• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1<!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.01//EN"
2                      "http://www.w3.org/TR/html4/strict.dtd">
3<html>
4<head>
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">
8</head>
9<body>
10
11<h1>
12  LLVM Programmer's Manual
13</h1>
14
15<ol>
16  <li><a href="#introduction">Introduction</a></li>
17  <li><a href="#general">General Information</a>
18    <ul>
19      <li><a href="#stl">The C++ Standard Template Library</a></li>
20<!--
21      <li>The <tt>-time-passes</tt> option</li>
22      <li>How to use the LLVM Makefile system</li>
23      <li>How to write a regression test</li>
24
25-->
26    </ul>
27  </li>
28  <li><a href="#apis">Important and useful LLVM APIs</a>
29    <ul>
30      <li><a href="#isa">The <tt>isa&lt;&gt;</tt>, <tt>cast&lt;&gt;</tt>
31and <tt>dyn_cast&lt;&gt;</tt> templates</a> </li>
32      <li><a href="#string_apis">Passing strings (the <tt>StringRef</tt>
33and <tt>Twine</tt> classes)</a>
34        <ul>
35          <li><a href="#StringRef">The <tt>StringRef</tt> class</a> </li>
36          <li><a href="#Twine">The <tt>Twine</tt> class</a> </li>
37        </ul>
38      </li>
39      <li><a href="#DEBUG">The <tt>DEBUG()</tt> macro and <tt>-debug</tt>
40option</a>
41        <ul>
42          <li><a href="#DEBUG_TYPE">Fine grained debug info with <tt>DEBUG_TYPE</tt>
43and the <tt>-debug-only</tt> option</a> </li>
44        </ul>
45      </li>
46      <li><a href="#Statistic">The <tt>Statistic</tt> class &amp; <tt>-stats</tt>
47option</a></li>
48<!--
49      <li>The <tt>InstVisitor</tt> template
50      <li>The general graph API
51-->
52      <li><a href="#ViewGraph">Viewing graphs while debugging code</a></li>
53    </ul>
54  </li>
55  <li><a href="#datastructure">Picking the Right Data Structure for a Task</a>
56    <ul>
57    <li><a href="#ds_sequential">Sequential Containers (std::vector, std::list, etc)</a>
58    <ul>
59      <li><a href="#dss_arrayref">llvm/ADT/ArrayRef.h</a></li>
60      <li><a href="#dss_fixedarrays">Fixed Size Arrays</a></li>
61      <li><a href="#dss_heaparrays">Heap Allocated Arrays</a></li>
62      <li><a href="#dss_tinyptrvector">"llvm/ADT/TinyPtrVector.h"</a></li>
63      <li><a href="#dss_smallvector">"llvm/ADT/SmallVector.h"</a></li>
64      <li><a href="#dss_vector">&lt;vector&gt;</a></li>
65      <li><a href="#dss_deque">&lt;deque&gt;</a></li>
66      <li><a href="#dss_list">&lt;list&gt;</a></li>
67      <li><a href="#dss_ilist">llvm/ADT/ilist.h</a></li>
68      <li><a href="#dss_packedvector">llvm/ADT/PackedVector.h</a></li>
69      <li><a href="#dss_other">Other Sequential Container Options</a></li>
70    </ul></li>
71    <li><a href="#ds_string">String-like containers</a>
72    <ul>
73      <li><a href="#dss_stringref">llvm/ADT/StringRef.h</a></li>
74      <li><a href="#dss_twine">llvm/ADT/Twine.h</a></li>
75      <li><a href="#dss_smallstring">llvm/ADT/SmallString.h</a></li>
76      <li><a href="#dss_stdstring">std::string</a></li>
77    </ul></li>
78    <li><a href="#ds_set">Set-Like Containers (std::set, SmallSet, SetVector, etc)</a>
79    <ul>
80      <li><a href="#dss_sortedvectorset">A sorted 'vector'</a></li>
81      <li><a href="#dss_smallset">"llvm/ADT/SmallSet.h"</a></li>
82      <li><a href="#dss_smallptrset">"llvm/ADT/SmallPtrSet.h"</a></li>
83      <li><a href="#dss_denseset">"llvm/ADT/DenseSet.h"</a></li>
84      <li><a href="#dss_FoldingSet">"llvm/ADT/FoldingSet.h"</a></li>
85      <li><a href="#dss_set">&lt;set&gt;</a></li>
86      <li><a href="#dss_setvector">"llvm/ADT/SetVector.h"</a></li>
87      <li><a href="#dss_uniquevector">"llvm/ADT/UniqueVector.h"</a></li>
88      <li><a href="#dss_otherset">Other Set-Like ContainerOptions</a></li>
89    </ul></li>
90    <li><a href="#ds_map">Map-Like Containers (std::map, DenseMap, etc)</a>
91    <ul>
92      <li><a href="#dss_sortedvectormap">A sorted 'vector'</a></li>
93      <li><a href="#dss_stringmap">"llvm/ADT/StringMap.h"</a></li>
94      <li><a href="#dss_indexedmap">"llvm/ADT/IndexedMap.h"</a></li>
95      <li><a href="#dss_densemap">"llvm/ADT/DenseMap.h"</a></li>
96      <li><a href="#dss_valuemap">"llvm/ADT/ValueMap.h"</a></li>
97      <li><a href="#dss_intervalmap">"llvm/ADT/IntervalMap.h"</a></li>
98      <li><a href="#dss_map">&lt;map&gt;</a></li>
99      <li><a href="#dss_inteqclasses">"llvm/ADT/IntEqClasses.h"</a></li>
100      <li><a href="#dss_othermap">Other Map-Like Container Options</a></li>
101    </ul></li>
102    <li><a href="#ds_bit">BitVector-like containers</a>
103    <ul>
104      <li><a href="#dss_bitvector">A dense bitvector</a></li>
105      <li><a href="#dss_smallbitvector">A "small" dense bitvector</a></li>
106      <li><a href="#dss_sparsebitvector">A sparse bitvector</a></li>
107    </ul></li>
108  </ul>
109  </li>
110  <li><a href="#common">Helpful Hints for Common Operations</a>
111    <ul>
112      <li><a href="#inspection">Basic Inspection and Traversal Routines</a>
113        <ul>
114          <li><a href="#iterate_function">Iterating over the <tt>BasicBlock</tt>s
115in a <tt>Function</tt></a> </li>
116          <li><a href="#iterate_basicblock">Iterating over the <tt>Instruction</tt>s
117in a <tt>BasicBlock</tt></a> </li>
118          <li><a href="#iterate_institer">Iterating over the <tt>Instruction</tt>s
119in a <tt>Function</tt></a> </li>
120          <li><a href="#iterate_convert">Turning an iterator into a
121class pointer</a> </li>
122          <li><a href="#iterate_complex">Finding call sites: a more
123complex example</a> </li>
124          <li><a href="#calls_and_invokes">Treating calls and invokes
125the same way</a> </li>
126          <li><a href="#iterate_chains">Iterating over def-use &amp;
127use-def chains</a> </li>
128          <li><a href="#iterate_preds">Iterating over predecessors &amp;
129successors of blocks</a></li>
130        </ul>
131      </li>
132      <li><a href="#simplechanges">Making simple changes</a>
133        <ul>
134          <li><a href="#schanges_creating">Creating and inserting new
135		 <tt>Instruction</tt>s</a> </li>
136          <li><a href="#schanges_deleting">Deleting 		 <tt>Instruction</tt>s</a> </li>
137          <li><a href="#schanges_replacing">Replacing an 		 <tt>Instruction</tt>
138with another <tt>Value</tt></a> </li>
139          <li><a href="#schanges_deletingGV">Deleting <tt>GlobalVariable</tt>s</a> </li>
140        </ul>
141      </li>
142      <li><a href="#create_types">How to Create Types</a></li>
143<!--
144    <li>Working with the Control Flow Graph
145    <ul>
146      <li>Accessing predecessors and successors of a <tt>BasicBlock</tt>
147      <li>
148      <li>
149    </ul>
150-->
151    </ul>
152  </li>
153
154  <li><a href="#threading">Threads and LLVM</a>
155  <ul>
156    <li><a href="#startmultithreaded">Entering and Exiting Multithreaded Mode
157        </a></li>
158    <li><a href="#shutdown">Ending execution with <tt>llvm_shutdown()</tt></a></li>
159    <li><a href="#managedstatic">Lazy initialization with <tt>ManagedStatic</tt></a></li>
160    <li><a href="#llvmcontext">Achieving Isolation with <tt>LLVMContext</tt></a></li>
161    <li><a href="#jitthreading">Threads and the JIT</a></li>
162  </ul>
163  </li>
164
165  <li><a href="#advanced">Advanced Topics</a>
166  <ul>
167
168  <li><a href="#SymbolTable">The <tt>ValueSymbolTable</tt> class</a></li>
169  <li><a href="#UserLayout">The <tt>User</tt> and owned <tt>Use</tt> classes' memory layout</a></li>
170  </ul></li>
171
172  <li><a href="#coreclasses">The Core LLVM Class Hierarchy Reference</a>
173    <ul>
174      <li><a href="#Type">The <tt>Type</tt> class</a> </li>
175      <li><a href="#Module">The <tt>Module</tt> class</a></li>
176      <li><a href="#Value">The <tt>Value</tt> class</a>
177      <ul>
178        <li><a href="#User">The <tt>User</tt> class</a>
179        <ul>
180          <li><a href="#Instruction">The <tt>Instruction</tt> class</a></li>
181          <li><a href="#Constant">The <tt>Constant</tt> class</a>
182          <ul>
183            <li><a href="#GlobalValue">The <tt>GlobalValue</tt> class</a>
184            <ul>
185              <li><a href="#Function">The <tt>Function</tt> class</a></li>
186              <li><a href="#GlobalVariable">The <tt>GlobalVariable</tt> class</a></li>
187            </ul>
188            </li>
189          </ul>
190          </li>
191        </ul>
192        </li>
193        <li><a href="#BasicBlock">The <tt>BasicBlock</tt> class</a></li>
194        <li><a href="#Argument">The <tt>Argument</tt> class</a></li>
195      </ul>
196      </li>
197    </ul>
198  </li>
199</ol>
200
201<div class="doc_author">
202  <p>Written by <a href="mailto:sabre@nondot.org">Chris Lattner</a>,
203                <a href="mailto:dhurjati@cs.uiuc.edu">Dinakar Dhurjati</a>,
204                <a href="mailto:ggreif@gmail.com">Gabor Greif</a>,
205                <a href="mailto:jstanley@cs.uiuc.edu">Joel Stanley</a>,
206                <a href="mailto:rspencer@x10sys.com">Reid Spencer</a> and
207                <a href="mailto:owen@apple.com">Owen Anderson</a></p>
208</div>
209
210<!-- *********************************************************************** -->
211<h2>
212  <a name="introduction">Introduction </a>
213</h2>
214<!-- *********************************************************************** -->
215
216<div>
217
218<p>This document is meant to highlight some of the important classes and
219interfaces available in the LLVM source-base.  This manual is not
220intended to explain what LLVM is, how it works, and what LLVM code looks
221like.  It assumes that you know the basics of LLVM and are interested
222in writing transformations or otherwise analyzing or manipulating the
223code.</p>
224
225<p>This document should get you oriented so that you can find your
226way in the continuously growing source code that makes up the LLVM
227infrastructure. Note that this manual is not intended to serve as a
228replacement for reading the source code, so if you think there should be
229a method in one of these classes to do something, but it's not listed,
230check the source.  Links to the <a href="/doxygen/">doxygen</a> sources
231are provided to make this as easy as possible.</p>
232
233<p>The first section of this document describes general information that is
234useful to know when working in the LLVM infrastructure, and the second describes
235the Core LLVM classes.  In the future this manual will be extended with
236information describing how to use extension libraries, such as dominator
237information, CFG traversal routines, and useful utilities like the <tt><a
238href="/doxygen/InstVisitor_8h-source.html">InstVisitor</a></tt> template.</p>
239
240</div>
241
242<!-- *********************************************************************** -->
243<h2>
244  <a name="general">General Information</a>
245</h2>
246<!-- *********************************************************************** -->
247
248<div>
249
250<p>This section contains general information that is useful if you are working
251in the LLVM source-base, but that isn't specific to any particular API.</p>
252
253<!-- ======================================================================= -->
254<h3>
255  <a name="stl">The C++ Standard Template Library</a>
256</h3>
257
258<div>
259
260<p>LLVM makes heavy use of the C++ Standard Template Library (STL),
261perhaps much more than you are used to, or have seen before.  Because of
262this, you might want to do a little background reading in the
263techniques used and capabilities of the library.  There are many good
264pages that discuss the STL, and several books on the subject that you
265can get, so it will not be discussed in this document.</p>
266
267<p>Here are some useful links:</p>
268
269<ol>
270
271<li><a href="http://www.dinkumware.com/manuals/#Standard C++ Library">Dinkumware
272C++ Library reference</a> - an excellent reference for the STL and other parts
273of the standard C++ library.</li>
274
275<li><a href="http://www.tempest-sw.com/cpp/">C++ In a Nutshell</a> - This is an
276O'Reilly book in the making.  It has a decent Standard Library
277Reference that rivals Dinkumware's, and is unfortunately no longer free since the
278book has been published.</li>
279
280<li><a href="http://www.parashift.com/c++-faq-lite/">C++ Frequently Asked
281Questions</a></li>
282
283<li><a href="http://www.sgi.com/tech/stl/">SGI's STL Programmer's Guide</a> -
284Contains a useful <a
285href="http://www.sgi.com/tech/stl/stl_introduction.html">Introduction to the
286STL</a>.</li>
287
288<li><a href="http://www.research.att.com/%7Ebs/C++.html">Bjarne Stroustrup's C++
289Page</a></li>
290
291<li><a href="http://64.78.49.204/">
292Bruce Eckel's Thinking in C++, 2nd ed. Volume 2 Revision 4.0 (even better, get
293the book).</a></li>
294
295</ol>
296
297<p>You are also encouraged to take a look at the <a
298href="CodingStandards.html">LLVM Coding Standards</a> guide which focuses on how
299to write maintainable code more than where to put your curly braces.</p>
300
301</div>
302
303<!-- ======================================================================= -->
304<h3>
305  <a name="stl">Other useful references</a>
306</h3>
307
308<div>
309
310<ol>
311<li><a href="http://www.fortran-2000.com/ArnaudRecipes/sharedlib.html">Using
312static and shared libraries across platforms</a></li>
313</ol>
314
315</div>
316
317</div>
318
319<!-- *********************************************************************** -->
320<h2>
321  <a name="apis">Important and useful LLVM APIs</a>
322</h2>
323<!-- *********************************************************************** -->
324
325<div>
326
327<p>Here we highlight some LLVM APIs that are generally useful and good to
328know about when writing transformations.</p>
329
330<!-- ======================================================================= -->
331<h3>
332  <a name="isa">The <tt>isa&lt;&gt;</tt>, <tt>cast&lt;&gt;</tt> and
333  <tt>dyn_cast&lt;&gt;</tt> templates</a>
334</h3>
335
336<div>
337
338<p>The LLVM source-base makes extensive use of a custom form of RTTI.
