist</tt> is augmented by an
1109 extra pointer, which serves as the back-link of the sentinel. This is the only
1110 field in the ghostly sentinel which can be legally accessed.</p>
1113 <!-- _______________________________________________________________________ -->
1114 <div class="doc_subsubsection">
1115 <a name="dss_other">Other Sequential Container options</a>
1118 <div class="doc_text">
1119 <p>Other STL containers are available, such as std::string.</p>
1121 <p>There are also various STL adapter classes such as std::queue,
1122 std::priority_queue, std::stack, etc. These provide simplified access to an
1123 underlying container but don't affect the cost of the container itself.</p>
1128 <!-- ======================================================================= -->
1129 <div class="doc_subsection">
1130 <a name="ds_set">Set-Like Containers (std::set, SmallSet, SetVector, etc)</a>
1133 <div class="doc_text">
1135 <p>Set-like containers are useful when you need to canonicalize multiple values
1136 into a single representation. There are several different choices for how to do
1137 this, providing various trade-offs.</p>
1142 <!-- _______________________________________________________________________ -->
1143 <div class="doc_subsubsection">
1144 <a name="dss_sortedvectorset">A sorted 'vector'</a>
1147 <div class="doc_text">
1149 <p>If you intend to insert a lot of elements, then do a lot of queries, a
1150 great approach is to use a vector (or other sequential container) with
1151 std::sort+std::unique to remove duplicates. This approach works really well if
1152 your usage pattern has these two distinct phases (insert then query), and can be
1153 coupled with a good choice of <a href="#ds_sequential">sequential container</a>.
1157 This combination provides the several nice properties: the result data is
1158 contiguous in memory (good for cache locality), has few allocations, is easy to
1159 address (iterators in the final vector are just indices or pointers), and can be
1160 efficiently queried with a standard binary or radix search.</p>
1164 <!-- _______________________________________________________________________ -->
1165 <div class="doc_subsubsection">
1166 <a name="dss_smallset">"llvm/ADT/SmallSet.h"</a>
1169 <div class="doc_text">
1171 <p>If you have a set-like data structure that is usually small and whose elements
1172 are reasonably small, a <tt>SmallSet<Type, N></tt> is a good choice. This set
1173 has space for N elements in place (thus, if the set is dynamically smaller than
1174 N, no malloc traffic is required) and accesses them with a simple linear search.
1175 When the set grows beyond 'N' elements, it allocates a more expensive representation that
1176 guarantees efficient access (for most types, it falls back to std::set, but for
1177 pointers it uses something far better, <a
1178 href="#dss_smallptrset">SmallPtrSet</a>).</p>
1180 <p>The magic of this class is that it handles small sets extremely efficiently,
1181 but gracefully handles extremely large sets without loss of efficiency. The
1182 drawback is that the interface is quite small: it supports insertion, queries
1183 and erasing, but does not support iteration.</p>
1187 <!-- _______________________________________________________________________ -->
1188 <div class="doc_subsubsection">
1189 <a name="dss_smallptrset">"llvm/ADT/SmallPtrSet.h"</a>
1192 <div class="doc_text">
1194 <p>SmallPtrSet has all the advantages of SmallSet (and a SmallSet of pointers is
1195 transparently implemented with a SmallPtrSet), but also supports iterators. If
1196 more than 'N' insertions are performed, a single quadratically
1197 probed hash table is allocated and grows as needed, providing extremely
1198 efficient access (constant time insertion/deleting/queries with low constant
1199 factors) and is very stingy with malloc traffic.</p>
1201 <p>Note that, unlike std::set, the iterators of SmallPtrSet are invalidated
1202 whenever an insertion occurs. Also, the values visited by the iterators are not
1203 visited in sorted order.</p>
1207 <!-- _______________________________________________________________________ -->
1208 <div class="doc_subsubsection">
1209 <a name="dss_denseset">"llvm/ADT/DenseSet.h"</a>
1212 <div class="doc_text">
1215 DenseSet is a simple quadratically probed hash table. It excels at supporting
1216 small values: it uses a single allocation to hold all of the pairs that
1217 are currently inserted in the set. DenseSet is a great way to unique small
1218 values that are not simple pointers (use <a
1219 href="#dss_smallptrset">SmallPtrSet</a> for pointers). Note that DenseSet has
1220 the same requirements for the value type that <a
1221 href="#dss_densemap">DenseMap</a> has.
1226 <!-- _______________________________________________________________________ -->
1227 <div class="doc_subsubsection">
1228 <a name="dss_FoldingSet">"llvm/ADT/FoldingSet.h"</a>
1231 <div class="doc_text">
1234 FoldingSet is an aggregate class that is really good at uniquing
1235 expensive-to-create or polymorphic objects. It is a combination of a chained
1236 hash table with intrusive links (uniqued objects are required to inherit from
1237 FoldingSetNode) that uses <a href="#dss_smallvector">SmallVector</a> as part of
1240 <p>Consider a case where you want to implement a "getOrCreateFoo" method for
1241 a complex object (for example, a node in the code generator). The client has a
1242 description of *what* it wants to generate (it knows the opcode and all the
1243 operands), but we don't want to 'new' a node, then try inserting it into a set
1244 only to find out it already exists, at which point we would have to delete it
1245 and return the node that already exists.
1248 <p>To support this style of client, FoldingSet perform a query with a
1249 FoldingSetNodeID (which wraps SmallVector) that can be used to describe the
1250 element that we want to query for. The query either returns the element
1251 matching the ID or it returns an opaque ID that indicates where insertion should
1252 take place. Construction of the ID usually does not require heap traffic.</p>
1254 <p>Because FoldingSet uses intrusive links, it can support polymorphic objects
1255 in the set (for example, you can have SDNode instances mixed with LoadSDNodes).
1256 Because the elements are individually allocated, pointers to the elements are
1257 stable: inserting or removing elements does not invalidate any pointers to other
1263 <!-- _______________________________________________________________________ -->
1264 <div class="doc_subsubsection">
1265 <a name="dss_set"><set></a>
1268 <div class="doc_text">
1270 <p><tt>std::set</tt> is a reasonable all-around set class, which is decent at
1271 many things but great at nothing. std::set allocates memory for each element
1272 inserted (thus it is very malloc intensive) and typically stores three pointers
1273 per element in the set (thus adding a large amount of per-element space
1274 overhead). It offers guaranteed log(n) performance, which is not particularly
1275 fast from a complexity standpoint (particularly if the elements of the set are
1276 expensive to compare, like strings), and has extremely high constant factors for
1277 lookup, insertion and removal.</p>
1279 <p>The advantages of std::set are that its iterators are stable (deleting or
1280 inserting an element from the set does not affect iterators or pointers to other
1281 elements) and that iteration over the set is guaranteed to be in sorted order.
1282 If the elements in the set are large, then the relative overhead of the pointers
1283 and malloc traffic is not a big deal, but if the elements of the set are small,
1284 std::set is almost never a good choice.</p>
1288 <!-- _______________________________________________________________________ -->
1289 <div class="doc_subsubsection">
1290 <a name="dss_setvector">"llvm/ADT/SetVector.h"</a>
1293 <div class="doc_text">
1294 <p>LLVM's SetVector<Type> is an adapter class that combines your choice of
1295 a set-like container along with a <a href="#ds_sequential">Sequential
1296 Container</a>. The important property
1297 that this provides is efficient insertion with uniquing (duplicate elements are
1298 ignored) with iteration support. It implements this by inserting elements into
1299 both a set-like container and the sequential container, using the set-like
1300 container for uniquing and the sequential container for iteration.
1303 <p>The difference between SetVector and other sets is that the order of
1304 iteration is guaranteed to match the order of insertion into the SetVector.
1305 This property is really important for things like sets of pointers. Because
1306 pointer values are non-deterministic (e.g. vary across runs of the program on
1307 different machines), iterating over the pointers in the set will
1308 not be in a well-defined order.</p>
1311 The drawback of SetVector is that it requires twice as much space as a normal
1312 set and has the sum of constant factors from the set-like container and the
1313 sequential container that it uses. Use it *only* if you need to iterate over
1314 the elements in a deterministic order. SetVector is also expensive to delete
1315 elements out of (linear time), unless you use it's "pop_back" method, which is
1319 <p>SetVector is an adapter class that defaults to using std::vector and std::set
1320 for the underlying containers, so it is quite expensive. However,
1321 <tt>"llvm/ADT/SetVector.h"</tt> also provides a SmallSetVector class, which
1322 defaults to using a SmallVector and SmallSet of a specified size. If you use
1323 this, and if your sets are dynamically smaller than N, you will save a lot of
1328 <!-- _______________________________________________________________________ -->
1329 <div class="doc_subsubsection">
1330 <a name="dss_uniquevector">"llvm/ADT/UniqueVector.h"</a>
1333 <div class="doc_text">
1336 UniqueVector is similar to <a href="#dss_setvector">SetVector</a>, but it
1337 retains a unique ID for each element inserted into the set. It internally
1338 contains a map and a vector, and it assigns a unique ID for each value inserted
1341 <p>UniqueVector is very expensive: its cost is the sum of the cost of
1342 maintaining both the map and vector, it has high complexity, high constant
1343 factors, and produces a lot of malloc traffic. It should be avoided.</p>
1348 <!-- _______________________________________________________________________ -->
1349 <div class="doc_subsubsection">
1350 <a name="dss_otherset">Other Set-Like Container Options</a>
1353 <div class="doc_text">
1356 The STL provides several other options, such as std::multiset and the various
1357 "hash_set" like containers (whether from C++ TR1 or from the SGI library). We
1358 never use hash_set and unordered_set because they are generally very expensive
1359 (each insertion requires a malloc) and very non-portable.
1362 <p>std::multiset is useful if you're not interested in elimination of
1363 duplicates, but has all the drawbacks of std::set. A sorted vector (where you
1364 don't delete duplicate entries) or some other approach is almost always
1369 <!-- ======================================================================= -->
1370 <div class="doc_subsection">
1371 <a name="ds_map">Map-Like Containers (std::map, DenseMap, etc)</a>
1374 <div class="doc_text">
1375 Map-like containers are useful when you want to associate data to a key. As
1376 usual, there are a lot of different ways to do this. :)
1379 <!-- _______________________________________________________________________ -->
1380 <div class="doc_subsubsection">
1381 <a name="dss_sortedvectormap">A sorted 'vector'</a>
1384 <div class="doc_text">
1387 If your usage pattern follows a strict insert-then-query approach, you can
1388 trivially use the same approach as <a href="#dss_sortedvectorset">sorted vectors
1389 for set-like containers</a>. The only difference is that your query function
1390 (which uses std::lower_bound to get efficient log(n) lookup) should only compare
1391 the key, not both the key and value. This yields the same advantages as sorted
1396 <!-- _______________________________________________________________________ -->
1397 <div class="doc_subsubsection">
1398 <a name="dss_stringmap">"llvm/ADT/StringMap.h"</a>
1401 <div class="doc_text">
1404 Strings are commonly used as keys in maps, and they are difficult to support
1405 efficiently: they are variable length, inefficient to hash and compare when
1406 long, expensive to copy, etc. StringMap is a specialized container designed to
1407 cope with these issues. It supports mapping an arbitrary range of bytes to an
1408 arbitrary other object.</p>
1410 <p>The StringMap implementation uses a quadratically-probed hash table, where
1411 the buckets store a pointer to the heap allocated entries (and some other
1412 stuff). The entries in the map must be heap allocated because the strings are
1413 variable length. The string data (key) and the element object (value) are
1414 stored in the same allocation with the string data immediately after the element
1415 object. This container guarantees the "<tt>(char*)(&Value+1)</tt>" points
1416 to the key string for a value.</p>
1418 <p>The StringMap is very fast for several reasons: quadratic probing is very
1419 cache efficient for lookups, the hash value of strings in buckets is not
1420 recomputed when lookup up an element, StringMap rarely has to touch the
1421 memory for unrelated objects when looking up a value (even when hash collisions
1422 happen), hash table growth does not recompute the hash values for strings
1423 already in the table, and each pair in the map is store in a single allocation
1424 (the string data is stored in the same allocation as the Value of a pair).</p>
1426 <p>StringMap also provides query methods that take byte ranges, so it only ever
1427 copies a string if a value is inserted into the table.</p>
1430 <!-- _______________________________________________________________________ -->
1431 <div class="doc_subsubsection">
1432 <a name="dss_indexedmap">"llvm/ADT/IndexedMap.h"</a>
1435 <div class="doc_text">
1437 IndexedMap is a specialized container for mapping small dense integers (or
1438 values that can be mapped to small dense integers) to some other type. It is
1439 internally implemented as a vector with a mapping function that maps the keys to
1440 the dense integer range.
1444 This is useful for cases like virtual registers in the LLVM code generator: they
1445 have a dense mapping that is offset by a compile-time constant (the first
1446 virtual register ID).</p>
1450 <!-- _______________________________________________________________________ -->
1451 <div class="doc_subsubsection">
1452 <a name="dss_densemap">"llvm/ADT/DenseMap.h"</a>
1455 <div class="doc_text">
1458 DenseMap is a simple quadratically probed hash table. It excels at supporting
1459 small keys and values: it uses a single allocation to hold all of the pairs that
1460 are currently inserted in the map. DenseMap is a great way to map pointers to
1461 pointers, or map other small types to each other.