339These templates have many similarities to the C++ <tt>dynamic_cast&lt;&gt;</tt>
340operator, but they don't have some drawbacks (primarily stemming from
341the fact that <tt>dynamic_cast&lt;&gt;</tt> only works on classes that
342have a v-table). Because they are used so often, you must know what they
343do and how they work. All of these templates are defined in the <a
344 href="/doxygen/Casting_8h-source.html"><tt>llvm/Support/Casting.h</tt></a>
345file (note that you very rarely have to include this file directly).</p>
346
347<dl>
348  <dt><tt>isa&lt;&gt;</tt>: </dt>
349
350  <dd><p>The <tt>isa&lt;&gt;</tt> operator works exactly like the Java
351  "<tt>instanceof</tt>" operator.  It returns true or false depending on whether
352  a reference or pointer points to an instance of the specified class.  This can
353  be very useful for constraint checking of various sorts (example below).</p>
354  </dd>
355
356  <dt><tt>cast&lt;&gt;</tt>: </dt>
357
358  <dd><p>The <tt>cast&lt;&gt;</tt> operator is a "checked cast" operation. It
359  converts a pointer or reference from a base class to a derived class, causing
360  an assertion failure if it is not really an instance of the right type.  This
361  should be used in cases where you have some information that makes you believe
362  that something is of the right type.  An example of the <tt>isa&lt;&gt;</tt>
363  and <tt>cast&lt;&gt;</tt> template is:</p>
364
365<div class="doc_code">
366<pre>
367static bool isLoopInvariant(const <a href="#Value">Value</a> *V, const Loop *L) {
368  if (isa&lt;<a href="#Constant">Constant</a>&gt;(V) || isa&lt;<a href="#Argument">Argument</a>&gt;(V) || isa&lt;<a href="#GlobalValue">GlobalValue</a>&gt;(V))
369    return true;
370
371  // <i>Otherwise, it must be an instruction...</i>
372  return !L-&gt;contains(cast&lt;<a href="#Instruction">Instruction</a>&gt;(V)-&gt;getParent());
373}
374</pre>
375</div>
376
377  <p>Note that you should <b>not</b> use an <tt>isa&lt;&gt;</tt> test followed
378  by a <tt>cast&lt;&gt;</tt>, for that use the <tt>dyn_cast&lt;&gt;</tt>
379  operator.</p>
380
381  </dd>
382
383  <dt><tt>dyn_cast&lt;&gt;</tt>:</dt>
384
385  <dd><p>The <tt>dyn_cast&lt;&gt;</tt> operator is a "checking cast" operation.
386  It checks to see if the operand is of the specified type, and if so, returns a
387  pointer to it (this operator does not work with references). If the operand is
388  not of the correct type, a null pointer is returned.  Thus, this works very
389  much like the <tt>dynamic_cast&lt;&gt;</tt> operator in C++, and should be
390  used in the same circumstances.  Typically, the <tt>dyn_cast&lt;&gt;</tt>
391  operator is used in an <tt>if</tt> statement or some other flow control
392  statement like this:</p>
393
394<div class="doc_code">
395<pre>
396if (<a href="#AllocationInst">AllocationInst</a> *AI = dyn_cast&lt;<a href="#AllocationInst">AllocationInst</a>&gt;(Val)) {
397  // <i>...</i>
398}
399</pre>
400</div>
401
402  <p>This form of the <tt>if</tt> statement effectively combines together a call
403  to <tt>isa&lt;&gt;</tt> and a call to <tt>cast&lt;&gt;</tt> into one
404  statement, which is very convenient.</p>
405
406  <p>Note that the <tt>dyn_cast&lt;&gt;</tt> operator, like C++'s
407  <tt>dynamic_cast&lt;&gt;</tt> or Java's <tt>instanceof</tt> operator, can be
408  abused.  In particular, you should not use big chained <tt>if/then/else</tt>
409  blocks to check for lots of different variants of classes.  If you find
410  yourself wanting to do this, it is much cleaner and more efficient to use the
411  <tt>InstVisitor</tt> class to dispatch over the instruction type directly.</p>
412
413  </dd>
414
415  <dt><tt>cast_or_null&lt;&gt;</tt>: </dt>
416
417  <dd><p>The <tt>cast_or_null&lt;&gt;</tt> operator works just like the
418  <tt>cast&lt;&gt;</tt> operator, except that it allows for a null pointer as an
419  argument (which it then propagates).  This can sometimes be useful, allowing
420  you to combine several null checks into one.</p></dd>
421
422  <dt><tt>dyn_cast_or_null&lt;&gt;</tt>: </dt>
423
424  <dd><p>The <tt>dyn_cast_or_null&lt;&gt;</tt> operator works just like the
425  <tt>dyn_cast&lt;&gt;</tt> operator, except that it allows for a null pointer
426  as an argument (which it then propagates).  This can sometimes be useful,
427  allowing you to combine several null checks into one.</p></dd>
428
429</dl>
430
431<p>These five templates can be used with any classes, whether they have a
432v-table or not.  To add support for these templates, you simply need to add
433<tt>classof</tt> static methods to the class you are interested casting
434to. Describing this is currently outside the scope of this document, but there
435are lots of examples in the LLVM source base.</p>
436
437</div>
438
439
440<!-- ======================================================================= -->
441<h3>
442  <a name="string_apis">Passing strings (the <tt>StringRef</tt>
443and <tt>Twine</tt> classes)</a>
444</h3>
445
446<div>
447
448<p>Although LLVM generally does not do much string manipulation, we do have
449several important APIs which take strings.  Two important examples are the
450Value class -- which has names for instructions, functions, etc. -- and the
451StringMap class which is used extensively in LLVM and Clang.</p>
452
453<p>These are generic classes, and they need to be able to accept strings which
454may have embedded null characters.  Therefore, they cannot simply take
455a <tt>const char *</tt>, and taking a <tt>const std::string&amp;</tt> requires
456clients to perform a heap allocation which is usually unnecessary.  Instead,
457many LLVM APIs use a <tt>StringRef</tt> or a <tt>const Twine&amp;</tt> for
458passing strings efficiently.</p>
459
460<!-- _______________________________________________________________________ -->
461<h4>
462  <a name="StringRef">The <tt>StringRef</tt> class</a>
463</h4>
464
465<div>
466
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
469on <tt>std:string</tt>, but does not require heap allocation.</p>
470
471<p>It can be implicitly constructed using a C style null-terminated string,
472an <tt>std::string</tt>, or explicitly with a character pointer and length.
473For example, the <tt>StringRef</tt> find function is declared as:</p>
474
475<pre class="doc_code">
476  iterator find(StringRef Key);
477</pre>
478
479<p>and clients can call it using any one of:</p>
480
481<pre class="doc_code">
482  Map.find("foo");                 <i>// Lookup "foo"</i>
483  Map.find(std::string("bar"));    <i>// Lookup "bar"</i>
484  Map.find(StringRef("\0baz", 4)); <i>// Lookup "\0baz"</i>
485</pre>
486
487<p>Similarly, APIs which need to return a string may return a <tt>StringRef</tt>
488instance, which can be used directly or converted to an <tt>std::string</tt>
489using the <tt>str</tt> member function.  See
490"<tt><a href="/doxygen/classllvm_1_1StringRef_8h-source.html">llvm/ADT/StringRef.h</a></tt>"
491for more information.</p>
492
493<p>You should rarely use the <tt>StringRef</tt> class directly, because it contains
494pointers to external memory it is not generally safe to store an instance of the
495class (unless you know that the external storage will not be freed). StringRef is
496small and pervasive enough in LLVM that it should always be passed by value.</p>
497
498</div>
499
500<!-- _______________________________________________________________________ -->
501<h4>
502  <a name="Twine">The <tt>Twine</tt> class</a>
503</h4>
504
505<div>
506
507<p>The <tt>Twine</tt> class is an efficient way for APIs to accept concatenated
508strings.  For example, a common LLVM paradigm is to name one instruction based on
509the name of another instruction with a suffix, for example:</p>
510
511<div class="doc_code">
512<pre>
513    New = CmpInst::Create(<i>...</i>, SO->getName() + ".cmp");
514</pre>
515</div>
516
517<p>The <tt>Twine</tt> class is effectively a
518lightweight <a href="http://en.wikipedia.org/wiki/Rope_(computer_science)">rope</a>
519which points to temporary (stack allocated) objects.  Twines can be implicitly
520constructed as the result of the plus operator applied to strings (i.e., a C
521strings, an <tt>std::string</tt>, or a <tt>StringRef</tt>).  The twine delays the
522actual concatenation of strings until it is actually required, at which point
523it can be efficiently rendered directly into a character array.  This avoids
524unnecessary heap allocation involved in constructing the temporary results of
525string concatenation. See
526"<tt><a href="/doxygen/classllvm_1_1Twine_8h-source.html">llvm/ADT/Twine.h</a></tt>"
527for more information.</p>
528
529<p>As with a <tt>StringRef</tt>, <tt>Twine</tt> objects point to external memory
530and should almost never be stored or mentioned directly.  They are intended
531solely for use when defining a function which should be able to efficiently
532accept concatenated strings.</p>
533
534</div>
535
536</div>
537
538<!-- ======================================================================= -->
539<h3>
540  <a name="DEBUG">The <tt>DEBUG()</tt> macro and <tt>-debug</tt> option</a>
541</h3>
542
543<div>
544
545<p>Often when working on your pass you will put a bunch of debugging printouts
546and other code into your pass.  After you get it working, you want to remove
547it, but you may need it again in the future (to work out new bugs that you run
548across).</p>
549
550<p> Naturally, because of this, you don't want to delete the debug printouts,
551but you don't want them to always be noisy.  A standard compromise is to comment
552them out, allowing you to enable them if you need them in the future.</p>
553
554<p>The "<tt><a href="/doxygen/Debug_8h-source.html">llvm/Support/Debug.h</a></tt>"
555file provides a macro named <tt>DEBUG()</tt> that is a much nicer solution to
556this 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
558tool) is run with the '<tt>-debug</tt>' command line argument:</p>
559
560<div class="doc_code">
561<pre>
562DEBUG(errs() &lt;&lt; "I am here!\n");
563</pre>
564</div>
565
566<p>Then you can run your pass like this:</p>
567
568<div class="doc_code">
569<pre>
570$ opt &lt; a.bc &gt; /dev/null -mypass
571<i>&lt;no output&gt;</i>
572$ opt &lt; a.bc &gt; /dev/null -mypass -debug
573I am here!
574</pre>
575</div>
576
577<p>Using the <tt>DEBUG()</tt> macro instead of a home-brewed solution allows you
578to not have to create "yet another" command line option for the debug output for
579your pass.  Note that <tt>DEBUG()</tt> macros are disabled for optimized builds,
580so they do not cause a performance impact at all (for the same reason, they
581should also not contain side-effects!).</p>
582
583<p>One additional nice thing about the <tt>DEBUG()</tt> macro is that you can
584enable 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
586program hasn't been started yet, you can always just run it with
587<tt>-debug</tt>.</p>
588
589<!-- _______________________________________________________________________ -->
590<h4>
591  <a name="DEBUG_TYPE">Fine grained debug info with <tt>DEBUG_TYPE</tt> and
592  the <tt>-debug-only</tt> option</a>
593</h4>
594
595<div>
596
597<p>Sometimes you may find yourself in a situation where enabling <tt>-debug</tt>
598just turns on <b>too much</b> information (such as when working on the code
599generator).  If you want to enable debug information with more fine-grained
600control, you define the <tt>DEBUG_TYPE</tt> macro and the <tt>-debug</tt> only
601option as follows:</p>
602
603<div class="doc_code">
604<pre>
605#undef  DEBUG_TYPE
606DEBUG(errs() &lt;&lt; "No debug type\n");
607#define DEBUG_TYPE "foo"
608DEBUG(errs() &lt;&lt; "'foo' debug type\n");
609#undef  DEBUG_TYPE
610#define DEBUG_TYPE "bar"
611DEBUG(errs() &lt;&lt; "'bar' debug type\n"));
612#undef  DEBUG_TYPE
613#define DEBUG_TYPE ""
614DEBUG(errs() &lt;&lt; "No debug type (2)\n");
615</pre>
616</div>
617
618<p>Then you can run your pass like this:</p>
619
620<div class="doc_code">
621<pre>
622$ opt &lt; a.bc &gt; /dev/null -mypass
623<i>&lt;no output&gt;</i>
624$ opt &lt; a.bc &gt; /dev/null -mypass -debug
625No debug type
626'foo' debug type
627'bar' debug type
628No debug type (2)
629$ opt &lt; a.bc &gt; /dev/null -mypass -debug-only=foo
630'foo' debug type
631$ opt &lt; a.bc &gt; /dev/null -mypass -debug-only=bar
632'bar' debug type
633</pre>
634</div>
635
636<p>Of course, in practice, you should only set <tt>DEBUG_TYPE</tt> at the top of
637a file, to specify the debug type for the entire module (if you do this before
638you <tt>#include "llvm/Support/Debug.h"</tt>, you don't have to insert the ugly
639<tt>#undef</tt>'s).  Also, you should use names more meaningful than "foo" and
640"bar", because there is no system in place to ensure that names do not
641conflict. If two different modules use the same string, they will all be turned
642on when the name is specified. This allows, for example, all debug information
643for instruction scheduling to be enabled with <tt>-debug-type=InstrSched</tt>,
644even if the source lives in multiple files.</p>
645
646<p>The <tt>DEBUG_WITH_TYPE</tt> macro is also available for situations where you
647would like to set <tt>DEBUG_TYPE</tt>, but only for one specific <tt>DEBUG</tt>
648statement. It takes an additional first parameter, which is the type to use. For
649example, the preceding example could be written as:</p>
650
651
652<div class="doc_code">
653<pre>
654DEBUG_WITH_TYPE("", errs() &lt;&lt; "No debug type\n");
655DEBUG_WITH_TYPE("foo", errs() &lt;&lt; "'foo' debug type\n");
656DEBUG_WITH_TYPE("bar", errs() &lt;&lt; "'bar' debug type\n"));
657DEBUG_WITH_TYPE("", errs() &lt;&lt; "No debug type (2)\n");
658</pre>
659</div>
660
661</div>
662
663</div>
664
665<!-- ======================================================================= -->
666<h3>
667  <a name="Statistic">The <tt>Statistic</tt> class &amp; <tt>-stats</tt>
668  option</a>
669</h3>
670
671<div>
672
673<p>The "<tt><a
674href="/doxygen/Statistic_8h-source.html">llvm/ADT/Statistic.h</a></tt>" file
675provides a class named <tt>Statistic</tt> that is used as a unified way to
676keep track of what the LLVM compiler is doing and how effective various
677optimizations are.  It is useful to see what optimizations are contributing to
678making a particular program run faster.</p>
679
680<p>Often you may run your pass on some big program, and you're interested to see
681how many times it makes a certain transformation.  Although you can do this with
682hand inspection, or some ad-hoc method, this is a real pain and not very useful
683for big programs.  Using the <tt>Statistic</tt> class makes it very easy to
684keep track of this information, and the calculated information is presented in a
685uniform manner with the rest of the passes being executed.</p>
686
687<p>There are many examples of <tt>Statistic</tt> uses, but the basics of using
688it are as follows:</p>
689
690<ol>
691    <li><p>Define your statistic like this:</p>
692
693<div class="doc_code">
694<pre>
695#define <a href="#DEBUG_TYPE">DEBUG_TYPE</a> "mypassname"   <i>// This goes before any #includes.</i>
696STATISTIC(NumXForms, "The # of times I did stuff");
697</pre>
698</div>
699
700  <p>The <tt>STATISTIC</tt> macro defines a static variable, whose name is
701    specified by the first argument.  The pass name is taken from the DEBUG_TYPE
702    macro, and the description is taken from the second argument.  The variable
703    defined ("NumXForms" in this case) acts like an unsigned integer.</p></li>
704
705    <li><p>Whenever you make a transformation, bump the counter:</p>
706
707<div class="doc_code">
708<pre>
709++NumXForms;   // <i>I did stuff!</i>
710</pre>
711</div>
712
713    </li>
714  </ol>
715
716  <p>That's all you have to do.  To get '<tt>opt</tt>' to print out the
717  statistics gathered, use the '<tt>-stats</tt>' option:</p>
718
719<div class="doc_code">
720<pre>
721$ opt -stats -mypassname &lt; program.bc &gt; /dev/null
722<i>... statistics output ...</i>
723</pre>
724</div>
725
726  <p> When running <tt>opt</tt> on a C file from the SPEC benchmark
727suite, it gives a report that looks like this:</p>
728
729<div class="doc_code">
730<pre>
731   7646 bitcodewriter   - Number of normal instructions
732    725 bitcodewriter   - Number of oversized instructions
733 129996 bitcodewriter   - Number of bitcode bytes written
734   2817 raise           - Number of insts DCEd or constprop'd
735   3213 raise           - Number of cast-of-self removed
736   5046 raise           - Number of expression trees converted
737     75 raise           - Number of other getelementptr's formed
738    138 raise           - Number of load/store peepholes
739     42 deadtypeelim    - Number of unused typenames removed from symtab
740    392 funcresolve     - Number of varargs functions resolved
741     27 globaldce       - Number of global variables removed
742      2 adce            - Number of basic blocks removed
743    134 cee             - Number of branches revectored
744     49 cee             - Number of setcc instruction eliminated
745    532 gcse            - Number of loads removed
746   2919 gcse            - Number of instructions removed
747     86 indvars         - Number of canonical indvars added
748     87 indvars         - Number of aux indvars removed
749     25 instcombine     - Number of dead inst eliminate
750    434 instcombine     - Number of insts combined
751    248 licm            - Number of load insts hoisted
752   1298 licm            - Number of insts hoisted to a loop pre-header
753      3 licm            - Number of insts hoisted to multiple loop preds (bad, no loop pre-header)
754     75 mem2reg         - Number of alloca's promoted
755   1444 cfgsimplify     - Number of blocks simplified
756</pre>
757</div>
758
759<p>Obviously, with so many optimizations, having a unified framework for this
760stuff is very nice.  Making your pass fit well into the framework makes it more
761maintainable and useful.</p>
762
763</div>
764
765<!-- ======================================================================= -->
766<h3>
767  <a name="ViewGraph">Viewing graphs while debugging code</a>
768</h3>
769
770<div>
771
772<p>Several of the important data structures in LLVM are graphs: for example
773CFGs made out of LLVM <a href="#BasicBlock">BasicBlock</a>s, CFGs made out of
774LLVM <a href="CodeGenerator.html#machinebasicblock">MachineBasicBlock</a>s, and
775<a href="CodeGenerator.html#selectiondag_intro">Instruction Selection
776DAGs</a>.  In many cases, while debugging various parts of the compiler, it is
777nice to instantly visualize these graphs.</p>
778
779<p>LLVM provides several callbacks that are available in a debug build to do
780exactly that.  If you call the <tt>Function::viewCFG()</tt> method, for example,
781the current LLVM tool will pop up a window containing the CFG for the function
782where each basic block is a node in the graph, and each node contains the
783instructions in the block.  Similarly, there also exists
784<tt>Function::viewCFGOnly()</tt> (does not include the instructions), the
785<tt>MachineFunction::viewCFG()</tt> and <tt>MachineFunction::viewCFGOnly()</tt>,
786and the <tt>SelectionDAG::viewGraph()</tt> methods.  Within GDB, for example,
787you can usually use something like <tt>call DAG.viewGraph()</tt> to pop
788up a window.  Alternatively, you can sprinkle calls to these functions in your
789code in places you want to debug.</p>
790
791<p>Getting this to work requires a small amount of configuration.  On Unix
792systems with X11, install the <a href="http://www.graphviz.org">graphviz</a>
793toolkit, and make sure 'dot' and 'gv' are in your path.  If you are running on
794Mac OS/X, download and install the Mac OS/X <a
795href="http://www.pixelglow.com/graphviz/">Graphviz program</a>, and add
796<tt>/Applications/Graphviz.app/Contents/MacOS/</tt> (or wherever you install
797it) to your path.  Once in your system and path are set up, rerun the LLVM
798configure script and rebuild LLVM to enable this functionality.</p>
799
800<p><tt>SelectionDAG</tt> has been extended to make it easier to locate
801<i>interesting</i> nodes in large complex graphs.  From gdb, if you
802<tt>call DAG.setGraphColor(<i>node</i>, "<i>color</i>")</tt>, then the
803next <tt>call DAG.viewGraph()</tt> would highlight the node in the
804specified color (choices of colors can be found at <a
805href="http://www.graphviz.org/doc/info/colors.html">colors</a>.) More
806complex node attributes can be provided with <tt>call
807DAG.setGraphAttrs(<i>node</i>, "<i>attributes</i>")</tt> (choices can be
808found at <a href="http://www.graphviz.org/doc/info/attrs.html">Graph
809Attributes</a>.)  If you want to restart and clear all the current graph
810attributes, then you can <tt>call DAG.clearGraphAttrs()</tt>. </p>
811
812<p>Note that graph visualization features are compiled out of Release builds
813to reduce file size.  This means that you need a Debug+Asserts or
814Release+Asserts build to use these features.</p>
815
816</div>
817
818</div>
819
820<!-- *********************************************************************** -->
821<h2>
822  <a name="datastructure">Picking the Right Data Structure for a Task</a>
823</h2>
824<!-- *********************************************************************** -->
825
826<div>
827
828<p>LLVM has a plethora of data structures in the <tt>llvm/ADT/</tt> directory,
829 and we commonly use STL data structures.  This section describes the trade-offs
830 you should consider when you pick one.</p>
831
832<p>
833The first step is a choose your own adventure: do you want a sequential
834container, a set-like container, or a map-like container?  The most important
835thing when choosing a container is the algorithmic properties of how you plan to
836access the container.  Based on that, you should use:</p>
837
838<ul>
839<li>a <a href="#ds_map">map-like</a> container if you need efficient look-up
840    of an value based on another value.  Map-like containers also support
841    efficient queries for containment (whether a key is in the map).  Map-like
842    containers generally do not support efficient reverse mapping (values to
843    keys).  If you need that, use two maps.  Some map-like containers also
844    support efficient iteration through the keys in sorted order.  Map-like
845    containers are the most expensive sort, only use them if you need one of
846    these capabilities.</li>
847
848<li>a <a href="#ds_set">set-like</a> container if you need to put a bunch of
849    stuff into a container that automatically eliminates duplicates.  Some
850    set-like containers support efficient iteration through the elements in
851    sorted order.  Set-like containers are more expensive than sequential
852    containers.
853</li>
854
855<li>a <a href="#ds_sequential">sequential</a> container provides
856    the most efficient way to add elements and keeps track of the order they are
857    added to the collection.  They permit duplicates and support efficient
858    iteration, but do not support efficient look-up based on a key.
859</li>
860
861<li>a <a href="#ds_string">string</a> container is a specialized sequential
862    container or reference structure that is used for character or byte
863    arrays.</li>
864
865<li>a <a href="#ds_bit">bit</a> container provides an efficient way to store and
866    perform set operations on sets of numeric id's, while automatically
867    eliminating duplicates.  Bit containers require a maximum of 1 bit for each
868    identifier you want to store.
869</li>
870</ul>
871
872<p>
873Once the proper category of container is determined, you can fine tune the
874memory use, constant factors, and cache behaviors of access by intelligently
875picking a member of the category.  Note that constant factors and cache behavior
876can be a big deal.  If you have a vector that usually only contains a few
877elements (but could contain many), for example, it's much better to use
878<a href="#dss_smallvector">SmallVector</a> than <a href="#dss_vector">vector</a>
879.  Doing so avoids (relatively) expensive malloc/free calls, which dwarf the
880cost of adding the elements to the container. </p>
881
882</div>
883
884
885<!-- ======================================================================= -->
886<h3>
887  <a name="ds_sequential">Sequential Containers (std::vector, std::list, etc)</a>
888</h3>
889
890<div>
891There are a variety of sequential containers available for you, based on your
892needs.  Pick the first in this section that will do what you want.