1465 There are several aspects of DenseMap that you should be aware of, however. The
1466 iterators in a densemap are invalidated whenever an insertion occurs, unlike
1467 map. Also, because DenseMap allocates space for a large number of key/value
1468 pairs (it starts with 64 by default), it will waste a lot of space if your keys
1469 or values are large. Finally, you must implement a partial specialization of
1470 DenseMapInfo for the key that you want, if it isn't already supported. This
1471 is required to tell DenseMap about two special marker values (which can never be
1472 inserted into the map) that it needs internally.</p>
1476 <!-- _______________________________________________________________________ -->
1477 <div class="doc_subsubsection">
1478 <a name="dss_map"><map></a>
1481 <div class="doc_text">
1484 std::map has similar characteristics to <a href="#dss_set">std::set</a>: it uses
1485 a single allocation per pair inserted into the map, it offers log(n) lookup with
1486 an extremely large constant factor, imposes a space penalty of 3 pointers per
1487 pair in the map, etc.</p>
1489 <p>std::map is most useful when your keys or values are very large, if you need
1490 to iterate over the collection in sorted order, or if you need stable iterators
1491 into the map (i.e. they don't get invalidated if an insertion or deletion of
1492 another element takes place).</p>
1496 <!-- _______________________________________________________________________ -->
1497 <div class="doc_subsubsection">
1498 <a name="dss_othermap">Other Map-Like Container Options</a>
1501 <div class="doc_text">
1504 The STL provides several other options, such as std::multimap and the various
1505 "hash_map" like containers (whether from C++ TR1 or from the SGI library). We
1506 never use hash_set and unordered_set because they are generally very expensive
1507 (each insertion requires a malloc) and very non-portable.</p>
1509 <p>std::multimap is useful if you want to map a key to multiple values, but has
1510 all the drawbacks of std::map. A sorted vector or some other approach is almost
1515 <!-- ======================================================================= -->
1516 <div class="doc_subsection">
1517 <a name="ds_string">String-like containers</a>
1520 <div class="doc_text">
1523 TODO: const char* vs stringref vs smallstring vs std::string. Describe twine,
1524 xref to #string_apis.
1529 <!-- ======================================================================= -->
1530 <div class="doc_subsection">
1531 <a name="ds_bit">Bit storage containers (BitVector, SparseBitVector)</a>
1534 <div class="doc_text">
1535 <p>Unlike the other containers, there are only two bit storage containers, and
1536 choosing when to use each is relatively straightforward.</p>
1538 <p>One additional option is
1539 <tt>std::vector<bool></tt>: we discourage its use for two reasons 1) the
1540 implementation in many common compilers (e.g. commonly available versions of
1541 GCC) is extremely inefficient and 2) the C++ standards committee is likely to
1542 deprecate this container and/or change it significantly somehow. In any case,
1543 please don't use it.</p>
1546 <!-- _______________________________________________________________________ -->
1547 <div class="doc_subsubsection">
1548 <a name="dss_bitvector">BitVector</a>
1551 <div class="doc_text">
1552 <p> The BitVector container provides a fixed size set of bits for manipulation.
1553 It supports individual bit setting/testing, as well as set operations. The set
1554 operations take time O(size of bitvector), but operations are performed one word
1555 at a time, instead of one bit at a time. This makes the BitVector very fast for
1556 set operations compared to other containers. Use the BitVector when you expect
1557 the number of set bits to be high (IE a dense set).
1561 <!-- _______________________________________________________________________ -->
1562 <div class="doc_subsubsection">
1563 <a name="dss_sparsebitvector">SparseBitVector</a>
1566 <div class="doc_text">
1567 <p> The SparseBitVector container is much like BitVector, with one major
1568 difference: Only the bits that are set, are stored. This makes the
1569 SparseBitVector much more space efficient than BitVector when the set is sparse,
1570 as well as making set operations O(number of set bits) instead of O(size of
1571 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
1572 (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).
1576 <!-- *********************************************************************** -->
1577 <div class="doc_section">
1578 <a name="common">Helpful Hints for Common Operations</a>
1580 <!-- *********************************************************************** -->
1582 <div class="doc_text">
1584 <p>This section describes how to perform some very simple transformations of
1585 LLVM code. This is meant to give examples of common idioms used, showing the
1586 practical side of LLVM transformations. <p> Because this is a "how-to" section,
1587 you should also read about the main classes that you will be working with. The
1588 <a href="#coreclasses">Core LLVM Class Hierarchy Reference</a> contains details
1589 and descriptions of the main classes that you should know about.</p>
1593 <!-- NOTE: this section should be heavy on example code -->
1594 <!-- ======================================================================= -->
1595 <div class="doc_subsection">
1596 <a name="inspection">Basic Inspection and Traversal Routines</a>
1599 <div class="doc_text">
1601 <p>The LLVM compiler infrastructure have many different data structures that may
1602 be traversed. Following the example of the C++ standard template library, the
1603 techniques used to traverse these various data structures are all basically the
1604 same. For a enumerable sequence of values, the <tt>XXXbegin()</tt> function (or
1605 method) returns an iterator to the start of the sequence, the <tt>XXXend()</tt>
1606 function returns an iterator pointing to one past the last valid element of the
1607 sequence, and there is some <tt>XXXiterator</tt> data type that is common
1608 between the two operations.</p>
1610 <p>Because the pattern for iteration is common across many different aspects of
1611 the program representation, the standard template library algorithms may be used
1612 on them, and it is easier to remember how to iterate. First we show a few common
1613 examples of the data structures that need to be traversed. Other data
1614 structures are traversed in very similar ways.</p>
1618 <!-- _______________________________________________________________________ -->
1619 <div class="doc_subsubsection">
1620 <a name="iterate_function">Iterating over the </a><a
1621 href="#BasicBlock"><tt>BasicBlock</tt></a>s in a <a
1622 href="#Function"><tt>Function</tt></a>
1625 <div class="doc_text">
1627 <p>It's quite common to have a <tt>Function</tt> instance that you'd like to
1628 transform in some way; in particular, you'd like to manipulate its
1629 <tt>BasicBlock</tt>s. To facilitate this, you'll need to iterate over all of
1630 the <tt>BasicBlock</tt>s that constitute the <tt>Function</tt>. The following is
1631 an example that prints the name of a <tt>BasicBlock</tt> and the number of
1632 <tt>Instruction</tt>s it contains:</p>
1634 <div class="doc_code">
1636 // <i>func is a pointer to a Function instance</i>
1637 for (Function::iterator i = func->begin(), e = func->end(); i != e; ++i)
1638 // <i>Print out the name of the basic block if it has one, and then the</i>
1639 // <i>number of instructions that it contains</i>
1640 llvm::cerr << "Basic block (name=" << i->getName() << ") has "
1641 << i->size() << " instructions.\n";
1645 <p>Note that i can be used as if it were a pointer for the purposes of
1646 invoking member functions of the <tt>Instruction</tt> class. This is
1647 because the indirection operator is overloaded for the iterator
1648 classes. In the above code, the expression <tt>i->size()</tt> is
1649 exactly equivalent to <tt>(*i).size()</tt> just like you'd expect.</p>
1653 <!-- _______________________________________________________________________ -->
1654 <div class="doc_subsubsection">
1655 <a name="iterate_basicblock">Iterating over the </a><a
1656 href="#Instruction"><tt>Instruction</tt></a>s in a <a
1657 href="#BasicBlock"><tt>BasicBlock</tt></a>
1660 <div class="doc_text">
1662 <p>Just like when dealing with <tt>BasicBlock</tt>s in <tt>Function</tt>s, it's
1663 easy to iterate over the individual instructions that make up
1664 <tt>BasicBlock</tt>s. Here's a code snippet that prints out each instruction in
1665 a <tt>BasicBlock</tt>:</p>
1667 <div class="doc_code">
1669 // <i>blk is a pointer to a BasicBlock instance</i>
1670 for (BasicBlock::iterator i = blk->begin(), e = blk->end(); i != e; ++i)
1671 // <i>The next statement works since operator<<(ostream&,...)</i>
1672 // <i>is overloaded for Instruction&</i>
1673 llvm::cerr << *i << "\n";
1677 <p>However, this isn't really the best way to print out the contents of a
1678 <tt>BasicBlock</tt>! Since the ostream operators are overloaded for virtually
1679 anything you'll care about, you could have just invoked the print routine on the
1680 basic block itself: <tt>llvm::cerr << *blk << "\n";</tt>.</p>
1684 <!-- _______________________________________________________________________ -->
1685 <div class="doc_subsubsection">
1686 <a name="iterate_institer">Iterating over the </a><a
1687 href="#Instruction"><tt>Instruction</tt></a>s in a <a
1688 href="#Function"><tt>Function</tt></a>
1691 <div class="doc_text">
1693 <p>If you're finding that you commonly iterate over a <tt>Function</tt>'s
1694 <tt>BasicBlock</tt>s and then that <tt>BasicBlock</tt>'s <tt>Instruction</tt>s,
1695 <tt>InstIterator</tt> should be used instead. You'll need to include <a
1696 href="/doxygen/InstIterator_8h-source.html"><tt>llvm/Support/InstIterator.h</tt></a>,
1697 and then instantiate <tt>InstIterator</tt>s explicitly in your code. Here's a
1698 small example that shows how to dump all instructions in a function to the standard error stream:<p>
1700 <div class="doc_code">
1702 #include "<a href="/doxygen/InstIterator_8h-source.html">llvm/Support/InstIterator.h</a>"
1704 // <i>F is a pointer to a Function instance</i>
1705 for (inst_iterator I = inst_begin(F), E = inst_end(F); I != E; ++I)
1706 llvm::cerr << *I << "\n";
1710 <p>Easy, isn't it? You can also use <tt>InstIterator</tt>s to fill a
1711 work list with its initial contents. For example, if you wanted to
1712 initialize a work list to contain all instructions in a <tt>Function</tt>
1713 F, all you would need to do is something like:</p>
1715 <div class="doc_code">
1717 std::set<Instruction*> worklist;
1718 // or better yet, SmallPtrSet<Instruction*, 64> worklist;
1720 for (inst_iterator I = inst_begin(F), E = inst_end(F); I != E; ++I)
1721 worklist.insert(&*I);
1725 <p>The STL set <tt>worklist</tt> would now contain all instructions in the
1726 <tt>Function</tt> pointed to by F.</p>
1730 <!-- _______________________________________________________________________ -->
1731 <div class="doc_subsubsection">
1732 <a name="iterate_convert">Turning an iterator into a class pointer (and
1736 <div class="doc_text">
1738 <p>Sometimes, it'll be useful to grab a reference (or pointer) to a class
1739 instance when all you've got at hand is an iterator. Well, extracting
1740 a reference or a pointer from an iterator is very straight-forward.
1741 Assuming that <tt>i</tt> is a <tt>BasicBlock::iterator</tt> and <tt>j</tt>
1742 is a <tt>BasicBlock::const_iterator</tt>:</p>
1744 <div class="doc_code">
1746 Instruction& inst = *i; // <i>Grab reference to instruction reference</i>
1747 Instruction* pinst = &*i; // <i>Grab pointer to instruction reference</i>
1748 const Instruction& inst = *j;
1752 <p>However, the iterators you'll be working with in the LLVM framework are
1753 special: they will automatically convert to a ptr-to-instance type whenever they
1754 need to. Instead of dereferencing the iterator and then taking the address of
1755 the result, you can simply assign the iterator to the proper pointer type and
1756 you get the dereference and address-of operation as a result of the assignment
1757 (behind the scenes, this is a result of overloading casting mechanisms). Thus
1758 the last line of the last example,</p>
1760 <div class="doc_code">
1762 Instruction *pinst = &*i;
1766 <p>is semantically equivalent to</p>
1768 <div class="doc_code">
1770 Instruction *pinst = i;
1774 <p>It's also possible to turn a class pointer into the corresponding iterator,
1775 and this is a constant time operation (very efficient). The following code
1776 snippet illustrates use of the conversion constructors provided by LLVM
1777 iterators. By using these, you can explicitly grab the iterator of something
1778 without actually obtaining it via iteration over some structure:</p>
1780 <div class="doc_code">
1782 void printNextInstruction(Instruction* inst) {
1783 BasicBlock::iterator it(inst);
1784 ++it; // <i>After this line, it refers to the instruction after *inst</i>
1785 if (it != inst->getParent()->end()) llvm::cerr << *it << "\n";
1792 <!--_______________________________________________________________________-->
1793 <div class="doc_subsubsection">
1794 <a name="iterate_complex">Finding call sites: a slightly more complex
1798 <div class="doc_text">
1800 <p>Say that you're writing a FunctionPass and would like to count all the
1801 locations in the entire module (that is, across every <tt>Function</tt>) where a
1802 certain function (i.e., some <tt>Function</tt>*) is already in scope. As you'll
1803 learn later, you may want to use an <tt>InstVisitor</tt> to accomplish this in a
1804 much more straight-forward manner, but this example will allow us to explore how
1805 you'd do it if you didn't have <tt>InstVisitor</tt> around. In pseudo-code, this
1806 is what we want to do:</p>
1808 <div class="doc_code">
1810 initialize callCounter to zero
1811 for each Function f in the Module
1812 for each BasicBlock b in f
1813 for each Instruction i in b
1814 if (i is a CallInst and calls the given function)
1815 increment callCounter
1819 <p>And the actual code is (remember, because we're writing a
1820 <tt>FunctionPass</tt>, our <tt>FunctionPass</tt>-derived class simply has to
1821 override the <tt>runOnFunction</tt> method):</p>
1823 <div class="doc_code">
1825 Function* targetFunc = ...;
1827 class OurFunctionPass : public FunctionPass {
1829 OurFunctionPass(): callCounter(0) { }
1831 virtual runOnFunction(Function& F) {
1832 for (Function::iterator b = F.begin(), be = F.end(); b != be; ++b) {
1833 for (BasicBlock::iterator i = b->begin(), ie = b->end(); i != ie; ++i) {
1834 if (<a href="#CallInst">CallInst</a>* callInst = <a href="#isa">dyn_cast</a><<a
1835 href="#CallInst">CallInst</a>>(&*i)) {
1836 // <i>We know we've encountered a call instruction, so we</i>
1837 // <i>need to determine if it's a call to the</i>
1838 // <i>function pointed to by m_func or not.</i>
1839 if (callInst->getCalledFunction() == targetFunc)
1847 unsigned callCounter;
1854 <!--_______________________________________________________________________-->
1855 <div class="doc_subsubsection">
1856 <a name="calls_and_invokes">Treating calls and invokes the same way</a>
1859 <div class="doc_text">
1861 <p>You may have noticed that the previous example was a bit oversimplified in
1862 that it did not deal with call sites generated by 'invoke' instructions. In
1863 this, and in other situations, you may find that you want to treat
1864 <tt>CallInst</tt>s and <tt>InvokeInst</tt>s the same way, even though their
1865 most-specific common base class is <tt>Instruction</tt>, which includes lots of
1866 less closely-related things. For these cases, LLVM provides a handy wrapper
1868 href="http://llvm.org/doxygen/classllvm_1_1CallSite.html"><tt>CallSite</tt></a>.