893
894<!-- _______________________________________________________________________ -->
895<h4>
896  <a name="dss_arrayref">llvm/ADT/ArrayRef.h</a>
897</h4>
898
899<div>
900<p>The llvm::ArrayRef class is the preferred class to use in an interface that
901   accepts a sequential list of elements in memory and just reads from them.  By
902   taking an ArrayRef, the API can be passed a fixed size array, an std::vector,
903   an llvm::SmallVector and anything else that is contiguous in memory.
904</p>
905</div>
906
907
908
909<!-- _______________________________________________________________________ -->
910<h4>
911  <a name="dss_fixedarrays">Fixed Size Arrays</a>
912</h4>
913
914<div>
915<p>Fixed size arrays are very simple and very fast.  They are good if you know
916exactly how many elements you have, or you have a (low) upper bound on how many
917you have.</p>
918</div>
919
920<!-- _______________________________________________________________________ -->
921<h4>
922  <a name="dss_heaparrays">Heap Allocated Arrays</a>
923</h4>
924
925<div>
926<p>Heap allocated arrays (new[] + delete[]) are also simple.  They are good if
927the number of elements is variable, if you know how many elements you will need
928before the array is allocated, and if the array is usually large (if not,
929consider a <a href="#dss_smallvector">SmallVector</a>).  The cost of a heap
930allocated array is the cost of the new/delete (aka malloc/free).  Also note that
931if you are allocating an array of a type with a constructor, the constructor and
932destructors will be run for every element in the array (re-sizable vectors only
933construct those elements actually used).</p>
934</div>
935
936<!-- _______________________________________________________________________ -->
937<h4>
938  <a name="dss_tinyptrvector">"llvm/ADT/TinyPtrVector.h"</a>
939</h4>
940
941
942<div>
943<p><tt>TinyPtrVector&lt;Type&gt;</tt> is a highly specialized collection class
944that is optimized to avoid allocation in the case when a vector has zero or one
945elements.  It has two major restrictions: 1) it can only hold values of pointer
946type, and 2) it cannot hold a null pointer.</p>
947
948<p>Since this container is highly specialized, it is rarely used.</p>
949
950</div>
951
952<!-- _______________________________________________________________________ -->
953<h4>
954  <a name="dss_smallvector">"llvm/ADT/SmallVector.h"</a>
955</h4>
956
957<div>
958<p><tt>SmallVector&lt;Type, N&gt;</tt> is a simple class that looks and smells
959just like <tt>vector&lt;Type&gt;</tt>:
960it supports efficient iteration, lays out elements in memory order (so you can
961do pointer arithmetic between elements), supports efficient push_back/pop_back
962operations, supports efficient random access to its elements, etc.</p>
963
964<p>The advantage of SmallVector is that it allocates space for
965some number of elements (N) <b>in the object itself</b>.  Because of this, if
966the SmallVector is dynamically smaller than N, no malloc is performed.  This can
967be a big win in cases where the malloc/free call is far more expensive than the
968code that fiddles around with the elements.</p>
969
970<p>This is good for vectors that are "usually small" (e.g. the number of
971predecessors/successors of a block is usually less than 8).  On the other hand,
972this makes the size of the SmallVector itself large, so you don't want to
973allocate lots of them (doing so will waste a lot of space).  As such,
974SmallVectors are most useful when on the stack.</p>
975
976<p>SmallVector also provides a nice portable and efficient replacement for
977<tt>alloca</tt>.</p>
978
979</div>
980
981<!-- _______________________________________________________________________ -->
982<h4>
983  <a name="dss_vector">&lt;vector&gt;</a>
984</h4>
985
986<div>
987<p>
988std::vector is well loved and respected.  It is useful when SmallVector isn't:
989when the size of the vector is often large (thus the small optimization will
990rarely be a benefit) or if you will be allocating many instances of the vector
991itself (which would waste space for elements that aren't in the container).
992vector is also useful when interfacing with code that expects vectors :).
993</p>
994
995<p>One worthwhile note about std::vector: avoid code like this:</p>
996
997<div class="doc_code">
998<pre>
999for ( ... ) {
1000   std::vector&lt;foo&gt; V;
1001   use V;
1002}
1003</pre>
1004</div>
1005
1006<p>Instead, write this as:</p>
1007
1008<div class="doc_code">
1009<pre>
1010std::vector&lt;foo&gt; V;
1011for ( ... ) {
1012   use V;
1013   V.clear();
1014}
1015</pre>
1016</div>
1017
1018<p>Doing so will save (at least) one heap allocation and free per iteration of
1019the loop.</p>
1020
1021</div>
1022
1023<!-- _______________________________________________________________________ -->
1024<h4>
1025  <a name="dss_deque">&lt;deque&gt;</a>
1026</h4>
1027
1028<div>
1029<p>std::deque is, in some senses, a generalized version of std::vector.  Like
1030std::vector, it provides constant time random access and other similar
1031properties, but it also provides efficient access to the front of the list.  It
1032does not guarantee continuity of elements within memory.</p>
1033
1034<p>In exchange for this extra flexibility, std::deque has significantly higher
1035constant factor costs than std::vector.  If possible, use std::vector or
1036something cheaper.</p>
1037</div>
1038
1039<!-- _______________________________________________________________________ -->
1040<h4>
1041  <a name="dss_list">&lt;list&gt;</a>
1042</h4>
1043
1044<div>
1045<p>std::list is an extremely inefficient class that is rarely useful.
1046It performs a heap allocation for every element inserted into it, thus having an
1047extremely high constant factor, particularly for small data types.  std::list
1048also only supports bidirectional iteration, not random access iteration.</p>
1049
1050<p>In exchange for this high cost, std::list supports efficient access to both
1051ends of the list (like std::deque, but unlike std::vector or SmallVector).  In
1052addition, the iterator invalidation characteristics of std::list are stronger
1053than that of a vector class: inserting or removing an element into the list does
1054not invalidate iterator or pointers to other elements in the list.</p>
1055</div>
1056
1057<!-- _______________________________________________________________________ -->
1058<h4>
1059  <a name="dss_ilist">llvm/ADT/ilist.h</a>
1060</h4>
1061
1062<div>
1063<p><tt>ilist&lt;T&gt;</tt> implements an 'intrusive' doubly-linked list.  It is
1064intrusive, because it requires the element to store and provide access to the
1065prev/next pointers for the list.</p>
1066
1067<p><tt>ilist</tt> has the same drawbacks as <tt>std::list</tt>, and additionally
1068requires an <tt>ilist_traits</tt> implementation for the element type, but it
1069provides some novel characteristics.  In particular, it can efficiently store
1070polymorphic objects, the traits class is informed when an element is inserted or
1071removed from the list, and <tt>ilist</tt>s are guaranteed to support a
1072constant-time splice operation.</p>
1073
1074<p>These properties are exactly what we want for things like
1075<tt>Instruction</tt>s and basic blocks, which is why these are implemented with
1076<tt>ilist</tt>s.</p>
1077
1078Related classes of interest are explained in the following subsections:
1079    <ul>
1080      <li><a href="#dss_ilist_traits">ilist_traits</a></li>
1081      <li><a href="#dss_iplist">iplist</a></li>
1082      <li><a href="#dss_ilist_node">llvm/ADT/ilist_node.h</a></li>
1083      <li><a href="#dss_ilist_sentinel">Sentinels</a></li>
1084    </ul>
1085</div>
1086
1087<!-- _______________________________________________________________________ -->
1088<h4>
1089  <a name="dss_packedvector">llvm/ADT/PackedVector.h</a>
1090</h4>
1091
1092<div>
1093<p>
1094Useful for storing a vector of values using only a few number of bits for each
1095value. Apart from the standard operations of a vector-like container, it can
1096also perform an 'or' set operation.
1097</p>
1098
1099<p>For example:</p>
1100
1101<div class="doc_code">
1102<pre>
1103enum State {
1104    None = 0x0,
1105    FirstCondition = 0x1,
1106    SecondCondition = 0x2,
1107    Both = 0x3
1108};
1109
1110State get() {
1111    PackedVector&lt;State, 2&gt; Vec1;
1112    Vec1.push_back(FirstCondition);
1113
1114    PackedVector&lt;State, 2&gt; Vec2;
1115    Vec2.push_back(SecondCondition);
1116
1117    Vec1 |= Vec2;
1118    return Vec1[0]; // returns 'Both'.
1119}
1120</pre>
1121</div>
1122
1123</div>
1124
1125<!-- _______________________________________________________________________ -->
1126<h4>
1127  <a name="dss_ilist_traits">ilist_traits</a>
1128</h4>
1129
1130<div>
1131<p><tt>ilist_traits&lt;T&gt;</tt> is <tt>ilist&lt;T&gt;</tt>'s customization
1132mechanism. <tt>iplist&lt;T&gt;</tt> (and consequently <tt>ilist&lt;T&gt;</tt>)
1133publicly derive from this traits class.</p>
1134</div>
1135
1136<!-- _______________________________________________________________________ -->
1137<h4>
1138  <a name="dss_iplist">iplist</a>
1139</h4>
1140
1141<div>
1142<p><tt>iplist&lt;T&gt;</tt> is <tt>ilist&lt;T&gt;</tt>'s base and as such
1143supports a slightly narrower interface. Notably, inserters from
1144<tt>T&amp;</tt> are absent.</p>
1145
1146<p><tt>ilist_traits&lt;T&gt;</tt> is a public base of this class and can be
1147used for a wide variety of customizations.</p>
1148</div>
1149
1150<!-- _______________________________________________________________________ -->
1151<h4>
1152  <a name="dss_ilist_node">llvm/ADT/ilist_node.h</a>
1153</h4>
1154
1155<div>
1156<p><tt>ilist_node&lt;T&gt;</tt> implements a the forward and backward links
1157that are expected by the <tt>ilist&lt;T&gt;</tt> (and analogous containers)
1158in the default manner.</p>
1159
1160<p><tt>ilist_node&lt;T&gt;</tt>s are meant to be embedded in the node type
1161<tt>T</tt>, usually <tt>T</tt> publicly derives from
1162<tt>ilist_node&lt;T&gt;</tt>.</p>
1163</div>
1164
1165<!-- _______________________________________________________________________ -->
1166<h4>
1167  <a name="dss_ilist_sentinel">Sentinels</a>
1168</h4>
1169
1170<div>
1171<p><tt>ilist</tt>s have another specialty that must be considered. To be a good
1172citizen in the C++ ecosystem, it needs to support the standard container
1173operations, such as <tt>begin</tt> and <tt>end</tt> iterators, etc. Also, the
1174<tt>operator--</tt> must work correctly on the <tt>end</tt> iterator in the
1175case of non-empty <tt>ilist</tt>s.</p>
1176
1177<p>The only sensible solution to this problem is to allocate a so-called
1178<i>sentinel</i> along with the intrusive list, which serves as the <tt>end</tt>
1179iterator, providing the back-link to the last element. However conforming to the
1180C++ convention it is illegal to <tt>operator++</tt> beyond the sentinel and it
1181also must not be dereferenced.</p>
1182
1183<p>These constraints allow for some implementation freedom to the <tt>ilist</tt>
1184how to allocate and store the sentinel. The corresponding policy is dictated
1185by <tt>ilist_traits&lt;T&gt;</tt>. By default a <tt>T</tt> gets heap-allocated
1186whenever the need for a sentinel arises.</p>
1187
1188<p>While the default policy is sufficient in most cases, it may break down when
1189<tt>T</tt> does not provide a default constructor. Also, in the case of many
1190instances of <tt>ilist</tt>s, the memory overhead of the associated sentinels
1191is wasted. To alleviate the situation with numerous and voluminous
1192<tt>T</tt>-sentinels, sometimes a trick is employed, leading to <i>ghostly
1193sentinels</i>.</p>
1194
1195<p>Ghostly sentinels are obtained by specially-crafted <tt>ilist_traits&lt;T&gt;</tt>
1196which superpose the sentinel with the <tt>ilist</tt> instance in memory. Pointer
1197arithmetic is used to obtain the sentinel, which is relative to the
1198<tt>ilist</tt>'s <tt>this</tt> pointer. The <tt>ilist</tt> is augmented by an
1199extra pointer, which serves as the back-link of the sentinel. This is the only
1200field in the ghostly sentinel which can be legally accessed.</p>
1201</div>
1202
1203<!-- _______________________________________________________________________ -->
1204<h4>
1205  <a name="dss_other">Other Sequential Container options</a>
1206</h4>
1207
1208<div>
1209<p>Other STL containers are available, such as std::string.</p>
1210
1211<p>There are also various STL adapter classes such as std::queue,
1212std::priority_queue, std::stack, etc.  These provide simplified access to an
1213underlying container but don't affect the cost of the container itself.</p>
1214
1215</div>
1216</div>
1217
1218<!-- ======================================================================= -->
1219<h3>
1220  <a name="ds_string">String-like containers</a>
1221</h3>
1222
1223<div>
1224
1225<p>
1226There are a variety of ways to pass around and use strings in C and C++, and
1227LLVM adds a few new options to choose from.  Pick the first option on this list
1228that will do what you need, they are ordered according to their relative cost.