1869 It is essentially a wrapper around an <tt>Instruction</tt> pointer, with some
1870 methods that provide functionality common to <tt>CallInst</tt>s and
1871 <tt>InvokeInst</tt>s.</p>
1873 <p>This class has "value semantics": it should be passed by value, not by
1874 reference and it should not be dynamically allocated or deallocated using
1875 <tt>operator new</tt> or <tt>operator delete</tt>. It is efficiently copyable,
1876 assignable and constructable, with costs equivalents to that of a bare pointer.
1877 If you look at its definition, it has only a single pointer member.</p>
1881 <!--_______________________________________________________________________-->
1882 <div class="doc_subsubsection">
1883 <a name="iterate_chains">Iterating over def-use & use-def chains</a>
1886 <div class="doc_text">
1888 <p>Frequently, we might have an instance of the <a
1889 href="/doxygen/classllvm_1_1Value.html">Value Class</a> and we want to
1890 determine which <tt>User</tt>s use the <tt>Value</tt>. The list of all
1891 <tt>User</tt>s of a particular <tt>Value</tt> is called a <i>def-use</i> chain.
1892 For example, let's say we have a <tt>Function*</tt> named <tt>F</tt> to a
1893 particular function <tt>foo</tt>. Finding all of the instructions that
1894 <i>use</i> <tt>foo</tt> is as simple as iterating over the <i>def-use</i> chain
1897 <div class="doc_code">
1901 for (Value::use_iterator i = F->use_begin(), e = F->use_end(); i != e; ++i)
1902 if (Instruction *Inst = dyn_cast<Instruction>(*i)) {
1903 llvm::cerr << "F is used in instruction:\n";
1904 llvm::cerr << *Inst << "\n";
1909 <p>Alternately, it's common to have an instance of the <a
1910 href="/doxygen/classllvm_1_1User.html">User Class</a> and need to know what
1911 <tt>Value</tt>s are used by it. The list of all <tt>Value</tt>s used by a
1912 <tt>User</tt> is known as a <i>use-def</i> chain. Instances of class
1913 <tt>Instruction</tt> are common <tt>User</tt>s, so we might want to iterate over
1914 all of the values that a particular instruction uses (that is, the operands of
1915 the particular <tt>Instruction</tt>):</p>
1917 <div class="doc_code">
1919 Instruction *pi = ...;
1921 for (User::op_iterator i = pi->op_begin(), e = pi->op_end(); i != e; ++i) {
1929 def-use chains ("finding all users of"): Value::use_begin/use_end
1930 use-def chains ("finding all values used"): User::op_begin/op_end [op=operand]
1935 <!--_______________________________________________________________________-->
1936 <div class="doc_subsubsection">
1937 <a name="iterate_preds">Iterating over predecessors &
1938 successors of blocks</a>
1941 <div class="doc_text">
1943 <p>Iterating over the predecessors and successors of a block is quite easy
1944 with the routines defined in <tt>"llvm/Support/CFG.h"</tt>. Just use code like
1945 this to iterate over all predecessors of BB:</p>
1947 <div class="doc_code">
1949 #include "llvm/Support/CFG.h"
1950 BasicBlock *BB = ...;
1952 for (pred_iterator PI = pred_begin(BB), E = pred_end(BB); PI != E; ++PI) {
1953 BasicBlock *Pred = *PI;
1959 <p>Similarly, to iterate over successors use
1960 succ_iterator/succ_begin/succ_end.</p>
1965 <!-- ======================================================================= -->
1966 <div class="doc_subsection">
1967 <a name="simplechanges">Making simple changes</a>
1970 <div class="doc_text">
1972 <p>There are some primitive transformation operations present in the LLVM
1973 infrastructure that are worth knowing about. When performing
1974 transformations, it's fairly common to manipulate the contents of basic
1975 blocks. This section describes some of the common methods for doing so
1976 and gives example code.</p>
1980 <!--_______________________________________________________________________-->
1981 <div class="doc_subsubsection">
1982 <a name="schanges_creating">Creating and inserting new
1983 <tt>Instruction</tt>s</a>
1986 <div class="doc_text">
1988 <p><i>Instantiating Instructions</i></p>
1990 <p>Creation of <tt>Instruction</tt>s is straight-forward: simply call the
1991 constructor for the kind of instruction to instantiate and provide the necessary
1992 parameters. For example, an <tt>AllocaInst</tt> only <i>requires</i> a
1993 (const-ptr-to) <tt>Type</tt>. Thus:</p>
1995 <div class="doc_code">
1997 AllocaInst* ai = new AllocaInst(Type::Int32Ty);
2001 <p>will create an <tt>AllocaInst</tt> instance that represents the allocation of
2002 one integer in the current stack frame, at run time. Each <tt>Instruction</tt>
2003 subclass is likely to have varying default parameters which change the semantics
2004 of the instruction, so refer to the <a
2005 href="/doxygen/classllvm_1_1Instruction.html">doxygen documentation for the subclass of
2006 Instruction</a> that you're interested in instantiating.</p>
2008 <p><i>Naming values</i></p>
2010 <p>It is very useful to name the values of instructions when you're able to, as
2011 this facilitates the debugging of your transformations. If you end up looking
2012 at generated LLVM machine code, you definitely want to have logical names
2013 associated with the results of instructions! By supplying a value for the
2014 <tt>Name</tt> (default) parameter of the <tt>Instruction</tt> constructor, you
2015 associate a logical name with the result of the instruction's execution at
2016 run time. For example, say that I'm writing a transformation that dynamically
2017 allocates space for an integer on the stack, and that integer is going to be
2018 used as some kind of index by some other code. To accomplish this, I place an
2019 <tt>AllocaInst</tt> at the first point in the first <tt>BasicBlock</tt> of some
2020 <tt>Function</tt>, and I'm intending to use it within the same
2021 <tt>Function</tt>. I might do:</p>
2023 <div class="doc_code">
2025 AllocaInst* pa = new AllocaInst(Type::Int32Ty, 0, "indexLoc");
2029 <p>where <tt>indexLoc</tt> is now the logical name of the instruction's
2030 execution value, which is a pointer to an integer on the run time stack.</p>
2032 <p><i>Inserting instructions</i></p>
2034 <p>There are essentially two ways to insert an <tt>Instruction</tt>
2035 into an existing sequence of instructions that form a <tt>BasicBlock</tt>:</p>
2038 <li>Insertion into an explicit instruction list
2040 <p>Given a <tt>BasicBlock* pb</tt>, an <tt>Instruction* pi</tt> within that
2041 <tt>BasicBlock</tt>, and a newly-created instruction we wish to insert
2042 before <tt>*pi</tt>, we do the following: </p>
2044 <div class="doc_code">
2046 BasicBlock *pb = ...;
2047 Instruction *pi = ...;
2048 Instruction *newInst = new Instruction(...);
2050 pb->getInstList().insert(pi, newInst); // <i>Inserts newInst before pi in pb</i>
2054 <p>Appending to the end of a <tt>BasicBlock</tt> is so common that
2055 the <tt>Instruction</tt> class and <tt>Instruction</tt>-derived
2056 classes provide constructors which take a pointer to a
2057 <tt>BasicBlock</tt> to be appended to. For example code that
2060 <div class="doc_code">
2062 BasicBlock *pb = ...;
2063 Instruction *newInst = new Instruction(...);
2065 pb->getInstList().push_back(newInst); // <i>Appends newInst to pb</i>
2071 <div class="doc_code">
2073 BasicBlock *pb = ...;
2074 Instruction *newInst = new Instruction(..., pb);
2078 <p>which is much cleaner, especially if you are creating
2079 long instruction streams.</p></li>
2081 <li>Insertion into an implicit instruction list
2083 <p><tt>Instruction</tt> instances that are already in <tt>BasicBlock</tt>s
2084 are implicitly associated with an existing instruction list: the instruction
2085 list of the enclosing basic block. Thus, we could have accomplished the same
2086 thing as the above code without being given a <tt>BasicBlock</tt> by doing:
2089 <div class="doc_code">
2091 Instruction *pi = ...;
2092 Instruction *newInst = new Instruction(...);
2094 pi->getParent()->getInstList().insert(pi, newInst);
2098 <p>In fact, this sequence of steps occurs so frequently that the
2099 <tt>Instruction</tt> class and <tt>Instruction</tt>-derived classes provide
2100 constructors which take (as a default parameter) a pointer to an
2101 <tt>Instruction</tt> which the newly-created <tt>Instruction</tt> should
2102 precede. That is, <tt>Instruction</tt> constructors are capable of
2103 inserting the newly-created instance into the <tt>BasicBlock</tt> of a
2104 provided instruction, immediately before that instruction. Using an
2105 <tt>Instruction</tt> constructor with a <tt>insertBefore</tt> (default)
2106 parameter, the above code becomes:</p>
2108 <div class="doc_code">
2110 Instruction* pi = ...;
2111 Instruction* newInst = new Instruction(..., pi);
2115 <p>which is much cleaner, especially if you're creating a lot of
2116 instructions and adding them to <tt>BasicBlock</tt>s.</p></li>
2121 <!--_______________________________________________________________________-->
2122 <div class="doc_subsubsection">
2123 <a name="schanges_deleting">Deleting <tt>Instruction</tt>s</a>
2126 <div class="doc_text">
2128 <p>Deleting an instruction from an existing sequence of instructions that form a
2129 <a href="#BasicBlock"><tt>BasicBlock</tt></a> is very straight-forward. First,
2130 you must have a pointer to the instruction that you wish to delete. Second, you
2131 need to obtain the pointer to that instruction's basic block. You use the
2132 pointer to the basic block to get its list of instructions and then use the
2133 erase function to remove your instruction. For example:</p>
2135 <div class="doc_code">
2137 <a href="#Instruction">Instruction</a> *I = .. ;
2138 I->eraseFromParent();
2144 <!--_______________________________________________________________________-->
2145 <div class="doc_subsubsection">
2146 <a name="schanges_replacing">Replacing an <tt>Instruction</tt> with another
2150 <div class="doc_text">
2152 <p><i>Replacing individual instructions</i></p>
2154 <p>Including "<a href="/doxygen/BasicBlockUtils_8h-source.html">llvm/Transforms/Utils/BasicBlockUtils.h</a>"
2155 permits use of two very useful replace functions: <tt>ReplaceInstWithValue</tt>
2156 and <tt>ReplaceInstWithInst</tt>.</p>
2158 <h4><a name="schanges_deleting">Deleting <tt>Instruction</tt>s</a></h4>
2161 <li><tt>ReplaceInstWithValue</tt>
2163 <p>This function replaces all uses of a given instruction with a value,
2164 and then removes the original instruction. The following example
2165 illustrates the replacement of the result of a particular
2166 <tt>AllocaInst</tt> that allocates memory for a single integer with a null
2167 pointer to an integer.</p>
2169 <div class="doc_code">
2171 AllocaInst* instToReplace = ...;
2172 BasicBlock::iterator ii(instToReplace);
2174 ReplaceInstWithValue(instToReplace->getParent()->getInstList(), ii,
2175 Constant::getNullValue(PointerType::getUnqual(Type::Int32Ty)));
2178 <li><tt>ReplaceInstWithInst</tt>
2180 <p>This function replaces a particular instruction with another
2181 instruction, inserting the new instruction into the basic block at the
2182 location where the old instruction was, and replacing any uses of the old
2183 instruction with the new instruction. The following example illustrates
2184 the replacement of one <tt>AllocaInst</tt> with another.</p>
2186 <div class="doc_code">
2188 AllocaInst* instToReplace = ...;
2189 BasicBlock::iterator ii(instToReplace);
2191 ReplaceInstWithInst(instToReplace->getParent()->getInstList(), ii,
2192 new AllocaInst(Type::Int32Ty, 0, "ptrToReplacedInt"));
2196 <p><i>Replacing multiple uses of <tt>User</tt>s and <tt>Value</tt>s</i></p>
2198 <p>You can use <tt>Value::replaceAllUsesWith</tt> and
2199 <tt>User::replaceUsesOfWith</tt> to change more than one use at a time. See the
2200 doxygen documentation for the <a href="/doxygen/classllvm_1_1Value.html">Value Class</a>
2201 and <a href="/doxygen/classllvm_1_1User.html">User Class</a>, respectively, for more
2204 <!-- Value::replaceAllUsesWith User::replaceUsesOfWith Point out:
2205 include/llvm/Transforms/Utils/ especially BasicBlockUtils.h with:
2206 ReplaceInstWithValue, ReplaceInstWithInst -->
2210 <!--_______________________________________________________________________-->
2211 <div class="doc_subsubsection">
2212 <a name="schanges_deletingGV">Deleting <tt>GlobalVariable</tt>s</a>
2215 <div class="doc_text">
2217 <p>Deleting a global variable from a module is just as easy as deleting an
2218 Instruction. First, you must have a pointer to the global variable that you wish
2219 to delete. You use this pointer to erase it from its parent, the module.