1229</p>
1230<p>
1231Note that is is generally preferred to <em>not</em> pass strings around as
1232"<tt>const char*</tt>"'s.  These have a number of problems, including the fact
1233that they cannot represent embedded nul ("\0") characters, and do not have a
1234length available efficiently.  The general replacement for '<tt>const
1235char*</tt>' is StringRef.
1236</p>
1237
1238<p>For more information on choosing string containers for APIs, please see
1239<a href="#string_apis">Passing strings</a>.</p>
1240
1241
1242<!-- _______________________________________________________________________ -->
1243<h4>
1244  <a name="dss_stringref">llvm/ADT/StringRef.h</a>
1245</h4>
1246
1247<div>
1248<p>
1249The StringRef class is a simple value class that contains a pointer to a
1250character and a length, and is quite related to the <a
1251href="#dss_arrayref">ArrayRef</a> class (but specialized for arrays of
1252characters).  Because StringRef carries a length with it, it safely handles
1253strings with embedded nul characters in it, getting the length does not require
1254a strlen call, and it even has very convenient APIs for slicing and dicing the
1255character range that it represents.
1256</p>
1257
1258<p>
1259StringRef is ideal for passing simple strings around that are known to be live,
1260either because they are C string literals, std::string, a C array, or a
1261SmallVector.  Each of these cases has an efficient implicit conversion to
1262StringRef, which doesn't result in a dynamic strlen being executed.
1263</p>
1264
1265<p>StringRef has a few major limitations which make more powerful string
1266containers useful:</p>
1267
1268<ol>
1269<li>You cannot directly convert a StringRef to a 'const char*' because there is
1270no way to add a trailing nul (unlike the .c_str() method on various stronger
1271classes).</li>
1272
1273
1274<li>StringRef doesn't own or keep alive the underlying string bytes.
1275As such it can easily lead to dangling pointers, and is not suitable for
1276embedding in datastructures in most cases (instead, use an std::string or
1277something like that).</li>
1278
1279<li>For the same reason, StringRef cannot be used as the return value of a
1280method if the method "computes" the result string.  Instead, use
1281std::string.</li>
1282
1283<li>StringRef's do not allow you to mutate the pointed-to string bytes and it
1284doesn't allow you to insert or remove bytes from the range.  For editing
1285operations like this, it interoperates with the <a
1286href="#dss_twine">Twine</a> class.</li>
1287</ol>
1288
1289<p>Because of its strengths and limitations, it is very common for a function to
1290take a StringRef and for a method on an object to return a StringRef that
1291points into some string that it owns.</p>
1292
1293</div>
1294
1295<!-- _______________________________________________________________________ -->
1296<h4>
1297  <a name="dss_twine">llvm/ADT/Twine.h</a>
1298</h4>
1299
1300<div>
1301  <p>
1302  The Twine class is used as an intermediary datatype for APIs that want to take
1303  a string that can be constructed inline with a series of concatenations.
1304  Twine works by forming recursive instances of the Twine datatype (a simple
1305  value object) on the stack as temporary objects, linking them together into a
1306  tree which is then linearized when the Twine is consumed.  Twine is only safe
1307  to use as the argument to a function, and should always be a const reference,
1308  e.g.:
1309  </p>
1310
1311  <pre>
1312    void foo(const Twine &amp;T);
1313    ...
1314    StringRef X = ...
1315    unsigned i = ...
1316    foo(X + "." + Twine(i));
1317  </pre>
1318
1319  <p>This example forms a string like "blarg.42" by concatenating the values
1320  together, and does not form intermediate strings containing "blarg" or
1321  "blarg.".
1322  </p>
1323
1324  <p>Because Twine is constructed with temporary objects on the stack, and
1325  because these instances are destroyed at the end of the current statement,
1326  it is an inherently dangerous API.  For example, this simple variant contains
1327  undefined behavior and will probably crash:</p>
1328
1329  <pre>
1330    void foo(const Twine &amp;T);
1331    ...
1332    StringRef X = ...
1333    unsigned i = ...
1334    const Twine &amp;Tmp = X + "." + Twine(i);
1335    foo(Tmp);
1336  </pre>
1337
1338  <p>... because the temporaries are destroyed before the call.  That said,
1339  Twine's are much more efficient than intermediate std::string temporaries, and
1340  they work really well with StringRef.  Just be aware of their limitations.</p>
1341
1342</div>
1343
1344
1345<!-- _______________________________________________________________________ -->
1346<h4>
1347  <a name="dss_smallstring">llvm/ADT/SmallString.h</a>
1348</h4>
1349
1350<div>
1351
1352<p>SmallString is a subclass of <a href="#dss_smallvector">SmallVector</a> that
1353adds some convenience APIs like += that takes StringRef's.  SmallString avoids
1354allocating memory in the case when the preallocated space is enough to hold its
1355data, and it calls back to general heap allocation when required.  Since it owns
1356its data, it is very safe to use and supports full mutation of the string.</p>
1357
1358<p>Like SmallVector's, the big downside to SmallString is their sizeof.  While
1359they are optimized for small strings, they themselves are not particularly
1360small.  This means that they work great for temporary scratch buffers on the
1361stack, but should not generally be put into the heap: it is very rare to
1362see a SmallString as the member of a frequently-allocated heap data structure
1363or returned by-value.
1364</p>
1365
1366</div>
1367
1368<!-- _______________________________________________________________________ -->
1369<h4>
1370  <a name="dss_stdstring">std::string</a>
1371</h4>
1372
1373<div>
1374
1375  <p>The standard C++ std::string class is a very general class that (like
1376  SmallString) owns its underlying data.  sizeof(std::string) is very reasonable
1377  so it can be embedded into heap data structures and returned by-value.
1378  On the other hand, std::string is highly inefficient for inline editing (e.g.
1379  concatenating a bunch of stuff together) and because it is provided by the
1380  standard library, its performance characteristics depend a lot of the host
1381  standard library (e.g. libc++ and MSVC provide a highly optimized string
1382  class, GCC contains a really slow implementation).
1383  </p>
1384
1385  <p>The major disadvantage of std::string is that almost every operation that
1386  makes them larger can allocate memory, which is slow.  As such, it is better
1387  to use SmallVector or Twine as a scratch buffer, but then use std::string to
1388  persist the result.</p>
1389
1390
1391</div>
1392
1393<!-- end of strings -->
1394</div>
1395
1396
1397<!-- ======================================================================= -->
1398<h3>
1399  <a name="ds_set">Set-Like Containers (std::set, SmallSet, SetVector, etc)</a>
1400</h3>
1401
1402<div>
1403
1404<p>Set-like containers are useful when you need to canonicalize multiple values
1405into a single representation.  There are several different choices for how to do
1406this, providing various trade-offs.</p>
1407
1408<!-- _______________________________________________________________________ -->
1409<h4>
1410  <a name="dss_sortedvectorset">A sorted 'vector'</a>
1411</h4>
1412
1413<div>
1414
1415<p>If you intend to insert a lot of elements, then do a lot of queries, a
1416great approach is to use a vector (or other sequential container) with
1417std::sort+std::unique to remove duplicates.  This approach works really well if
1418your usage pattern has these two distinct phases (insert then query), and can be
1419coupled with a good choice of <a href="#ds_sequential">sequential container</a>.
1420</p>
1421
1422<p>
1423This combination provides the several nice properties: the result data is
1424contiguous in memory (good for cache locality), has few allocations, is easy to
1425address (iterators in the final vector are just indices or pointers), and can be
1426efficiently queried with a standard binary or radix search.</p>
1427
1428</div>
1429
1430<!-- _______________________________________________________________________ -->
1431<h4>
1432  <a name="dss_smallset">"llvm/ADT/SmallSet.h"</a>
1433</h4>
1434
1435<div>
1436
1437<p>If you have a set-like data structure that is usually small and whose elements
1438are reasonably small, a <tt>SmallSet&lt;Type, N&gt;</tt> is a good choice.  This set
1439has space for N elements in place (thus, if the set is dynamically smaller than
1440N, no malloc traffic is required) and accesses them with a simple linear search.
1441When the set grows beyond 'N' elements, it allocates a more expensive representation that
1442guarantees efficient access (for most types, it falls back to std::set, but for
1443pointers it uses something far better, <a
1444href="#dss_smallptrset">SmallPtrSet</a>).</p>
1445
1446<p>The magic of this class is that it handles small sets extremely efficiently,
1447but gracefully handles extremely large sets without loss of efficiency.  The
1448drawback is that the interface is quite small: it supports insertion, queries
1449and erasing, but does not support iteration.</p>
1450
1451</div>
1452
1453<!-- _______________________________________________________________________ -->
1454<h4>
1455  <a name="dss_smallptrset">"llvm/ADT/SmallPtrSet.h"</a>
1456</h4>
1457
1458<div>
1459
1460<p>SmallPtrSet has all the advantages of <tt>SmallSet</tt> (and a <tt>SmallSet</tt> of pointers is
1461transparently implemented with a <tt>SmallPtrSet</tt>), but also supports iterators.  If
1462more than 'N' insertions are performed, a single quadratically
1463probed hash table is allocated and grows as needed, providing extremely
1464efficient access (constant time insertion/deleting/queries with low constant
1465factors) and is very stingy with malloc traffic.</p>
1466
1467<p>Note that, unlike <tt>std::set</tt>, the iterators of <tt>SmallPtrSet</tt> are invalidated
1468whenever an insertion occurs.  Also, the values visited by the iterators are not
1469visited in sorted order.</p>
1470
1471</div>
1472
1473<!-- _______________________________________________________________________ -->
1474<h4>
1475  <a name="dss_denseset">"llvm/ADT/DenseSet.h"</a>
1476</h4>
1477
1478<div>
1479
1480<p>
1481DenseSet is a simple quadratically probed hash table.  It excels at supporting
1482small values: it uses a single allocation to hold all of the pairs that
1483are currently inserted in the set.  DenseSet is a great way to unique small
1484values that are not simple pointers (use <a
1485href="#dss_smallptrset">SmallPtrSet</a> for pointers).  Note that DenseSet has
1486the same requirements for the value type that <a
1487href="#dss_densemap">DenseMap</a> has.
1488</p>
1489
1490</div>
1491
1492<!-- _______________________________________________________________________ -->
1493<h4>
1494  <a name="dss_FoldingSet">"llvm/ADT/FoldingSet.h"</a>
1495</h4>
1496
1497<div>
1498
1499<p>
1500FoldingSet is an aggregate class that is really good at uniquing
1501expensive-to-create or polymorphic objects.  It is a combination of a chained
1502hash table with intrusive links (uniqued objects are required to inherit from
1503FoldingSetNode) that uses <a href="#dss_smallvector">SmallVector</a> as part of
1504its ID process.</p>
1505
1506<p>Consider a case where you want to implement a "getOrCreateFoo" method for
1507a complex object (for example, a node in the code generator).  The client has a
1508description of *what* it wants to generate (it knows the opcode and all the
1509operands), but we don't want to 'new' a node, then try inserting it into a set
1510only to find out it already exists, at which point we would have to delete it
1511and return the node that already exists.
1512</p>
1513
1514<p>To support this style of client, FoldingSet perform a query with a
1515FoldingSetNodeID (which wraps SmallVector) that can be used to describe the
1516element that we want to query for.  The query either returns the element
1517matching the ID or it returns an opaque ID that indicates where insertion should
1518take place.  Construction of the ID usually does not require heap traffic.</p>
1519
1520<p>Because FoldingSet uses intrusive links, it can support polymorphic objects
1521in the set (for example, you can have SDNode instances mixed with LoadSDNodes).
1522Because the elements are individually allocated, pointers to the elements are
1523stable: inserting or removing elements does not invalidate any pointers to other
1524elements.
1525</p>
1526
1527</div>
1528
1529<!-- _______________________________________________________________________ -->
1530<h4>
1531  <a name="dss_set">&lt;set&gt;</a>
1532</h4>
1533
1534<div>
1535
1536<p><tt>std::set</tt> is a reasonable all-around set class, which is decent at
1537many things but great at nothing.  std::set allocates memory for each element
1538inserted (thus it is very malloc intensive) and typically stores three pointers
1539per element in the set (thus adding a large amount of per-element space
1540overhead).  It offers guaranteed log(n) performance, which is not particularly
1541fast from a complexity standpoint (particularly if the elements of the set are
1542expensive to compare, like strings), and has extremely high constant factors for
1543lookup, insertion and removal.</p>
1544
1545<p>The advantages of std::set are that its iterators are stable (deleting or
1546inserting an element from the set does not affect iterators or pointers to other
1547elements) and that iteration over the set is guaranteed to be in sorted order.
1548If the elements in the set are large, then the relative overhead of the pointers
1549and malloc traffic is not a big deal, but if the elements of the set are small,
1550std::set is almost never a good choice.</p>
1551
1552</div>
1553
1554<!-- _______________________________________________________________________ -->
1555<h4>
1556  <a name="dss_setvector">"llvm/ADT/SetVector.h"</a>
1557</h4>
1558
1559<div>
1560<p>LLVM's SetVector&lt;Type&gt; is an adapter class that combines your choice of
1561a set-like container along with a <a href="#ds_sequential">Sequential
1562Container</a>.  The important property
1563that this provides is efficient insertion with uniquing (duplicate elements are
1564ignored) with iteration support.  It implements this by inserting elements into
1565both a set-like container and the sequential container, using the set-like
1566container for uniquing and the sequential container for iteration.
1567</p>
1568
1569<p>The difference between SetVector and other sets is that the order of
1570iteration is guaranteed to match the order of insertion into the SetVector.
1571This property is really important for things like sets of pointers.  Because
1572pointer values are non-deterministic (e.g. vary across runs of the program on
1573different machines), iterating over the pointers in the set will
1574not be in a well-defined order.</p>
1575
1576<p>
1577The drawback of SetVector is that it requires twice as much space as a normal
1578set and has the sum of constant factors from the set-like container and the
1579sequential container that it uses.  Use it *only* if you need to iterate over
1580the elements in a deterministic order.  SetVector is also expensive to delete
1581elements out of (linear time), unless you use it's "pop_back" method, which is
1582faster.
1583</p>
1584
1585<p><tt>SetVector</tt> is an adapter class that defaults to
1586   using <tt>std::vector</tt> and a size 16 <tt>SmallSet</tt> for the underlying
1587   containers, so it is quite expensive. However,
1588   <tt>"llvm/ADT/SetVector.h"</tt> also provides a <tt>SmallSetVector</tt>
1589   class, which defaults to using a <tt>SmallVector</tt> and <tt>SmallSet</tt>
1590   of a specified size. If you use this, and if your sets are dynamically
1591   smaller than <tt>N</tt>, you will save a lot of heap traffic.</p>
1592
1593</div>
1594
1595<!-- _______________________________________________________________________ -->
1596<h4>
1597  <a name="dss_uniquevector">"llvm/ADT/UniqueVector.h"</a>
1598</h4>
1599
1600<div>
1601
1602<p>
1603UniqueVector is similar to <a href="#dss_setvector">SetVector</a>, but it
1604retains a unique ID for each element inserted into the set.  It internally
1605contains a map and a vector, and it assigns a unique ID for each value inserted
1606into the set.</p>
1607
1608<p>UniqueVector is very expensive: its cost is the sum of the cost of
1609maintaining both the map and vector, it has high complexity, high constant
1610factors, and produces a lot of malloc traffic.  It should be avoided.</p>
1611
1612</div>
1613
1614
1615<!-- _______________________________________________________________________ -->
1616<h4>
1617  <a name="dss_otherset">Other Set-Like Container Options</a>
1618</h4>
1619
1620<div>
1621
1622<p>
1623The STL provides several other options, such as std::multiset and the various
1624"hash_set" like containers (whether from C++ TR1 or from the SGI library). We
1625never use hash_set and unordered_set because they are generally very expensive
1626(each insertion requires a malloc) and very non-portable.
1627</p>
1628
1629<p>std::multiset is useful if you're not interested in elimination of
1630duplicates, but has all the drawbacks of std::set.  A sorted vector (where you
1631don't delete duplicate entries) or some other approach is almost always
1632better.</p>
1633
1634</div>
1635
1636</div>
1637
1638<!-- ======================================================================= -->
1639<h3>
1640  <a name="ds_map">Map-Like Containers (std::map, DenseMap, etc)</a>
1641</h3>
1642
1643<div>
1644Map-like containers are useful when you want to associate data to a key.  As
1645usual, there are a lot of different ways to do this. :)
1646
1647<!-- _______________________________________________________________________ -->
1648<h4>
1649  <a name="dss_sortedvectormap">A sorted 'vector'</a>
1650</h4>
1651
1652<div>
1653
1654<p>
1655If your usage pattern follows a strict insert-then-query approach, you can
1656trivially use the same approach as <a href="#dss_sortedvectorset">sorted vectors
1657for set-like containers</a>.  The only difference is that your query function
1658(which uses std::lower_bound to get efficient log(n) lookup) should only compare
1659the key, not both the key and value.  This yields the same advantages as sorted
1660vectors for sets.
1661</p>
1662</div>
1663
1664<!-- _______________________________________________________________________ -->
1665<h4>
1666  <a name="dss_stringmap">"llvm/ADT/StringMap.h"</a>
1667</h4>
1668
1669<div>
1670
1671<p>
1672Strings are commonly used as keys in maps, and they are difficult to support
1673efficiently: they are variable length, inefficient to hash and compare when
1674long, expensive to copy, etc.  StringMap is a specialized container designed to
1675cope with these issues.  It supports mapping an arbitrary range of bytes to an
1676arbitrary other object.</p>
1677
1678<p>The StringMap implementation uses a quadratically-probed hash table, where
1679the buckets store a pointer to the heap allocated entries (and some other
1680stuff).  The entries in the map must be heap allocated because the strings are
1681variable length.  The string data (key) and the element object (value) are
1682stored in the same allocation with the string data immediately after the element
1683object.  This container guarantees the "<tt>(char*)(&amp;Value+1)</tt>" points
1684to the key string for a value.</p>
1685
1686<p>The StringMap is very fast for several reasons: quadratic probing is very
1687cache efficient for lookups, the hash value of strings in buckets is not
1688recomputed when looking up an element, StringMap rarely has to touch the
1689memory for unrelated objects when looking up a value (even when hash collisions
1690happen), hash table growth does not recompute the hash values for strings
1691already in the table, and each pair in the map is store in a single allocation
1692(the string data is stored in the same allocation as the Value of a pair).</p>
1693
1694<p>StringMap also provides query methods that take byte ranges, so it only ever
1695copies a string if a value is inserted into the table.</p>
1696</div>
1697
1698<!-- _______________________________________________________________________ -->
1699<h4>
1700  <a name="dss_indexedmap">"llvm/ADT/IndexedMap.h"</a>
1701</h4>
1702
1703<div>
1704<p>
1705IndexedMap is a specialized container for mapping small dense integers (or
1706values that can be mapped to small dense integers) to some other type.  It is
1707internally implemented as a vector with a mapping function that maps the keys to
1708the dense integer range.
1709</p>
1710
1711<p>
1712This is useful for cases like virtual registers in the LLVM code generator: they
1713have a dense mapping that is offset by a compile-time constant (the first
1714virtual register ID).</p>
1715
1716</div>
1717
1718<!-- _______________________________________________________________________ -->
1719<h4>
1720  <a name="dss_densemap">"llvm/ADT/DenseMap.h"</a>
1721</h4>
1722
1723<div>
1724
1725<p>
1726DenseMap is a simple quadratically probed hash table.  It excels at supporting
1727small keys and values: it uses a single allocation to hold all of the pairs that
1728are currently inserted in the map.  DenseMap is a great way to map pointers to
1729pointers, or map other small types to each other.
1730</p>
1731
1732<p>
1733There are several aspects of DenseMap that you should be aware of, however.  The
1734iterators in a densemap are invalidated whenever an insertion occurs, unlike
1735map.  Also, because DenseMap allocates space for a large number of key/value
1736pairs (it starts with 64 by default), it will waste a lot of space if your keys
1737or values are large.  Finally, you must implement a partial specialization of
1738DenseMapInfo for the key that you want, if it isn't already supported.  This
1739is required to tell DenseMap about two special marker values (which can never be
1740inserted into the map) that it needs internally.</p>
1741
1742</div>
1743
1744<!-- _______________________________________________________________________ -->
1745<h4>
1746  <a name="dss_valuemap">"llvm/ADT/ValueMap.h"</a>
1747</h4>
1748
1749<div>
1750
1751<p>
1752ValueMap is a wrapper around a <a href="#dss_densemap">DenseMap</a> mapping
1753Value*s (or subclasses) to another type.  When a Value is deleted or RAUW'ed,
1754ValueMap will update itself so the new version of the key is mapped to the same
1755value, just as if the key were a WeakVH.  You can configure exactly how this
1756happens, and what else happens on these two events, by passing
1757a <code>Config</code> parameter to the ValueMap template.</p>
1758
1759</div>
1760
1761<!-- _______________________________________________________________________ -->
1762<h4>
1763  <a name="dss_intervalmap">"llvm/ADT/IntervalMap.h"</a>
1764</h4>
1765
1766<div>
1767
1768<p> IntervalMap is a compact map for small keys and values. It maps key
1769intervals instead of single keys, and it will automatically coalesce adjacent
1770intervals. When then map only contains a few intervals, they are stored in the
1771map object itself to avoid allocations.</p>
1772
1773<p> The IntervalMap iterators are quite big, so they should not be passed around
1774as STL iterators. The heavyweight iterators allow a smaller data structure.</p>
1775
1776</div>
1777
1778<!-- _______________________________________________________________________ -->
1779<h4>
1780  <a name="dss_map">&lt;map&gt;</a>
1781</h4>
1782
1783<div>
1784
1785<p>
1786std::map has similar characteristics to <a href="#dss_set">std::set</a>: it uses
1787a single allocation per pair inserted into the map, it offers log(n) lookup with
1788an extremely large constant factor, imposes a space penalty of 3 pointers per
1789pair in the map, etc.</p>
1790
1791<p>std::map is most useful when your keys or values are very large, if you need
1792to iterate over the collection in sorted order, or if you need stable iterators
1793into the map (i.e. they don't get invalidated if an insertion or deletion of
1794another element takes place).</p>
1795
1796</div>
1797
1798<!-- _______________________________________________________________________ -->
1799<h4>
1800  <a name="dss_inteqclasses">"llvm/ADT/IntEqClasses.h"</a>
1801</h4>
1802
1803<div>
1804
1805<p>IntEqClasses provides a compact representation of equivalence classes of
1806small integers. Initially, each integer in the range 0..n-1 has its own
1807equivalence class. Classes can be joined by passing two class representatives to
1808the join(a, b) method. Two integers are in the same class when findLeader()
1809returns the same representative.</p>
1810
1811<p>Once all equivalence classes are formed, the map can be compressed so each
1812integer 0..n-1 maps to an equivalence class number in the range 0..m-1, where m
1813is the total number of equivalence classes. The map must be uncompressed before
1814it can be edited again.</p>
1815
1816</div>
1817
1818<!-- _______________________________________________________________________ -->
1819<h4>
1820  <a name="dss_othermap">Other Map-Like Container Options</a>
1821</h4>
1822
1823<div>
1824
1825<p>
1826The STL provides several other options, such as std::multimap and the various
1827"hash_map" like containers (whether from C++ TR1 or from the SGI library). We
1828never use hash_set and unordered_set because they are generally very expensive
1829(each insertion requires a malloc) and very non-portable.</p>
1830
1831<p>std::multimap is useful if you want to map a key to multiple values, but has
1832all the drawbacks of std::map.  A sorted vector or some other approach is almost
1833always better.</p>
1834
1835</div>
1836
1837</div>
1838
1839<!-- ======================================================================= -->
1840<h3>
1841  <a name="ds_bit">Bit storage containers (BitVector, SparseBitVector)</a>
1842</h3>
1843
1844<div>
1845<p>Unlike the other containers, there are only two bit storage containers, and
1846choosing when to use each is relatively straightforward.</p>
1847
1848<p>One additional option is
1849<tt>std::vector&lt;bool&gt;</tt>: we discourage its use for two reasons 1) the
1850implementation in many common compilers (e.g. commonly available versions of
1851GCC) is extremely inefficient and 2) the C++ standards committee is likely to
1852deprecate this container and/or change it significantly somehow.  In any case,
1853please don't use it.</p>
1854
1855<!-- _______________________________________________________________________ -->
1856<h4>
1857  <a name="dss_bitvector">BitVector</a>
1858</h4>
1859
1860<div>
1861<p> The BitVector container provides a dynamic size set of bits for manipulation.
1862It supports individual bit setting/testing, as well as set operations.  The set
1863operations take time O(size of bitvector), but operations are performed one word
1864at a time, instead of one bit at a time.  This makes the BitVector very fast for
1865set operations compared to other containers.  Use the BitVector when you expect
1866the number of set bits to be high (IE a dense set).
1867</p>
1868</div>
1869
1870<!-- _______________________________________________________________________ -->
1871<h4>
1872  <a name="dss_smallbitvector">SmallBitVector</a>
1873</h4>
1874
1875<div>
1876<p> The SmallBitVector container provides the same interface as BitVector, but
1877it is optimized for the case where only a small number of bits, less than
187825 or so, are needed. It also transparently supports larger bit counts, but
1879slightly less efficiently than a plain BitVector, so SmallBitVector should
1880only be used when larger counts are rare.
1881</p>
1882
1883<p>
1884At this time, SmallBitVector does not support set operations (and, or, xor),
1885and its operator[] does not provide an assignable lvalue.