2222 <div class="doc_code">
2224 <a href="#GlobalVariable">GlobalVariable</a> *GV = .. ;
2226 GV->eraseFromParent();
2232 <!-- ======================================================================= -->
2233 <div class="doc_subsection">
2234 <a name="create_types">How to Create Types</a>
2237 <div class="doc_text">
2239 <p>In generating IR, you may need some complex types. If you know these types
2240 statically, you can use <tt>TypeBuilder<...>::get()</tt>, defined
2241 in <tt>llvm/Support/TypeBuilder.h</tt>, to retrieve them. <tt>TypeBuilder</tt>
2242 has two forms depending on whether you're building types for cross-compilation
2243 or native library use. <tt>TypeBuilder<T, true></tt> requires
2244 that <tt>T</tt> be independent of the host environment, meaning that it's built
2246 the <a href="/doxygen/namespacellvm_1_1types.html"><tt>llvm::types</tt></a>
2247 namespace and pointers, functions, arrays, etc. built of
2248 those. <tt>TypeBuilder<T, false></tt> additionally allows native C types
2249 whose size may depend on the host compiler. For example,</p>
2251 <div class="doc_code">
2253 FunctionType *ft = TypeBuilder<types::i<8>(types::i<32>*), true>::get();
2257 <p>is easier to read and write than the equivalent</p>
2259 <div class="doc_code">
2261 std::vector<const Type*> params;
2262 params.push_back(PointerType::getUnqual(Type::Int32Ty));
2263 FunctionType *ft = FunctionType::get(Type::Int8Ty, params, false);
2267 <p>See the <a href="/doxygen/TypeBuilder_8h-source.html#l00001">class
2268 comment</a> for more details.</p>
2272 <!-- *********************************************************************** -->
2273 <div class="doc_section">
2274 <a name="threading">Threads and LLVM</a>
2276 <!-- *********************************************************************** -->
2278 <div class="doc_text">
2280 This section describes the interaction of the LLVM APIs with multithreading,
2281 both on the part of client applications, and in the JIT, in the hosted
2286 Note that LLVM's support for multithreading is still relatively young. Up
2287 through version 2.5, the execution of threaded hosted applications was
2288 supported, but not threaded client access to the APIs. While this use case is
2289 now supported, clients <em>must</em> adhere to the guidelines specified below to
2290 ensure proper operation in multithreaded mode.
2294 Note that, on Unix-like platforms, LLVM requires the presence of GCC's atomic
2295 intrinsics in order to support threaded operation. If you need a
2296 multhreading-capable LLVM on a platform without a suitably modern system
2297 compiler, consider compiling LLVM and LLVM-GCC in single-threaded mode, and
2298 using the resultant compiler to build a copy of LLVM with multithreading
2303 <!-- ======================================================================= -->
2304 <div class="doc_subsection">
2305 <a name="startmultithreaded">Entering and Exiting Multithreaded Mode</a>
2308 <div class="doc_text">
2311 In order to properly protect its internal data structures while avoiding
2312 excessive locking overhead in the single-threaded case, the LLVM must intialize
2313 certain data structures necessary to provide guards around its internals. To do
2314 so, the client program must invoke <tt>llvm_start_multithreaded()</tt> before
2315 making any concurrent LLVM API calls. To subsequently tear down these
2316 structures, use the <tt>llvm_stop_multithreaded()</tt> call. You can also use
2317 the <tt>llvm_is_multithreaded()</tt> call to check the status of multithreaded
2322 Note that both of these calls must be made <em>in isolation</em>. That is to
2323 say that no other LLVM API calls may be executing at any time during the
2324 execution of <tt>llvm_start_multithreaded()</tt> or <tt>llvm_stop_multithreaded
2325 </tt>. It's is the client's responsibility to enforce this isolation.
2329 The return value of <tt>llvm_start_multithreaded()</tt> indicates the success or
2330 failure of the initialization. Failure typically indicates that your copy of
2331 LLVM was built without multithreading support, typically because GCC atomic
2332 intrinsics were not found in your system compiler. In this case, the LLVM API
2333 will not be safe for concurrent calls. However, it <em>will</em> be safe for
2334 hosting threaded applications in the JIT, though care must be taken to ensure
2335 that side exits and the like do not accidentally result in concurrent LLVM API
2340 <!-- ======================================================================= -->
2341 <div class="doc_subsection">
2342 <a name="shutdown">Ending Execution with <tt>llvm_shutdown()</tt></a>
2345 <div class="doc_text">
2347 When you are done using the LLVM APIs, you should call <tt>llvm_shutdown()</tt>
2348 to deallocate memory used for internal structures. This will also invoke
2349 <tt>llvm_stop_multithreaded()</tt> if LLVM is operating in multithreaded mode.
2350 As such, <tt>llvm_shutdown()</tt> requires the same isolation guarantees as
2351 <tt>llvm_stop_multithreaded()</tt>.
2355 Note that, if you use scope-based shutdown, you can use the
2356 <tt>llvm_shutdown_obj</tt> class, which calls <tt>llvm_shutdown()</tt> in its
2360 <!-- ======================================================================= -->
2361 <div class="doc_subsection">
2362 <a name="managedstatic">Lazy Initialization with <tt>ManagedStatic</tt></a>
2365 <div class="doc_text">
2367 <tt>ManagedStatic</tt> is a utility class in LLVM used to implement static
2368 initialization of static resources, such as the global type tables. Before the
2369 invocation of <tt>llvm_shutdown()</tt>, it implements a simple lazy
2370 initialization scheme. Once <tt>llvm_start_multithreaded()</tt> returns,
2371 however, it uses double-checked locking to implement thread-safe lazy
2376 Note that, because no other threads are allowed to issue LLVM API calls before
2377 <tt>llvm_start_multithreaded()</tt> returns, it is possible to have
2378 <tt>ManagedStatic</tt>s of <tt>llvm::sys::Mutex</tt>s.
2382 The <tt>llvm_acquire_global_lock()</tt> and <tt>llvm_release_global_lock</tt>
2383 APIs provide access to the global lock used to implement the double-checked
2384 locking for lazy initialization. These should only be used internally to LLVM,
2385 and only if you know what you're doing!
2389 <!-- *********************************************************************** -->
2390 <div class="doc_section">
2391 <a name="advanced">Advanced Topics</a>
2393 <!-- *********************************************************************** -->
2395 <div class="doc_text">
2397 This section describes some of the advanced or obscure API's that most clients
2398 do not need to be aware of. These API's tend manage the inner workings of the
2399 LLVM system, and only need to be accessed in unusual circumstances.
2403 <!-- ======================================================================= -->
2404 <div class="doc_subsection">
2405 <a name="TypeResolve">LLVM Type Resolution</a>
2408 <div class="doc_text">
2411 The LLVM type system has a very simple goal: allow clients to compare types for
2412 structural equality with a simple pointer comparison (aka a shallow compare).
2413 This goal makes clients much simpler and faster, and is used throughout the LLVM
2418 Unfortunately achieving this goal is not a simple matter. In particular,
2419 recursive types and late resolution of opaque types makes the situation very
2420 difficult to handle. Fortunately, for the most part, our implementation makes
2421 most clients able to be completely unaware of the nasty internal details. The
2422 primary case where clients are exposed to the inner workings of it are when
2423 building a recursive type. In addition to this case, the LLVM bitcode reader,
2424 assembly parser, and linker also have to be aware of the inner workings of this
2429 For our purposes below, we need three concepts. First, an "Opaque Type" is
2430 exactly as defined in the <a href="LangRef.html#t_opaque">language
2431 reference</a>. Second an "Abstract Type" is any type which includes an
2432 opaque type as part of its type graph (for example "<tt>{ opaque, i32 }</tt>").
2433 Third, a concrete type is a type that is not an abstract type (e.g. "<tt>{ i32,
2439 <!-- ______________________________________________________________________ -->
2440 <div class="doc_subsubsection">
2441 <a name="BuildRecType">Basic Recursive Type Construction</a>
2444 <div class="doc_text">
2447 Because the most common question is "how do I build a recursive type with LLVM",
2448 we answer it now and explain it as we go. Here we include enough to cause this
2449 to be emitted to an output .ll file:
2452 <div class="doc_code">
2454 %mylist = type { %mylist*, i32 }
2459 To build this, use the following LLVM APIs:
2462 <div class="doc_code">
2464 // <i>Create the initial outer struct</i>
2465 <a href="#PATypeHolder">PATypeHolder</a> StructTy = OpaqueType::get();
2466 std::vector<const Type*> Elts;
2467 Elts.push_back(PointerType::getUnqual(StructTy));
2468 Elts.push_back(Type::Int32Ty);
2469 StructType *NewSTy = StructType::get(Elts);
2471 // <i>At this point, NewSTy = "{ opaque*, i32 }". Tell VMCore that</i>
2472 // <i>the struct and the opaque type are actually the same.</i>
2473 cast<OpaqueType>(StructTy.get())-><a href="#refineAbstractTypeTo">refineAbstractTypeTo</a>(NewSTy);
2475 // <i>NewSTy is potentially invalidated, but StructTy (a <a href="#PATypeHolder">PATypeHolder</a>) is</i>
2476 // <i>kept up-to-date</i>
2477 NewSTy = cast<StructType>(StructTy.get());
2479 // <i>Add a name for the type to the module symbol table (optional)</i>
2480 MyModule->addTypeName("mylist", NewSTy);
2485 This code shows the basic approach used to build recursive types: build a
2486 non-recursive type using 'opaque', then use type unification to close the cycle.
2487 The type unification step is performed by the <tt><a
2488 href="#refineAbstractTypeTo">refineAbstractTypeTo</a></tt> method, which is
2489 described next. After that, we describe the <a
2490 href="#PATypeHolder">PATypeHolder class</a>.
2495 <!-- ______________________________________________________________________ -->
2496 <div class="doc_subsubsection">
2497 <a name="refineAbstractTypeTo">The <tt>refineAbstractTypeTo</tt> method</a>
2500 <div class="doc_text">
2502 The <tt>refineAbstractTypeTo</tt> method starts the type unification process.
2503 While this method is actually a member of the DerivedType class, it is most
2504 often used on OpaqueType instances. Type unification is actually a recursive
2505 process. After unification, types can become structurally isomorphic to
2506 existing types, and all duplicates are deleted (to preserve pointer equality).
2510 In the example above, the OpaqueType object is definitely deleted.
2511 Additionally, if there is an "{ \2*, i32}" type already created in the system,
2512 the pointer and struct type created are <b>also</b> deleted. Obviously whenever
2513 a type is deleted, any "Type*" pointers in the program are invalidated. As
2514 such, it is safest to avoid having <i>any</i> "Type*" pointers to abstract types
2515 live across a call to <tt>refineAbstractTypeTo</tt> (note that non-abstract
2516 types can never move or be deleted). To deal with this, the <a
2517 href="#PATypeHolder">PATypeHolder</a> class is used to maintain a stable
2518 reference to a possibly refined type, and the <a
2519 href="#AbstractTypeUser">AbstractTypeUser</a> class is used to update more
2520 complex datastructures.
2525 <!-- ______________________________________________________________________ -->
2526 <div class="doc_subsubsection">
2527 <a name="PATypeHolder">The PATypeHolder Class</a>
2530 <div class="doc_text">
2532 PATypeHolder is a form of a "smart pointer" for Type objects. When VMCore
2533 happily goes about nuking types that become isomorphic to existing types, it
2534 automatically updates all PATypeHolder objects to point to the new type. In the
2535 example above, this allows the code to maintain a pointer to the resultant
2536 resolved recursive type, even though the Type*'s are potentially invalidated.
2540 PATypeHolder is an extremely light-weight object that uses a lazy union-find
2541 implementation to update pointers. For example the pointer from a Value to its
2542 Type is maintained by PATypeHolder objects.
2547 <!-- ______________________________________________________________________ -->
2548 <div class="doc_subsubsection">
2549 <a name="AbstractTypeUser">The AbstractTypeUser Class</a>
2552 <div class="doc_text">
2555 Some data structures need more to perform more complex updates when types get
2556 resolved. To support this, a class can derive from the AbstractTypeUser class.
2558 allows it to get callbacks when certain types are resolved. To register to get
2559 callbacks for a particular type, the DerivedType::{add/remove}AbstractTypeUser
2560 methods can be called on a type. Note that these methods only work for <i>
2561 abstract</i> types. Concrete types (those that do not include any opaque
2562 objects) can never be refined.
2567 <!-- ======================================================================= -->
2568 <div class="doc_subsection">
2569 <a name="SymbolTable">The <tt>ValueSymbolTable</tt> and
2570 <tt>TypeSymbolTable</tt> classes</a>
2573 <div class="doc_text">
2574 <p>The <tt><a href="http://llvm.org/doxygen/classllvm_1_1ValueSymbolTable.html">
2575 ValueSymbolTable</a></tt> class provides a symbol table that the <a
2576 href="#Function"><tt>Function</tt></a> and <a href="#Module">
2577 <tt>Module</tt></a> classes use for naming value definitions. The symbol table
2578 can provide a name for any <a href="#Value"><tt>Value</tt></a>.
2579 The <tt><a href="http://llvm.org/doxygen/classllvm_1_1TypeSymbolTable.html">
2580 TypeSymbolTable</a></tt> class is used by the <tt>Module</tt> class to store
2581 names for types.</p>
2583 <p>Note that the <tt>SymbolTable</tt> class should not be directly accessed
2584 by most clients. It should only be used when iteration over the symbol table
2585 names themselves are required, which is very special purpose. Note that not
2587 <tt><a href="#Value">Value</a></tt>s have names, and those without names (i.e. they have
2588 an empty name) do not exist in the symbol table.
2591 <p>These symbol tables support iteration over the values/types in the symbol
2592 table with <tt>begin/end/iterator</tt> and supports querying to see if a
2593 specific name is in the symbol table (with <tt>lookup</tt>). The
2594 <tt>ValueSymbolTable</tt> class exposes no public mutator methods, instead,
2595 simply call <tt>setName</tt> on a value, which will autoinsert it into the
2596 appropriate symbol table. For types, use the Module::addTypeName method to
2597 insert entries into the symbol table.</p>
2603 <!-- ======================================================================= -->
2604 <div class="doc_subsection">
2605 <a name="UserLayout">The <tt>User</tt> and owned <tt>Use</tt> classes' memory layout</a>
2608 <div class="doc_text">
2609 <p>The <tt><a href="http://llvm.org/doxygen/classllvm_1_1User.html">
2610 User</a></tt> class provides a basis for expressing the ownership of <tt>User</tt>
2611 towards other <tt><a href="http://llvm.org/doxygen/classllvm_1_1Value.html">
2612 Value</a></tt>s. The <tt><a href="http://llvm.org/doxygen/classllvm_1_1Use.html">
2613 Use</a></tt> helper class is employed to do the bookkeeping and to facilitate <i>O(1)</i>
2614 addition and removal.</p>
2616 <!-- ______________________________________________________________________ -->
2617 <div class="doc_subsubsection">
2618 <a name="Use2User">Interaction and relationship between <tt>User</tt> and <tt>Use</tt> objects</a>
2621 <div class="doc_text">
2623 A subclass of <tt>User</tt> can choose between incorporating its <tt>Use</tt> objects
2624 or refer to them out-of-line by means of a pointer. A mixed variant
2625 (some <tt>Use</tt>s inline others hung off) is impractical and breaks the invariant
2626 that the <tt>Use</tt> objects belonging to the same <tt>User</tt> form a contiguous array.
2631 We have 2 different layouts in the <tt>User</tt> (sub)classes:
2634 The <tt>Use</tt> object(s) are inside (resp. at fixed offset) of the <tt>User</tt>
2635 object and there are a fixed number of them.</p>
2638 The <tt>Use</tt> object(s) are referenced by a pointer to an
2639 array from the <tt>User</tt> object and there may be a variable
2643 As of v2.4 each layout still possesses a direct pointer to the
2644 start of the array of <tt>Use</tt>s. Though not mandatory for layout a),
2645 we stick to this redundancy for the sake of simplicity.
2646 The <tt>User</tt> object also stores the number of <tt>Use</tt> objects it
2647 has. (Theoretically this information can also be calculated
2648 given the scheme presented below.)</p>
2650 Special forms of allocation operators (<tt>operator new</tt>)
2651 enforce the following memory layouts:</p>
2654 <li><p>Layout a) is modelled by prepending the <tt>User</tt> object by the <tt>Use[]</tt> array.</p>
2657 ...---.---.---.---.-------...
2658 | P | P | P | P | User
2659 '''---'---'---'---'-------'''
2662 <li><p>Layout b) is modelled by pointing at the <tt>Use[]</tt> array.</p>
2674 <i>(In the above figures '<tt>P</tt>' stands for the <tt>Use**</tt> that
2675 is stored in each <tt>Use</tt> object in the member <tt>Use::Prev</tt>)</i>
2677 <!-- ______________________________________________________________________ -->
2678 <div class="doc_subsubsection">
2679 <a name="Waymarking">The waymarking algorithm</a>
2682 <div class="doc_text">
2684 Since the <tt>Use</tt> objects are deprived of the direct (back)pointer to
2685 their <tt>User</tt> objects, there must be a fast and exact method to
2686 recover it. This is accomplished by the following scheme:</p>
2689 A bit-encoding in the 2 LSBits (least significant bits) of the <tt>Use::Prev</tt> allows to find the
2690 start of the <tt>User</tt> object:
2692 <li><tt>00</tt> —> binary digit 0</li>
2693 <li><tt>01</tt> —> binary digit 1</li>
2694 <li><tt>10</tt> —> stop and calculate (<tt>s</tt>)</li>
2695 <li><tt>11</tt> —> full stop (<tt>S</tt>)</li>
2698 Given a <tt>Use*</tt>, all we have to do is to walk till we get
2699 a stop and we either have a <tt>User</tt> immediately behind or
2700 we have to walk to the next stop picking up digits
2701 and calculating the offset:</p>
2703 .---.---.---.---.---.---.---.---.---.---.---.---.---.---.---.---.----------------
2704 | 1 | s | 1 | 0 | 1 | 0 | s | 1 | 1 | 0 | s | 1 | 1 | s | 1 | S | User (or User*)
2705 '---'---'---'---'---'---'---'---'---'---'---'---'---'---'---'---'----------------
2706 |+15 |+10 |+6 |+3 |+1
2709 | | |______________________>
2710 | |______________________________________>
2711 |__________________________________________________________>
2714 Only the significant number of bits need to be stored between the
2715 stops, so that the <i>worst case is 20 memory accesses</i> when there are
2716 1000 <tt>Use</tt> objects associated with a <tt>User</tt>.</p>
2718 <!-- ______________________________________________________________________ -->
2719 <div class="doc_subsubsection">
2720 <a name="ReferenceImpl">Reference implementation</a>
2723 <div class="doc_text">
2725 The following literate Haskell fragment demonstrates the concept:</p>
2728 <div class="doc_code">
2730 > import Test.QuickCheck
2732 > digits :: Int -> [Char] -> [Char]
2733 > digits 0 acc = '0' : acc
2734 > digits 1 acc = '1' : acc
2735 > digits n acc = digits (n `div` 2) $ digits (n `mod` 2) acc
2737 > dist :: Int -> [Char] -> [Char]
2740 > dist 1 acc = let r = dist 0 acc in 's' : digits (length r) r
2741 > dist n acc = dist (n - 1) $ dist 1 acc
2743 > takeLast n ss = reverse $ take n $ reverse ss
2745 > test = takeLast 40 $ dist 20 []
2750 Printing <test> gives: <tt>"1s100000s11010s10100s1111s1010s110s11s1S"</tt></p>
2752 The reverse algorithm computes the length of the string just by examining
2753 a certain prefix:</p>
2755 <div class="doc_code">
2757 > pref :: [Char] -> Int
2759 > pref ('s':'1':rest) = decode 2 1 rest
2760 > pref (_:rest) = 1 + pref rest
2762 > decode walk acc ('0':rest) = decode (walk + 1) (acc * 2) rest
2763 > decode walk acc ('1':rest) = decode (walk + 1) (acc * 2 + 1) rest
2764 > decode walk acc _ = walk + acc
2769 Now, as expected, printing <pref test> gives <tt>40</tt>.</p>
2771 We can <i>quickCheck</i> this with following property:</p>
2773 <div class="doc_code">
2775 > testcase = dist 2000 []
2776 > testcaseLength = length testcase
2778 > identityProp n = n > 0 && n <= testcaseLength ==> length arr == pref arr
2779 > where arr = takeLast n testcase
2784 As expected <quickCheck identityProp> gives:</p>
2787 *Main> quickCheck identityProp
2788 OK, passed 100 tests.
2791 Let's be a bit more exhaustive:</p>
2793 <div class="doc_code">
2796 > deepCheck p = check (defaultConfig { configMaxTest = 500 }) p
2801 And here is the result of <deepCheck identityProp>:</p>
2804 *Main> deepCheck identityProp
2805 OK, passed 500 tests.
2808 <!-- ______________________________________________________________________ -->
2809 <div class="doc_subsubsection">
2810 <a name="Tagging">Tagging considerations</a>
2814 To maintain the invariant that the 2 LSBits of each <tt>Use**</tt> in <tt>Use</tt>
2815 never change after being set up, setters of <tt>Use::Prev</tt> must re-tag the
2816 new <tt>Use**</tt> on every modification. Accordingly getters must strip the
2819 For layout b) instead of the <tt>User</tt> we find a pointer (<tt>User*</tt> with LSBit set).
2820 Following this pointer brings us to the <tt>User</tt>. A portable trick ensures
2821 that the first bytes of <tt>User</tt> (if interpreted as a pointer) never has
2822 the LSBit set. (Portability is relying on the fact that all known compilers place the
2823 <tt>vptr</tt> in the first word of the instances.)</p>
2827 <!-- *********************************************************************** -->
2828 <div class="doc_section">
2829 <a name="coreclasses">The Core LLVM Class Hierarchy Reference </a>
2831 <!-- *********************************************************************** -->
2833 <div class="doc_text">
2834 <p><tt>#include "<a href="/doxygen/Type_8h-source.html">llvm/Type.h</a>"</tt>
2835 <br>doxygen info: <a href="/doxygen/classllvm_1_1Type.html">Type Class</a></p>
2837 <p>The Core LLVM classes are the primary means of representing the program
2838 being inspected or transformed. The core LLVM classes are defined in
2839 header files in the <tt>include/llvm/</tt> directory, and implemented in
2840 the <tt>lib/VMCore</tt> directory.</p>
2844 <!-- ======================================================================= -->
2845 <div class="doc_subsection">
2846 <a name="Type">The <tt>Type</tt> class and Derived Types</a>
2849 <div class="doc_text">
2851 <p><tt>Type</tt> is a superclass of all type classes. Every <tt>Value</tt> has
2852 a <tt>Type</tt>. <tt>Type</tt> cannot be instantiated directly but only
2853 through its subclasses. Certain primitive types (<tt>VoidType</tt>,
2854 <tt>LabelType</tt>, <tt>FloatType</tt> and <tt>DoubleType</tt>) have hidden
2855 subclasses. They are hidden because they offer no useful functionality beyond
2856 what the <tt>Type</tt> class offers except to distinguish themselves from
2857 other subclasses of <tt>Type</tt>.</p>
2858 <p>All other types are subclasses of <tt>DerivedType</tt>. Types can be
2859 named, but this is not a requirement. There exists exactly
2860 one instance of a given shape at any one time. This allows type equality to
2861 be performed with address equality of the Type Instance. That is, given two
2862 <tt>Type*</tt> values, the types are identical if the pointers are identical.
2866 <!-- _______________________________________________________________________ -->
2867 <div class="doc_subsubsection">
2868 <a name="m_Type">Important Public Methods</a>
2871 <div class="doc_text">
2874 <li><tt>bool isInteger() const</tt>: Returns true for any integer type.</li>
2876 <li><tt>bool isFloatingPoint()</tt>: Return true if this is one of the two
2877 floating point types.</li>
2879 <li><tt>bool isAbstract()</tt>: Return true if the type is abstract (contains
2880 an OpaqueType anywhere in its definition).</li>
2882 <li><tt>bool isSized()</tt>: Return true if the type has known size. Things
2883 that don't have a size are abstract types, labels and void.</li>
2888 <!-- _______________________________________________________________________ -->
2889 <div class="doc_subsubsection">
2890 <a name="derivedtypes">Important Derived Types</a>
2892 <div class="doc_text">
2894 <dt><tt>IntegerType</tt></dt>
2895 <dd>Subclass of DerivedType that represents integer types of any bit width.
2896 Any bit width between <tt>IntegerType::MIN_INT_BITS</tt> (1) and
2897 <tt>IntegerType::MAX_INT_BITS</tt> (~8 million) can be represented.