1886</p>
1887</div>
1888
1889<!-- _______________________________________________________________________ -->
1890<h4>
1891  <a name="dss_sparsebitvector">SparseBitVector</a>
1892</h4>
1893
1894<div>
1895<p> The SparseBitVector container is much like BitVector, with one major
1896difference: Only the bits that are set, are stored.  This makes the
1897SparseBitVector much more space efficient than BitVector when the set is sparse,
1898as well as making set operations O(number of set bits) instead of O(size of
1899universe).  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
1900(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).
1901</p>
1902</div>
1903
1904</div>
1905
1906</div>
1907
1908<!-- *********************************************************************** -->
1909<h2>
1910  <a name="common">Helpful Hints for Common Operations</a>
1911</h2>
1912<!-- *********************************************************************** -->
1913
1914<div>
1915
1916<p>This section describes how to perform some very simple transformations of
1917LLVM code.  This is meant to give examples of common idioms used, showing the
1918practical side of LLVM transformations.  <p> Because this is a "how-to" section,
1919you should also read about the main classes that you will be working with.  The
1920<a href="#coreclasses">Core LLVM Class Hierarchy Reference</a> contains details
1921and descriptions of the main classes that you should know about.</p>
1922
1923<!-- NOTE: this section should be heavy on example code -->
1924<!-- ======================================================================= -->
1925<h3>
1926  <a name="inspection">Basic Inspection and Traversal Routines</a>
1927</h3>
1928
1929<div>
1930
1931<p>The LLVM compiler infrastructure have many different data structures that may
1932be traversed.  Following the example of the C++ standard template library, the
1933techniques used to traverse these various data structures are all basically the
1934same.  For a enumerable sequence of values, the <tt>XXXbegin()</tt> function (or
1935method) returns an iterator to the start of the sequence, the <tt>XXXend()</tt>
1936function returns an iterator pointing to one past the last valid element of the
1937sequence, and there is some <tt>XXXiterator</tt> data type that is common
1938between the two operations.</p>
1939
1940<p>Because the pattern for iteration is common across many different aspects of
1941the program representation, the standard template library algorithms may be used
1942on them, and it is easier to remember how to iterate. First we show a few common
1943examples of the data structures that need to be traversed.  Other data
1944structures are traversed in very similar ways.</p>
1945
1946<!-- _______________________________________________________________________ -->
1947<h4>
1948  <a name="iterate_function">Iterating over the </a><a
1949  href="#BasicBlock"><tt>BasicBlock</tt></a>s in a <a
1950  href="#Function"><tt>Function</tt></a>
1951</h4>
1952
1953<div>
1954
1955<p>It's quite common to have a <tt>Function</tt> instance that you'd like to
1956transform in some way; in particular, you'd like to manipulate its
1957<tt>BasicBlock</tt>s.  To facilitate this, you'll need to iterate over all of
1958the <tt>BasicBlock</tt>s that constitute the <tt>Function</tt>. The following is
1959an example that prints the name of a <tt>BasicBlock</tt> and the number of
1960<tt>Instruction</tt>s it contains:</p>
1961
1962<div class="doc_code">
1963<pre>
1964// <i>func is a pointer to a Function instance</i>
1965for (Function::iterator i = func-&gt;begin(), e = func-&gt;end(); i != e; ++i)
1966  // <i>Print out the name of the basic block if it has one, and then the</i>
1967  // <i>number of instructions that it contains</i>
1968  errs() &lt;&lt; "Basic block (name=" &lt;&lt; i-&gt;getName() &lt;&lt; ") has "
1969             &lt;&lt; i-&gt;size() &lt;&lt; " instructions.\n";
1970</pre>
1971</div>
1972
1973<p>Note that i can be used as if it were a pointer for the purposes of
1974invoking member functions of the <tt>Instruction</tt> class.  This is
1975because the indirection operator is overloaded for the iterator
1976classes.  In the above code, the expression <tt>i-&gt;size()</tt> is
1977exactly equivalent to <tt>(*i).size()</tt> just like you'd expect.</p>
1978
1979</div>
1980
1981<!-- _______________________________________________________________________ -->
1982<h4>
1983  <a name="iterate_basicblock">Iterating over the </a><a
1984  href="#Instruction"><tt>Instruction</tt></a>s in a <a
1985  href="#BasicBlock"><tt>BasicBlock</tt></a>
1986</h4>
1987
1988<div>
1989
1990<p>Just like when dealing with <tt>BasicBlock</tt>s in <tt>Function</tt>s, it's
1991easy to iterate over the individual instructions that make up
1992<tt>BasicBlock</tt>s. Here's a code snippet that prints out each instruction in
1993a <tt>BasicBlock</tt>:</p>
1994
1995<div class="doc_code">
1996<pre>
1997// <i>blk is a pointer to a BasicBlock instance</i>
1998for (BasicBlock::iterator i = blk-&gt;begin(), e = blk-&gt;end(); i != e; ++i)
1999   // <i>The next statement works since operator&lt;&lt;(ostream&amp;,...)</i>
2000   // <i>is overloaded for Instruction&amp;</i>
2001   errs() &lt;&lt; *i &lt;&lt; "\n";
2002</pre>
2003</div>
2004
2005<p>However, this isn't really the best way to print out the contents of a
2006<tt>BasicBlock</tt>!  Since the ostream operators are overloaded for virtually
2007anything you'll care about, you could have just invoked the print routine on the
2008basic block itself: <tt>errs() &lt;&lt; *blk &lt;&lt; "\n";</tt>.</p>
2009
2010</div>
2011
2012<!-- _______________________________________________________________________ -->
2013<h4>
2014  <a name="iterate_institer">Iterating over the </a><a
2015  href="#Instruction"><tt>Instruction</tt></a>s in a <a
2016  href="#Function"><tt>Function</tt></a>
2017</h4>
2018
2019<div>
2020
2021<p>If you're finding that you commonly iterate over a <tt>Function</tt>'s
2022<tt>BasicBlock</tt>s and then that <tt>BasicBlock</tt>'s <tt>Instruction</tt>s,
2023<tt>InstIterator</tt> should be used instead. You'll need to include <a
2024href="/doxygen/InstIterator_8h-source.html"><tt>llvm/Support/InstIterator.h</tt></a>,
2025and then instantiate <tt>InstIterator</tt>s explicitly in your code.  Here's a
2026small example that shows how to dump all instructions in a function to the standard error stream:<p>
2027
2028<div class="doc_code">
2029<pre>
2030#include "<a href="/doxygen/InstIterator_8h-source.html">llvm/Support/InstIterator.h</a>"
2031
2032// <i>F is a pointer to a Function instance</i>
2033for (inst_iterator I = inst_begin(F), E = inst_end(F); I != E; ++I)
2034  errs() &lt;&lt; *I &lt;&lt; "\n";
2035</pre>
2036</div>
2037
2038<p>Easy, isn't it?  You can also use <tt>InstIterator</tt>s to fill a
2039work list with its initial contents.  For example, if you wanted to
2040initialize a work list to contain all instructions in a <tt>Function</tt>
2041F, all you would need to do is something like:</p>
2042
2043<div class="doc_code">
2044<pre>
2045std::set&lt;Instruction*&gt; worklist;
2046// or better yet, SmallPtrSet&lt;Instruction*, 64&gt; worklist;
2047
2048for (inst_iterator I = inst_begin(F), E = inst_end(F); I != E; ++I)
2049   worklist.insert(&amp;*I);
2050</pre>
2051</div>
2052
2053<p>The STL set <tt>worklist</tt> would now contain all instructions in the
2054<tt>Function</tt> pointed to by F.</p>
2055
2056</div>
2057
2058<!-- _______________________________________________________________________ -->
2059<h4>
2060  <a name="iterate_convert">Turning an iterator into a class pointer (and
2061  vice-versa)</a>
2062</h4>
2063
2064<div>
2065
2066<p>Sometimes, it'll be useful to grab a reference (or pointer) to a class
2067instance when all you've got at hand is an iterator.  Well, extracting
2068a reference or a pointer from an iterator is very straight-forward.
2069Assuming that <tt>i</tt> is a <tt>BasicBlock::iterator</tt> and <tt>j</tt>
2070is a <tt>BasicBlock::const_iterator</tt>:</p>
2071
2072<div class="doc_code">
2073<pre>
2074Instruction&amp; inst = *i;   // <i>Grab reference to instruction reference</i>
2075Instruction* pinst = &amp;*i; // <i>Grab pointer to instruction reference</i>
2076const Instruction&amp; inst = *j;
2077</pre>
2078</div>
2079
2080<p>However, the iterators you'll be working with in the LLVM framework are
2081special: they will automatically convert to a ptr-to-instance type whenever they
2082need to.  Instead of dereferencing the iterator and then taking the address of
2083the result, you can simply assign the iterator to the proper pointer type and
2084you get the dereference and address-of operation as a result of the assignment
2085(behind the scenes, this is a result of overloading casting mechanisms).  Thus
2086the last line of the last example,</p>
2087
2088<div class="doc_code">
2089<pre>
2090Instruction *pinst = &amp;*i;
2091</pre>
2092</div>
2093
2094<p>is semantically equivalent to</p>
2095
2096<div class="doc_code">
2097<pre>
2098Instruction *pinst = i;
2099</pre>
2100</div>
2101
2102<p>It's also possible to turn a class pointer into the corresponding iterator,
2103and this is a constant time operation (very efficient).  The following code
2104snippet illustrates use of the conversion constructors provided by LLVM
2105iterators.  By using these, you can explicitly grab the iterator of something
2106without actually obtaining it via iteration over some structure:</p>
2107
2108<div class="doc_code">
2109<pre>
2110void printNextInstruction(Instruction* inst) {
2111  BasicBlock::iterator it(inst);
2112  ++it; // <i>After this line, it refers to the instruction after *inst</i>
2113  if (it != inst-&gt;getParent()-&gt;end()) errs() &lt;&lt; *it &lt;&lt; "\n";
2114}
2115</pre>
2116</div>
2117
2118<p>Unfortunately, these implicit conversions come at a cost; they prevent
2119these iterators from conforming to standard iterator conventions, and thus
2120from being usable with standard algorithms and containers. For example, they
2121prevent the following code, where <tt>B</tt> is a <tt>BasicBlock</tt>,
2122from compiling:</p>
2123
2124<div class="doc_code">
2125<pre>
2126  llvm::SmallVector&lt;llvm::Instruction *, 16&gt;(B-&gt;begin(), B-&gt;end());
2127</pre>
2128</div>
2129
2130<p>Because of this, these implicit conversions may be removed some day,
2131and <tt>operator*</tt> changed to return a pointer instead of a reference.</p>
2132
2133</div>
2134
2135<!--_______________________________________________________________________-->
2136<h4>
2137  <a name="iterate_complex">Finding call sites: a slightly more complex
2138  example</a>
2139</h4>
2140
2141<div>
2142
2143<p>Say that you're writing a FunctionPass and would like to count all the
2144locations in the entire module (that is, across every <tt>Function</tt>) where a
2145certain function (i.e., some <tt>Function</tt>*) is already in scope.  As you'll
2146learn later, you may want to use an <tt>InstVisitor</tt> to accomplish this in a
2147much more straight-forward manner, but this example will allow us to explore how
2148you'd do it if you didn't have <tt>InstVisitor</tt> around. In pseudo-code, this
2149is what we want to do:</p>
2150
2151<div class="doc_code">
2152<pre>
2153initialize callCounter to zero
2154for each Function f in the Module
2155  for each BasicBlock b in f
2156    for each Instruction i in b
2157      if (i is a CallInst and calls the given function)
2158        increment callCounter
2159</pre>
2160</div>
2161
2162<p>And the actual code is (remember, because we're writing a
2163<tt>FunctionPass</tt>, our <tt>FunctionPass</tt>-derived class simply has to
2164override the <tt>runOnFunction</tt> method):</p>
2165
2166<div class="doc_code">
2167<pre>
2168Function* targetFunc = ...;
2169
2170class OurFunctionPass : public FunctionPass {
2171  public:
2172    OurFunctionPass(): callCounter(0) { }
2173
2174    virtual runOnFunction(Function&amp; F) {
2175      for (Function::iterator b = F.begin(), be = F.end(); b != be; ++b) {
2176        for (BasicBlock::iterator i = b-&gt;begin(), ie = b-&gt;end(); i != ie; ++i) {
2177          if (<a href="#CallInst">CallInst</a>* callInst = <a href="#isa">dyn_cast</a>&lt;<a
2178 href="#CallInst">CallInst</a>&gt;(&amp;*i)) {
2179            // <i>We know we've encountered a call instruction, so we</i>
2180            // <i>need to determine if it's a call to the</i>
2181            // <i>function pointed to by m_func or not.</i>
2182            if (callInst-&gt;getCalledFunction() == targetFunc)
2183              ++callCounter;
2184          }
2185        }
2186      }
2187    }
2188
2189  private:
2190    unsigned callCounter;
2191};
2192</pre>
2193</div>
2194
2195</div>
2196
2197<!--_______________________________________________________________________-->
2198<h4>
2199  <a name="calls_and_invokes">Treating calls and invokes the same way</a>
2200</h4>
2201
2202<div>
2203
2204<p>You may have noticed that the previous example was a bit oversimplified in
2205that it did not deal with call sites generated by 'invoke' instructions. In
2206this, and in other situations, you may find that you want to treat
2207<tt>CallInst</tt>s and <tt>InvokeInst</tt>s the same way, even though their
2208most-specific common base class is <tt>Instruction</tt>, which includes lots of
2209less closely-related things. For these cases, LLVM provides a handy wrapper
2210class called <a
2211href="http://llvm.org/doxygen/classllvm_1_1CallSite.html"><tt>CallSite</tt></a>.