2899 <li><tt>static const IntegerType* get(unsigned NumBits)</tt>: get an integer
2900 type of a specific bit width.</li>
2901 <li><tt>unsigned getBitWidth() const</tt>: Get the bit width of an integer
2905 <dt><tt>SequentialType</tt></dt>
2906 <dd>This is subclassed by ArrayType and PointerType
2908 <li><tt>const Type * getElementType() const</tt>: Returns the type of each
2909 of the elements in the sequential type. </li>
2912 <dt><tt>ArrayType</tt></dt>
2913 <dd>This is a subclass of SequentialType and defines the interface for array
2916 <li><tt>unsigned getNumElements() const</tt>: Returns the number of
2917 elements in the array. </li>
2920 <dt><tt>PointerType</tt></dt>
2921 <dd>Subclass of SequentialType for pointer types.</dd>
2922 <dt><tt>VectorType</tt></dt>
2923 <dd>Subclass of SequentialType for vector types. A
2924 vector type is similar to an ArrayType but is distinguished because it is
2925 a first class type wherease ArrayType is not. Vector types are used for
2926 vector operations and are usually small vectors of of an integer or floating
2928 <dt><tt>StructType</tt></dt>
2929 <dd>Subclass of DerivedTypes for struct types.</dd>
2930 <dt><tt><a name="FunctionType">FunctionType</a></tt></dt>
2931 <dd>Subclass of DerivedTypes for function types.
2933 <li><tt>bool isVarArg() const</tt>: Returns true if its a vararg
2935 <li><tt> const Type * getReturnType() const</tt>: Returns the
2936 return type of the function.</li>
2937 <li><tt>const Type * getParamType (unsigned i)</tt>: Returns
2938 the type of the ith parameter.</li>
2939 <li><tt> const unsigned getNumParams() const</tt>: Returns the
2940 number of formal parameters.</li>
2943 <dt><tt>OpaqueType</tt></dt>
2944 <dd>Sublcass of DerivedType for abstract types. This class
2945 defines no content and is used as a placeholder for some other type. Note
2946 that OpaqueType is used (temporarily) during type resolution for forward
2947 references of types. Once the referenced type is resolved, the OpaqueType
2948 is replaced with the actual type. OpaqueType can also be used for data
2949 abstraction. At link time opaque types can be resolved to actual types
2950 of the same name.</dd>
2956 <!-- ======================================================================= -->
2957 <div class="doc_subsection">
2958 <a name="Module">The <tt>Module</tt> class</a>
2961 <div class="doc_text">
2964 href="/doxygen/Module_8h-source.html">llvm/Module.h</a>"</tt><br> doxygen info:
2965 <a href="/doxygen/classllvm_1_1Module.html">Module Class</a></p>
2967 <p>The <tt>Module</tt> class represents the top level structure present in LLVM
2968 programs. An LLVM module is effectively either a translation unit of the
2969 original program or a combination of several translation units merged by the
2970 linker. The <tt>Module</tt> class keeps track of a list of <a
2971 href="#Function"><tt>Function</tt></a>s, a list of <a
2972 href="#GlobalVariable"><tt>GlobalVariable</tt></a>s, and a <a
2973 href="#SymbolTable"><tt>SymbolTable</tt></a>. Additionally, it contains a few
2974 helpful member functions that try to make common operations easy.</p>
2978 <!-- _______________________________________________________________________ -->
2979 <div class="doc_subsubsection">
2980 <a name="m_Module">Important Public Members of the <tt>Module</tt> class</a>
2983 <div class="doc_text">
2986 <li><tt>Module::Module(std::string name = "")</tt></li>
2989 <p>Constructing a <a href="#Module">Module</a> is easy. You can optionally
2990 provide a name for it (probably based on the name of the translation unit).</p>
2993 <li><tt>Module::iterator</tt> - Typedef for function list iterator<br>
2994 <tt>Module::const_iterator</tt> - Typedef for const_iterator.<br>
2996 <tt>begin()</tt>, <tt>end()</tt>
2997 <tt>size()</tt>, <tt>empty()</tt>
2999 <p>These are forwarding methods that make it easy to access the contents of
3000 a <tt>Module</tt> object's <a href="#Function"><tt>Function</tt></a>
3003 <li><tt>Module::FunctionListType &getFunctionList()</tt>
3005 <p> Returns the list of <a href="#Function"><tt>Function</tt></a>s. This is
3006 necessary to use when you need to update the list or perform a complex
3007 action that doesn't have a forwarding method.</p>
3009 <p><!-- Global Variable --></p></li>
3015 <li><tt>Module::global_iterator</tt> - Typedef for global variable list iterator<br>
3017 <tt>Module::const_global_iterator</tt> - Typedef for const_iterator.<br>
3019 <tt>global_begin()</tt>, <tt>global_end()</tt>
3020 <tt>global_size()</tt>, <tt>global_empty()</tt>
3022 <p> These are forwarding methods that make it easy to access the contents of
3023 a <tt>Module</tt> object's <a
3024 href="#GlobalVariable"><tt>GlobalVariable</tt></a> list.</p></li>
3026 <li><tt>Module::GlobalListType &getGlobalList()</tt>
3028 <p>Returns the list of <a
3029 href="#GlobalVariable"><tt>GlobalVariable</tt></a>s. This is necessary to
3030 use when you need to update the list or perform a complex action that
3031 doesn't have a forwarding method.</p>
3033 <p><!-- Symbol table stuff --> </p></li>
3039 <li><tt><a href="#SymbolTable">SymbolTable</a> *getSymbolTable()</tt>
3041 <p>Return a reference to the <a href="#SymbolTable"><tt>SymbolTable</tt></a>
3042 for this <tt>Module</tt>.</p>
3044 <p><!-- Convenience methods --></p></li>
3050 <li><tt><a href="#Function">Function</a> *getFunction(const std::string
3051 &Name, const <a href="#FunctionType">FunctionType</a> *Ty)</tt>
3053 <p>Look up the specified function in the <tt>Module</tt> <a
3054 href="#SymbolTable"><tt>SymbolTable</tt></a>. If it does not exist, return
3055 <tt>null</tt>.</p></li>
3057 <li><tt><a href="#Function">Function</a> *getOrInsertFunction(const
3058 std::string &Name, const <a href="#FunctionType">FunctionType</a> *T)</tt>
3060 <p>Look up the specified function in the <tt>Module</tt> <a
3061 href="#SymbolTable"><tt>SymbolTable</tt></a>. If it does not exist, add an
3062 external declaration for the function and return it.</p></li>
3064 <li><tt>std::string getTypeName(const <a href="#Type">Type</a> *Ty)</tt>
3066 <p>If there is at least one entry in the <a
3067 href="#SymbolTable"><tt>SymbolTable</tt></a> for the specified <a
3068 href="#Type"><tt>Type</tt></a>, return it. Otherwise return the empty
3071 <li><tt>bool addTypeName(const std::string &Name, const <a
3072 href="#Type">Type</a> *Ty)</tt>
3074 <p>Insert an entry in the <a href="#SymbolTable"><tt>SymbolTable</tt></a>
3075 mapping <tt>Name</tt> to <tt>Ty</tt>. If there is already an entry for this
3076 name, true is returned and the <a
3077 href="#SymbolTable"><tt>SymbolTable</tt></a> is not modified.</p></li>
3083 <!-- ======================================================================= -->
3084 <div class="doc_subsection">
3085 <a name="Value">The <tt>Value</tt> class</a>
3088 <div class="doc_text">
3090 <p><tt>#include "<a href="/doxygen/Value_8h-source.html">llvm/Value.h</a>"</tt>
3092 doxygen info: <a href="/doxygen/classllvm_1_1Value.html">Value Class</a></p>
3094 <p>The <tt>Value</tt> class is the most important class in the LLVM Source
3095 base. It represents a typed value that may be used (among other things) as an
3096 operand to an instruction. There are many different types of <tt>Value</tt>s,
3097 such as <a href="#Constant"><tt>Constant</tt></a>s,<a
3098 href="#Argument"><tt>Argument</tt></a>s. Even <a
3099 href="#Instruction"><tt>Instruction</tt></a>s and <a
3100 href="#Function"><tt>Function</tt></a>s are <tt>Value</tt>s.</p>
3102 <p>A particular <tt>Value</tt> may be used many times in the LLVM representation
3103 for a program. For example, an incoming argument to a function (represented
3104 with an instance of the <a href="#Argument">Argument</a> class) is "used" by
3105 every instruction in the function that references the argument. To keep track
3106 of this relationship, the <tt>Value</tt> class keeps a list of all of the <a
3107 href="#User"><tt>User</tt></a>s that is using it (the <a
3108 href="#User"><tt>User</tt></a> class is a base class for all nodes in the LLVM
3109 graph that can refer to <tt>Value</tt>s). This use list is how LLVM represents
3110 def-use information in the program, and is accessible through the <tt>use_</tt>*
3111 methods, shown below.</p>
3113 <p>Because LLVM is a typed representation, every LLVM <tt>Value</tt> is typed,
3114 and this <a href="#Type">Type</a> is available through the <tt>getType()</tt>
3115 method. In addition, all LLVM values can be named. The "name" of the
3116 <tt>Value</tt> is a symbolic string printed in the LLVM code:</p>
3118 <div class="doc_code">
3120 %<b>foo</b> = add i32 1, 2
3124 <p><a name="nameWarning">The name of this instruction is "foo".</a> <b>NOTE</b>
3125 that the name of any value may be missing (an empty string), so names should
3126 <b>ONLY</b> be used for debugging (making the source code easier to read,
3127 debugging printouts), they should not be used to keep track of values or map
3128 between them. For this purpose, use a <tt>std::map</tt> of pointers to the
3129 <tt>Value</tt> itself instead.</p>
3131 <p>One important aspect of LLVM is that there is no distinction between an SSA
3132 variable and the operation that produces it. Because of this, any reference to
3133 the value produced by an instruction (or the value available as an incoming
3134 argument, for example) is represented as a direct pointer to the instance of
3136 represents this value. Although this may take some getting used to, it
3137 simplifies the representation and makes it easier to manipulate.</p>
3141 <!-- _______________________________________________________________________ -->
3142 <div class="doc_subsubsection">
3143 <a name="m_Value">Important Public Members of the <tt>Value</tt> class</a>
3146 <div class="doc_text">
3149 <li><tt>Value::use_iterator</tt> - Typedef for iterator over the
3151 <tt>Value::use_const_iterator</tt> - Typedef for const_iterator over
3153 <tt>unsigned use_size()</tt> - Returns the number of users of the
3155 <tt>bool use_empty()</tt> - Returns true if there are no users.<br>
3156 <tt>use_iterator use_begin()</tt> - Get an iterator to the start of
3158 <tt>use_iterator use_end()</tt> - Get an iterator to the end of the
3160 <tt><a href="#User">User</a> *use_back()</tt> - Returns the last
3161 element in the list.
3162 <p> These methods are the interface to access the def-use
3163 information in LLVM. As with all other iterators in LLVM, the naming
3164 conventions follow the conventions defined by the <a href="#stl">STL</a>.</p>
3166 <li><tt><a href="#Type">Type</a> *getType() const</tt>
3167 <p>This method returns the Type of the Value.</p>
3169 <li><tt>bool hasName() const</tt><br>
3170 <tt>std::string getName() const</tt><br>
3171 <tt>void setName(const std::string &Name)</tt>
3172 <p> This family of methods is used to access and assign a name to a <tt>Value</tt>,
3173 be aware of the <a href="#nameWarning">precaution above</a>.</p>
3175 <li><tt>void replaceAllUsesWith(Value *V)</tt>
3177 <p>This method traverses the use list of a <tt>Value</tt> changing all <a
3178 href="#User"><tt>User</tt>s</a> of the current value to refer to
3179 "<tt>V</tt>" instead. For example, if you detect that an instruction always
3180 produces a constant value (for example through constant folding), you can
3181 replace all uses of the instruction with the constant like this:</p>
3183 <div class="doc_code">
3185 Inst->replaceAllUsesWith(ConstVal);
3193 <!-- ======================================================================= -->
3194 <div class="doc_subsection">
3195 <a name="User">The <tt>User</tt> class</a>
3198 <div class="doc_text">
3201 <tt>#include "<a href="/doxygen/User_8h-source.html">llvm/User.h</a>"</tt><br>
3202 doxygen info: <a href="/doxygen/classllvm_1_1User.html">User Class</a><br>
3203 Superclass: <a href="#Value"><tt>Value</tt></a></p>
3205 <p>The <tt>User</tt> class is the common base class of all LLVM nodes that may
3206 refer to <a href="#Value"><tt>Value</tt></a>s. It exposes a list of "Operands"
3207 that are all of the <a href="#Value"><tt>Value</tt></a>s that the User is
3208 referring to. The <tt>User</tt> class itself is a subclass of
3211 <p>The operands of a <tt>User</tt> point directly to the LLVM <a
3212 href="#Value"><tt>Value</tt></a> that it refers to. Because LLVM uses Static
3213 Single Assignment (SSA) form, there can only be one definition referred to,
3214 allowing this direct connection. This connection provides the use-def
3215 information in LLVM.</p>
3219 <!-- _______________________________________________________________________ -->
3220 <div class="doc_subsubsection">
3221 <a name="m_User">Important Public Members of the <tt>User</tt> class</a>
3224 <div class="doc_text">
3226 <p>The <tt>User</tt> class exposes the operand list in two ways: through
3227 an index access interface and through an iterator based interface.</p>
3230 <li><tt>Value *getOperand(unsigned i)</tt><br>
3231 <tt>unsigned getNumOperands()</tt>
3232 <p> These two methods expose the operands of the <tt>User</tt> in a
3233 convenient form for direct access.</p></li>
3235 <li><tt>User::op_iterator</tt> - Typedef for iterator over the operand
3237 <tt>op_iterator op_begin()</tt> - Get an iterator to the start of
3238 the operand list.<br>
3239 <tt>op_iterator op_end()</tt> - Get an iterator to the end of the
3241 <p> Together, these methods make up the iterator based interface to
3242 the operands of a <tt>User</tt>.</p></li>
3247 <!-- ======================================================================= -->
3248 <div class="doc_subsection">
3249 <a name="Instruction">The <tt>Instruction</tt> class</a>
3252 <div class="doc_text">
3254 <p><tt>#include "</tt><tt><a
3255 href="/doxygen/Instruction_8h-source.html">llvm/Instruction.h</a>"</tt><br>
3256 doxygen info: <a href="/doxygen/classllvm_1_1Instruction.html">Instruction Class</a><br>
3257 Superclasses: <a href="#User"><tt>User</tt></a>, <a
3258 href="#Value"><tt>Value</tt></a></p>
3260 <p>The <tt>Instruction</tt> class is the common base class for all LLVM
3261 instructions. It provides only a few methods, but is a very commonly used
3262 class. The primary data tracked by the <tt>Instruction</tt> class itself is the
3263 opcode (instruction type) and the parent <a
3264 href="#BasicBlock"><tt>BasicBlock</tt></a> the <tt>Instruction</tt> is embedded
3265 into. To represent a specific type of instruction, one of many subclasses of
3266 <tt>Instruction</tt> are used.</p>
3268 <p> Because the <tt>Instruction</tt> class subclasses the <a
3269 href="#User"><tt>User</tt></a> class, its operands can be accessed in the same
3270 way as for other <a href="#User"><tt>User</tt></a>s (with the
3271 <tt>getOperand()</tt>/<tt>getNumOperands()</tt> and
3272 <tt>op_begin()</tt>/<tt>op_end()</tt> methods).</p> <p> An important file for
3273 the <tt>Instruction</tt> class is the <tt>llvm/Instruction.def</tt> file. This
3274 file contains some meta-data about the various different types of instructions
3275 in LLVM. It describes the enum values that are used as opcodes (for example
3276 <tt>Instruction::Add</tt> and <tt>Instruction::ICmp</tt>), as well as the
3277 concrete sub-classes of <tt>Instruction</tt> that implement the instruction (for
3278 example <tt><a href="#BinaryOperator">BinaryOperator</a></tt> and <tt><a
3279 href="#CmpInst">CmpInst</a></tt>). Unfortunately, the use of macros in
3280 this file confuses doxygen, so these enum values don't show up correctly in the
3281 <a href="/doxygen/classllvm_1_1Instruction.html">doxygen output</a>.</p>
3285 <!-- _______________________________________________________________________ -->
3286 <div class="doc_subsubsection">
3287 <a name="s_Instruction">Important Subclasses of the <tt>Instruction</tt>
3290 <div class="doc_text">
3292 <li><tt><a name="BinaryOperator">BinaryOperator</a></tt>
3293 <p>This subclasses represents all two operand instructions whose operands
3294 must be the same type, except for the comparison instructions.</p></li>
3295 <li><tt><a name="CastInst">CastInst</a></tt>
3296 <p>This subclass is the parent of the 12 casting instructions. It provides
3297 common operations on cast instructions.</p>
3298 <li><tt><a name="CmpInst">CmpInst</a></tt>
3299 <p>This subclass respresents the two comparison instructions,
3300 <a href="LangRef.html#i_icmp">ICmpInst</a> (integer opreands), and
3301 <a href="LangRef.html#i_fcmp">FCmpInst</a> (floating point operands).</p>
3302 <li><tt><a name="TerminatorInst">TerminatorInst</a></tt>
3303 <p>This subclass is the parent of all terminator instructions (those which
3304 can terminate a block).</p>
3308 <!-- _______________________________________________________________________ -->
3309 <div class="doc_subsubsection">
3310 <a name="m_Instruction">Important Public Members of the <tt>Instruction</tt>
3314 <div class="doc_text">
3317 <li><tt><a href="#BasicBlock">BasicBlock</a> *getParent()</tt>
3318 <p>Returns the <a href="#BasicBlock"><tt>BasicBlock</tt></a> that
3319 this <tt>Instruction</tt> is embedded into.</p></li>
3320 <li><tt>bool mayWriteToMemory()</tt>
3321 <p>Returns true if the instruction writes to memory, i.e. it is a
3322 <tt>call</tt>,<tt>free</tt>,<tt>invoke</tt>, or <tt>store</tt>.</p></li>
3323 <li><tt>unsigned getOpcode()</tt>
3324 <p>Returns the opcode for the <tt>Instruction</tt>.</p></li>
3325 <li><tt><a href="#Instruction">Instruction</a> *clone() const</tt>
3326 <p>Returns another instance of the specified instruction, identical
3327 in all ways to the original except that the instruction has no parent
3328 (ie it's not embedded into a <a href="#BasicBlock"><tt>BasicBlock</tt></a>),
3329 and it has no name</p></li>
3334 <!-- ======================================================================= -->
3335 <div class="doc_subsection">
3336 <a name="Constant">The <tt>Constant</tt> class and subclasses</a>
3339 <div class="doc_text">
3341 <p>Constant represents a base class for different types of constants. It
3342 is subclassed by ConstantInt, ConstantArray, etc. for representing
3343 the various types of Constants. <a href="#GlobalValue">GlobalValue</a> is also
3344 a subclass, which represents the address of a global variable or function.
3349 <!-- _______________________________________________________________________ -->
3350 <div class="doc_subsubsection">Important Subclasses of Constant </div>
3351 <div class="doc_text">
3353 <li>ConstantInt : This subclass of Constant represents an integer constant of
3356 <li><tt>const APInt& getValue() const</tt>: Returns the underlying
3357 value of this constant, an APInt value.</li>
3358 <li><tt>int64_t getSExtValue() const</tt>: Converts the underlying APInt
3359 value to an int64_t via sign extension. If the value (not the bit width)
3360 of the APInt is too large to fit in an int64_t, an assertion will result.
3361 For this reason, use of this method is discouraged.</li>
3362 <li><tt>uint64_t getZExtValue() const</tt>: Converts the underlying APInt
3363 value to a uint64_t via zero extension. IF the value (not the bit width)
3364 of the APInt is too large to fit in a uint64_t, an assertion will result.
3365 For this reason, use of this method is discouraged.</li>
3366 <li><tt>static ConstantInt* get(const APInt& Val)</tt>: Returns the
3367 ConstantInt object that represents the value provided by <tt>Val</tt>.
3368 The type is implied as the IntegerType that corresponds to the bit width
3369 of <tt>Val</tt>.</li>
3370 <li><tt>static ConstantInt* get(const Type *Ty, uint64_t Val)</tt>:
3371 Returns the ConstantInt object that represents the value provided by
3372 <tt>Val</tt> for integer type <tt>Ty</tt>.</li>
3375 <li>ConstantFP : This class represents a floating point constant.
3377 <li><tt>double getValue() const</tt>: Returns the underlying value of
3378 this constant. </li>
3381 <li>ConstantArray : This represents a constant array.
3383 <li><tt>const std::vector<Use> &getValues() const</tt>: Returns
3384 a vector of component constants that makeup this array. </li>
3387 <li>ConstantStruct : This represents a constant struct.
3389 <li><tt>const std::vector<Use> &getValues() const</tt>: Returns
3390 a vector of component constants that makeup this array. </li>
3393 <li>GlobalValue : This represents either a global variable or a function. In
3394 either case, the value is a constant fixed address (after linking).
3400 <!-- ======================================================================= -->
3401 <div class="doc_subsection">
3402 <a name="GlobalValue">The <tt>GlobalValue</tt> class</a>
3405 <div class="doc_text">
3408 href="/doxygen/GlobalValue_8h-source.html">llvm/GlobalValue.h</a>"</tt><br>
3409 doxygen info: <a href="/doxygen/classllvm_1_1GlobalValue.html">GlobalValue
3411 Superclasses: <a href="#Constant"><tt>Constant</tt></a>,
3412 <a href="#User"><tt>User</tt></a>, <a href="#Value"><tt>Value</tt></a></p>
3414 <p>Global values (<a href="#GlobalVariable"><tt>GlobalVariable</tt></a>s or <a
3415 href="#Function"><tt>Function</tt></a>s) are the only LLVM values that are
3416 visible in the bodies of all <a href="#Function"><tt>Function</tt></a>s.
3417 Because they are visible at global scope, they are also subject to linking with
3418 other globals defined in different translation units. To control the linking
3419 process, <tt>GlobalValue</tt>s know their linkage rules. Specifically,
3420 <tt>GlobalValue</tt>s know whether they have internal or external linkage, as
3421 defined by the <tt>LinkageTypes</tt> enumeration.</p>
3423 <p>If a <tt>GlobalValue</tt> has internal linkage (equivalent to being
3424 <tt>static</tt> in C), it is not visible to code outside the current translation
3425 unit, and does not participate in linking. If it has external linkage, it is
3426 visible to external code, and does participate in linking. In addition to
3427 linkage information, <tt>GlobalValue</tt>s keep track of which <a
3428 href="#Module"><tt>Module</tt></a> they are currently part of.</p>
3430 <p>Because <tt>GlobalValue</tt>s are memory objects, they are always referred to
3431 by their <b>address</b>. As such, the <a href="#Type"><tt>Type</tt></a> of a
3432 global is always a pointer to its contents. It is important to remember this
3433 when using the <tt>GetElementPtrInst</tt> instruction because this pointer must
3434 be dereferenced first. For example, if you have a <tt>GlobalVariable</tt> (a
3435 subclass of <tt>GlobalValue)</tt> that is an array of 24 ints, type <tt>[24 x
3436 i32]</tt>, then the <tt>GlobalVariable</tt> is a pointer to that array. Although
3437 the address of the first element of this array and the value of the
3438 <tt>GlobalVariable</tt> are the same, they have different types. The
3439 <tt>GlobalVariable</tt>'s type is <tt>[24 x i32]</tt>. The first element's type
3440 is <tt>i32.</tt> Because of this, accessing a global value requires you to
3441 dereference the pointer with <tt>GetElementPtrInst</tt> first, then its elements
3442 can be accessed. This is explained in the <a href="LangRef.html#globalvars">LLVM
3443 Language Reference Manual</a>.</p>
3447 <!-- _______________________________________________________________________ -->
3448 <div class="doc_subsubsection">
3449 <a name="m_GlobalValue">Important Public Members of the <tt>GlobalValue</tt>
3453 <div class="doc_text">
3456 <li><tt>bool hasInternalLinkage() const</tt><br>
3457 <tt>bool hasExternalLinkage() const</tt><br>
3458 <tt>void setInternalLinkage(bool HasInternalLinkage)</tt>
3459 <p> These methods manipulate the linkage characteristics of the <tt>GlobalValue</tt>.</p>
3462 <li><tt><a href="#Module">Module</a> *getParent()</tt>
3463 <p> This returns the <a href="#Module"><tt>Module</tt></a> that the
3464 GlobalValue is currently embedded into.</p></li>
3469 <!-- ======================================================================= -->
3470 <div class="doc_subsection">
3471 <a name="Function">The <tt>Function</tt> class</a>
3474 <div class="doc_text">
3477 href="/doxygen/Function_8h-source.html">llvm/Function.h</a>"</tt><br> doxygen
3478 info: <a href="/doxygen/classllvm_1_1Function.html">Function Class</a><br>
3479 Superclasses: <a href="#GlobalValue"><tt>GlobalValue</tt></a>,
3480 <a href="#Constant"><tt>Constant</tt></a>,
3481 <a href="#User"><tt>User</tt></a>,
3482 <a href="#Value"><tt>Value</tt></a></p>
3484 <p>The <tt>Function</tt> class represents a single procedure in LLVM. It is
3485 actually one of the more complex classes in the LLVM heirarchy because it must
3486 keep track of a large amount of data. The <tt>Function</tt> class keeps track
3487 of a list of <a href="#BasicBlock"><tt>BasicBlock</tt></a>s, a list of formal
3488 <a href="#Argument"><tt>Argument</tt></a>s, and a
3489 <a href="#SymbolTable"><tt>SymbolTable</tt></a>.</p>
3491 <p>The list of <a href="#BasicBlock"><tt>BasicBlock</tt></a>s is the most
3492 commonly used part of <tt>Function</tt> objects. The list imposes an implicit
3493 ordering of the blocks in the function, which indicate how the code will be
3494 layed out by the backend. Additionally, the first <a
3495 href="#BasicBlock"><tt>BasicBlock</tt></a> is the implicit entry node for the
3496 <tt>Function</tt>. It is not legal in LLVM to explicitly branch to this initial
3497 block. There are no implicit exit nodes, and in fact there may be multiple exit
3498 nodes from a single <tt>Function</tt>. If the <a
3499 href="#BasicBlock"><tt>BasicBlock</tt></a> list is empty, this indicates that
3500 the <tt>Function</tt> is actually a function declaration: the actual body of the
3501 function hasn't been linked in yet.</p>
3503 <p>In addition to a list of <a href="#BasicBlock"><tt>BasicBlock</tt></a>s, the
3504 <tt>Function</tt> class also keeps track of the list of formal <a
3505 href="#Argument"><tt>Argument</tt></a>s that the function receives. This
3506 container manages the lifetime of the <a href="#Argument"><tt>Argument</tt></a>
3507 nodes, just like the <a href="#BasicBlock"><tt>BasicBlock</tt></a> list does for
3508 the <a href="#BasicBlock"><tt>BasicBlock</tt></a>s.</p>
3510 <p>The <a href="#SymbolTable"><tt>SymbolTable</tt></a> is a very rarely used
3511 LLVM feature that is only used when you have to look up a value by name. Aside
3512 from that, the <a href="#SymbolTable"><tt>SymbolTable</tt></a> is used
3513 internally to make sure that there are not conflicts between the names of <a
3514 href="#Instruction"><tt>Instruction</tt></a>s, <a
3515 href="#BasicBlock"><tt>BasicBlock</tt></a>s, or <a
3516 href="#Argument"><tt>Argument</tt></a>s in the function body.</p>
3518 <p>Note that <tt>Function</tt> is a <a href="#GlobalValue">GlobalValue</a>
3519 and therefore also a <a href="#Constant">Constant</a>. The value of the function
3520 is its address (after linking) which is guaranteed to be constant.</p>
3523 <!-- _______________________________________________________________________ -->
3524 <div class="doc_subsubsection">
3525 <a name="m_Function">Important Public Members of the <tt>Function</tt>
3529 <div class="doc_text">
3532 <li><tt>Function(const </tt><tt><a href="#FunctionType">FunctionType</a>
3533 *Ty, LinkageTypes Linkage, const std::string &N = "", Module* Parent = 0)</tt>
3535 <p>Constructor used when you need to create new <tt>Function</tt>s to add
3536 the the program. The constructor must specify the type of the function to
3537 create and what type of linkage the function should have. The <a
3538 href="#FunctionType"><tt>FunctionType</tt></a> argument
3539 specifies the formal arguments and return value for the function. The same
3540 <a href="#FunctionType"><tt>FunctionType</tt></a> value can be used to
3541 create multiple functions. The <tt>Parent</tt> argument specifies the Module
3542 in which the function is defined. If this argument is provided, the function
3543 will automatically be inserted into that module's list of
3546 <li><tt>bool isDeclaration()</tt>
3548 <p>Return whether or not the <tt>Function</tt> has a body defined. If the
3549 function is "external", it does not have a body, and thus must be resolved
3550 by linking with a function defined in a different translation unit.</p></li>
3552 <li><tt>Function::iterator</tt> - Typedef for basic block list iterator<br>
3553 <tt>Function::const_iterator</tt> - Typedef for const_iterator.<br>
3555 <tt>begin()</tt>, <tt>end()</tt>
3556 <tt>size()</tt>, <tt>empty()</tt>
3558 <p>These are forwarding methods that make it easy to access the contents of
3559 a <tt>Function</tt> object's <a href="#BasicBlock"><tt>BasicBlock</tt></a>
3562 <li><tt>Function::BasicBlockListType &getBasicBlockList()</tt>
3564 <p>Returns the list of <a href="#BasicBlock"><tt>BasicBlock</tt></a>s. This
3565 is necessary to use when you need to update the list or perform a complex
3566 action that doesn't have a forwarding method.</p></li>
3568 <li><tt>Function::arg_iterator</tt> - Typedef for the argument list
3570 <tt>Function::const_arg_iterator</tt> - Typedef for const_iterator.<br>
3572 <tt>arg_begin()</tt>, <tt>arg_end()</tt>
3573 <tt>arg_size()</tt>, <tt>arg_empty()</tt>
3575 <p>These are forwarding methods that make it easy to access the contents of
3576 a <tt>Function</tt> object's <a href="#Argument"><tt>Argument</tt></a>
3579 <li><tt>Function::ArgumentListType &getArgumentList()</tt>
3581 <p>Returns the list of <a href="#Argument"><tt>Argument</tt></a>s. This is
3582 necessary to use when you need to update the list or perform a complex
3583 action that doesn't have a forwarding method.</p></li>
3585 <li><tt><a href="#BasicBlock">BasicBlock</a> &getEntryBlock()</tt>
3587 <p>Returns the entry <a href="#BasicBlock"><tt>BasicBlock</tt></a> for the
3588 function. Because the entry block for the function is always the first
3589 block, this returns the first block of the <tt>Function</tt>.</p></li>
3591 <li><tt><a href="#Type">Type</a> *getReturnType()</tt><br>
3592 <tt><a href="#FunctionType">FunctionType</a> *getFunctionType()</tt>
3594 <p>This traverses the <a href="#Type"><tt>Type</tt></a> of the
3595 <tt>Function</tt> and returns the return type of the function, or the <a
3596 href="#FunctionType"><tt>FunctionType</tt></a> of the actual
3599 <li><tt><a href="#SymbolTable">SymbolTable</a> *getSymbolTable()</tt>
3601 <p> Return a pointer to the <a href="#SymbolTable"><tt>SymbolTable</tt></a>
3602 for this <tt>Function</tt>.</p></li>
3607 <!-- ======================================================================= -->
3608 <div class="doc_subsection">
3609 <a name="GlobalVariable">The <tt>GlobalVariable</tt> class</a>
3612 <div class="doc_text">
3615 href="/doxygen/GlobalVariable_8h-source.html">llvm/GlobalVariable.h</a>"</tt>
3617 doxygen info: <a href="/doxygen/classllvm_1_1GlobalVariable.html">GlobalVariable
3619 Superclasses: <a href="#GlobalValue"><tt>GlobalValue</tt></a>,
3620 <a href="#Constant"><tt>Constant</tt></a>,
3621 <a href="#User"><tt>User</tt></a>,
3622 <a href="#Value"><tt>Value</tt></a></p>
3624 <p>Global variables are represented with the (suprise suprise)
3625 <tt>GlobalVariable</tt> class. Like functions, <tt>GlobalVariable</tt>s are also
3626 subclasses of <a href="#GlobalValue"><tt>GlobalValue</tt></a>, and as such are
3627 always referenced by their address (global values must live in memory, so their
3628 "name" refers to their constant address). See
3629 <a href="#GlobalValue"><tt>GlobalValue</tt></a> for more on this. Global
3630 variables may have an initial value (which must be a
3631 <a href="#Constant"><tt>Constant</tt></a>), and if they have an initializer,
3632 they may be marked as "constant" themselves (indicating that their contents
3633 never change at runtime).</p>
3636 <!-- _______________________________________________________________________ -->
3637 <div class="doc_subsubsection">
3638 <a name="m_GlobalVariable">Important Public Members of the
3639 <tt>GlobalVariable</tt> class</a>
3642 <div class="doc_text">
3645 <li><tt>GlobalVariable(const </tt><tt><a href="#Type">Type</a> *Ty, bool
3646 isConstant, LinkageTypes& Linkage, <a href="#Constant">Constant</a>
3647 *Initializer = 0, const std::string &Name = "", Module* Parent = 0)</tt>
3649 <p>Create a new global variable of the specified type. If
3650 <tt>isConstant</tt> is true then the global variable will be marked as
3651 unchanging for the program. The Linkage parameter specifies the type of
3652 linkage (internal, external, weak, linkonce, appending) for the variable.
3653 If the linkage is InternalLinkage, WeakAnyLinkage, WeakODRLinkage,
3654 LinkOnceAnyLinkage or LinkOnceODRLinkage, then the resultant
3655 global variable will have internal linkage. AppendingLinkage concatenates
3656 together all instances (in different translation units) of the variable
3657 into a single variable but is only applicable to arrays. See
3658 the <a href="LangRef.html#modulestructure">LLVM Language Reference</a> for
3659 further details on linkage types. Optionally an initializer, a name, and the
3660 module to put the variable into may be specified for the global variable as
3663 <li><tt>bool isConstant() const</tt>
3665 <p>Returns true if this is a global variable that is known not to
3666 be modified at runtime.</p></li>
3668 <li><tt>bool hasInitializer()</tt>
3670 <p>Returns true if this <tt>GlobalVariable</tt> has an intializer.</p></li>
3672 <li><tt><a href="#Constant">Constant</a> *getInitializer()</tt>
3674 <p>Returns the intial value for a <tt>GlobalVariable</tt>. It is not legal
3675 to call this method if there is no initializer.</p></li>
3681 <!-- ======================================================================= -->
3682 <div class="doc_subsection">
3683 <a name="BasicBlock">The <tt>BasicBlock</tt> class</a>
3686 <div class="doc_text">
3689 href="/doxygen/BasicBlock_8h-source.html">llvm/BasicBlock.h</a>"</tt><br>
3690 doxygen info: <a href="/doxygen/classllvm_1_1BasicBlock.html">BasicBlock
3692 Superclass: <a href="#Value"><tt>Value</tt></a></p>
3694 <p>This class represents a single entry multiple exit section of the code,
3695 commonly known as a basic block by the compiler community. The
3696 <tt>BasicBlock</tt> class maintains a list of <a
3697 href="#Instruction"><tt>Instruction</tt></a>s, which form the body of the block.
3698 Matching the language definition, the last element of this list of instructions
3699 is always a terminator instruction (a subclass of the <a
3700 href="#TerminatorInst"><tt>TerminatorInst</tt></a> class).</p>
3702 <p>In addition to tracking the list of instructions that make up the block, the
3703 <tt>BasicBlock</tt> class also keeps track of the <a
3704 href="#Function"><tt>Function</tt></a> that it is embedded into.</p>
3706 <p>Note that <tt>BasicBlock</tt>s themselves are <a
3707 href="#Value"><tt>Value</tt></a>s, because they are referenced by instructions
3708 like branches and can go in the switch tables. <tt>BasicBlock</tt>s have type
3713 <!-- _______________________________________________________________________ -->
3714 <div class="doc_subsubsection">
3715 <a name="m_BasicBlock">Important Public Members of the <tt>BasicBlock</tt>
3719 <div class="doc_text">
3722 <li><tt>BasicBlock(const std::string &Name = "", </tt><tt><a
3723 href="#Function">Function</a> *Parent = 0)</tt>
3725 <p>The <tt>BasicBlock</tt> constructor is used to create new basic blocks for
3726 insertion into a function. The constructor optionally takes a name for the new
3727 block, and a <a href="#Function"><tt>Function</tt></a> to insert it into. If
3728 the <tt>Parent</tt> parameter is specified, the new <tt>BasicBlock</tt> is
3729 automatically inserted at the end of the specified <a
3730 href="#Function"><tt>Function</tt></a>, if not specified, the BasicBlock must be
3731 manually inserted into the <a href="#Function"><tt>Function</tt></a>.</p></li>
3733 <li><tt>BasicBlock::iterator</tt> - Typedef for instruction list iterator<br>
3734 <tt>BasicBlock::const_iterator</tt> - Typedef for const_iterator.<br>
3735 <tt>begin()</tt>, <tt>end()</tt>, <tt>front()</tt>, <tt>back()</tt>,
3736 <tt>size()</tt>, <tt>empty()</tt>
3737 STL-style functions for accessing the instruction list.
3739 <p>These methods and typedefs are forwarding functions that have the same
3740 semantics as the standard library methods of the same names. These methods
3741 expose the underlying instruction list of a basic block in a way that is easy to
3742 manipulate. To get the full complement of container operations (including
3743 operations to update the list), you must use the <tt>getInstList()</tt>
3746 <li><tt>BasicBlock::InstListType &getInstList()</tt>
3748 <p>This method is used to get access to the underlying container that actually
3749 holds the Instructions. This method must be used when there isn't a forwarding
3750 function in the <tt>BasicBlock</tt> class for the operation that you would like
3751 to perform. Because there are no forwarding functions for "updating"
3752 operations, you need to use this if you want to update the contents of a
3753 <tt>BasicBlock</tt>.</p></li>
3755 <li><tt><a href="#Function">Function</a> *getParent()</tt>
3757 <p> Returns a pointer to <a href="#Function"><tt>Function</tt></a> the block is
3758 embedded into, or a null pointer if it is homeless.</p></li>
3760 <li><tt><a href="#TerminatorInst">TerminatorInst</a> *getTerminator()</tt>
3762 <p> Returns a pointer to the terminator instruction that appears at the end of
3763 the <tt>BasicBlock</tt>. If there is no terminator instruction, or if the last
3764 instruction in the block is not a terminator, then a null pointer is
3772 <!-- ======================================================================= -->
3773 <div class="doc_subsection">
3774 <a name="Argument">The <tt>Argument</tt> class</a>
3777 <div class="doc_text">
3779 <p>This subclass of Value defines the interface for incoming formal
3780 arguments to a function. A Function maintains a list of its formal
3781 arguments. An argument has a pointer to the parent Function.</p>
3785 <!-- *********************************************************************** -->
3788 <a href="http://jigsaw.w3.org/css-validator/check/referer"><img
3789 src="http://jigsaw.w3.org/css-validator/images/vcss-blue" alt="Valid CSS"></a>
3790 <a href="http://validator.w3.org/check/referer"><img
3791 src="http://www.w3.org/Icons/valid-html401" alt="Valid HTML 4.01 Strict"></a>
3793 <a href="mailto:dhurjati@cs.uiuc.edu">Dinakar Dhurjati</a> and
3794 <a href="mailto:sabre@nondot.org">Chris Lattner</a><br>
3795 <a href="http://llvm.org">The LLVM Compiler Infrastructure</a><br>
3796 Last modified: $Date$