2212It is essentially a wrapper around an <tt>Instruction</tt> pointer, with some
2213methods that provide functionality common to <tt>CallInst</tt>s and
2214<tt>InvokeInst</tt>s.</p>
2215
2216<p>This class has "value semantics": it should be passed by value, not by
2217reference and it should not be dynamically allocated or deallocated using
2218<tt>operator new</tt> or <tt>operator delete</tt>. It is efficiently copyable,
2219assignable and constructable, with costs equivalents to that of a bare pointer.
2220If you look at its definition, it has only a single pointer member.</p>
2221
2222</div>
2223
2224<!--_______________________________________________________________________-->
2225<h4>
2226  <a name="iterate_chains">Iterating over def-use &amp; use-def chains</a>
2227</h4>
2228
2229<div>
2230
2231<p>Frequently, we might have an instance of the <a
2232href="/doxygen/classllvm_1_1Value.html">Value Class</a> and we want to
2233determine which <tt>User</tt>s use the <tt>Value</tt>.  The list of all
2234<tt>User</tt>s of a particular <tt>Value</tt> is called a <i>def-use</i> chain.
2235For example, let's say we have a <tt>Function*</tt> named <tt>F</tt> to a
2236particular function <tt>foo</tt>. Finding all of the instructions that
2237<i>use</i> <tt>foo</tt> is as simple as iterating over the <i>def-use</i> chain
2238of <tt>F</tt>:</p>
2239
2240<div class="doc_code">
2241<pre>
2242Function *F = ...;
2243
2244for (Value::use_iterator i = F-&gt;use_begin(), e = F-&gt;use_end(); i != e; ++i)
2245  if (Instruction *Inst = dyn_cast&lt;Instruction&gt;(*i)) {
2246    errs() &lt;&lt; "F is used in instruction:\n";
2247    errs() &lt;&lt; *Inst &lt;&lt; "\n";
2248  }
2249</pre>
2250</div>
2251
2252<p>Note that dereferencing a <tt>Value::use_iterator</tt> is not a very cheap
2253operation. Instead of performing <tt>*i</tt> above several times, consider
2254doing it only once in the loop body and reusing its result.</p>
2255
2256<p>Alternatively, it's common to have an instance of the <a
2257href="/doxygen/classllvm_1_1User.html">User Class</a> and need to know what
2258<tt>Value</tt>s are used by it.  The list of all <tt>Value</tt>s used by a
2259<tt>User</tt> is known as a <i>use-def</i> chain.  Instances of class
2260<tt>Instruction</tt> are common <tt>User</tt>s, so we might want to iterate over
2261all of the values that a particular instruction uses (that is, the operands of
2262the particular <tt>Instruction</tt>):</p>
2263
2264<div class="doc_code">
2265<pre>
2266Instruction *pi = ...;
2267
2268for (User::op_iterator i = pi-&gt;op_begin(), e = pi-&gt;op_end(); i != e; ++i) {
2269  Value *v = *i;
2270  // <i>...</i>
2271}
2272</pre>
2273</div>
2274
2275<p>Declaring objects as <tt>const</tt> is an important tool of enforcing
2276mutation free algorithms (such as analyses, etc.). For this purpose above
2277iterators come in constant flavors as <tt>Value::const_use_iterator</tt>
2278and <tt>Value::const_op_iterator</tt>.  They automatically arise when
2279calling <tt>use/op_begin()</tt> on <tt>const Value*</tt>s or
2280<tt>const User*</tt>s respectively.  Upon dereferencing, they return
2281<tt>const Use*</tt>s. Otherwise the above patterns remain unchanged.</p>
2282
2283</div>
2284
2285<!--_______________________________________________________________________-->
2286<h4>
2287  <a name="iterate_preds">Iterating over predecessors &amp;
2288successors of blocks</a>
2289</h4>
2290
2291<div>
2292
2293<p>Iterating over the predecessors and successors of a block is quite easy
2294with the routines defined in <tt>"llvm/Support/CFG.h"</tt>.  Just use code like
2295this to iterate over all predecessors of BB:</p>
2296
2297<div class="doc_code">
2298<pre>
2299#include "llvm/Support/CFG.h"
2300BasicBlock *BB = ...;
2301
2302for (pred_iterator PI = pred_begin(BB), E = pred_end(BB); PI != E; ++PI) {
2303  BasicBlock *Pred = *PI;
2304  // <i>...</i>
2305}
2306</pre>
2307</div>
2308
2309<p>Similarly, to iterate over successors use
2310succ_iterator/succ_begin/succ_end.</p>
2311
2312</div>
2313
2314</div>
2315
2316<!-- ======================================================================= -->
2317<h3>
2318  <a name="simplechanges">Making simple changes</a>
2319</h3>
2320
2321<div>
2322
2323<p>There are some primitive transformation operations present in the LLVM
2324infrastructure that are worth knowing about.  When performing
2325transformations, it's fairly common to manipulate the contents of basic
2326blocks. This section describes some of the common methods for doing so
2327and gives example code.</p>
2328
2329<!--_______________________________________________________________________-->
2330<h4>
2331  <a name="schanges_creating">Creating and inserting new
2332  <tt>Instruction</tt>s</a>
2333</h4>
2334
2335<div>
2336
2337<p><i>Instantiating Instructions</i></p>
2338
2339<p>Creation of <tt>Instruction</tt>s is straight-forward: simply call the
2340constructor for the kind of instruction to instantiate and provide the necessary
2341parameters. For example, an <tt>AllocaInst</tt> only <i>requires</i> a
2342(const-ptr-to) <tt>Type</tt>. Thus:</p>
2343
2344<div class="doc_code">
2345<pre>
2346AllocaInst* ai = new AllocaInst(Type::Int32Ty);
2347</pre>
2348</div>
2349
2350<p>will create an <tt>AllocaInst</tt> instance that represents the allocation of
2351one integer in the current stack frame, at run time. Each <tt>Instruction</tt>
2352subclass is likely to have varying default parameters which change the semantics
2353of the instruction, so refer to the <a
2354href="/doxygen/classllvm_1_1Instruction.html">doxygen documentation for the subclass of
2355Instruction</a> that you're interested in instantiating.</p>
2356
2357<p><i>Naming values</i></p>
2358
2359<p>It is very useful to name the values of instructions when you're able to, as
2360this facilitates the debugging of your transformations.  If you end up looking
2361at generated LLVM machine code, you definitely want to have logical names
2362associated with the results of instructions!  By supplying a value for the
2363<tt>Name</tt> (default) parameter of the <tt>Instruction</tt> constructor, you
2364associate a logical name with the result of the instruction's execution at
2365run time.  For example, say that I'm writing a transformation that dynamically
2366allocates space for an integer on the stack, and that integer is going to be
2367used as some kind of index by some other code.  To accomplish this, I place an
2368<tt>AllocaInst</tt> at the first point in the first <tt>BasicBlock</tt> of some
2369<tt>Function</tt>, and I'm intending to use it within the same
2370<tt>Function</tt>. I might do:</p>
2371
2372<div class="doc_code">
2373<pre>
2374AllocaInst* pa = new AllocaInst(Type::Int32Ty, 0, "indexLoc");
2375</pre>
2376</div>
2377
2378<p>where <tt>indexLoc</tt> is now the logical name of the instruction's
2379execution value, which is a pointer to an integer on the run time stack.</p>
2380
2381<p><i>Inserting instructions</i></p>
2382
2383<p>There are essentially two ways to insert an <tt>Instruction</tt>
2384into an existing sequence of instructions that form a <tt>BasicBlock</tt>:</p>
2385
2386<ul>
2387  <li>Insertion into an explicit instruction list
2388
2389    <p>Given a <tt>BasicBlock* pb</tt>, an <tt>Instruction* pi</tt> within that
2390    <tt>BasicBlock</tt>, and a newly-created instruction we wish to insert
2391    before <tt>*pi</tt>, we do the following: </p>
2392
2393<div class="doc_code">
2394<pre>
2395BasicBlock *pb = ...;
2396Instruction *pi = ...;
2397Instruction *newInst = new Instruction(...);
2398
2399pb-&gt;getInstList().insert(pi, newInst); // <i>Inserts newInst before pi in pb</i>
2400</pre>
2401</div>
2402
2403    <p>Appending to the end of a <tt>BasicBlock</tt> is so common that
2404    the <tt>Instruction</tt> class and <tt>Instruction</tt>-derived
2405    classes provide constructors which take a pointer to a
2406    <tt>BasicBlock</tt> to be appended to. For example code that
2407    looked like: </p>
2408
2409<div class="doc_code">
2410<pre>
2411BasicBlock *pb = ...;
2412Instruction *newInst = new Instruction(...);
2413
2414pb-&gt;getInstList().push_back(newInst); // <i>Appends newInst to pb</i>
2415</pre>
2416</div>
2417
2418    <p>becomes: </p>
2419
2420<div class="doc_code">
2421<pre>
2422BasicBlock *pb = ...;
2423Instruction *newInst = new Instruction(..., pb);
2424</pre>
2425</div>
2426
2427    <p>which is much cleaner, especially if you are creating
2428    long instruction streams.</p></li>
2429
2430  <li>Insertion into an implicit instruction list
2431
2432    <p><tt>Instruction</tt> instances that are already in <tt>BasicBlock</tt>s
2433    are implicitly associated with an existing instruction list: the instruction
2434    list of the enclosing basic block. Thus, we could have accomplished the same
2435    thing as the above code without being given a <tt>BasicBlock</tt> by doing:
2436    </p>
2437
2438<div class="doc_code">
2439<pre>
2440Instruction *pi = ...;
2441Instruction *newInst = new Instruction(...);
2442
2443pi-&gt;getParent()-&gt;getInstList().insert(pi, newInst);
2444</pre>
2445</div>
2446
2447    <p>In fact, this sequence of steps occurs so frequently that the
2448    <tt>Instruction</tt> class and <tt>Instruction</tt>-derived classes provide
2449    constructors which take (as a default parameter) a pointer to an
2450    <tt>Instruction</tt> which the newly-created <tt>Instruction</tt> should
2451    precede.  That is, <tt>Instruction</tt> constructors are capable of
2452    inserting the newly-created instance into the <tt>BasicBlock</tt> of a
2453    provided instruction, immediately before that instruction.  Using an
2454    <tt>Instruction</tt> constructor with a <tt>insertBefore</tt> (default)
2455    parameter, the above code becomes:</p>
2456
2457<div class="doc_code">
2458<pre>
2459Instruction* pi = ...;
2460Instruction* newInst = new Instruction(..., pi);
2461</pre>
2462</div>
2463
2464    <p>which is much cleaner, especially if you're creating a lot of
2465    instructions and adding them to <tt>BasicBlock</tt>s.</p></li>
2466</ul>
2467
2468</div>
2469
2470<!--_______________________________________________________________________-->
2471<h4>
2472  <a name="schanges_deleting">Deleting <tt>Instruction</tt>s</a>
2473</h4>
2474
2475<div>
2476
2477<p>Deleting an instruction from an existing sequence of instructions that form a
2478<a href="#BasicBlock"><tt>BasicBlock</tt></a> is very straight-forward: just
2479call the instruction's eraseFromParent() method.  For example:</p>
2480
2481<div class="doc_code">
2482<pre>
2483<a href="#Instruction">Instruction</a> *I = .. ;
2484I-&gt;eraseFromParent();
2485</pre>
2486</div>
2487
2488<p>This unlinks the instruction from its containing basic block and deletes
2489it.  If you'd just like to unlink the instruction from its containing basic
2490block but not delete it, you can use the <tt>removeFromParent()</tt> method.</p>
2491
2492</div>
2493
2494<!--_______________________________________________________________________-->
2495<h4>
2496  <a name="schanges_replacing">Replacing an <tt>Instruction</tt> with another
2497  <tt>Value</tt></a>
2498</h4>
2499
2500<div>
2501
2502<p><i>Replacing individual instructions</i></p>
2503
2504<p>Including "<a href="/doxygen/BasicBlockUtils_8h-source.html">llvm/Transforms/Utils/BasicBlockUtils.h</a>"
2505permits use of two very useful replace functions: <tt>ReplaceInstWithValue</tt>
2506and <tt>ReplaceInstWithInst</tt>.</p>
2507
2508<h5><a name="schanges_deleting">Deleting <tt>Instruction</tt>s</a></h5>
2509
2510<ul>
2511  <li><tt>ReplaceInstWithValue</tt>
2512
2513    <p>This function replaces all uses of a given instruction with a value,
2514    and then removes the original instruction. The following example
2515    illustrates the replacement of the result of a particular
2516    <tt>AllocaInst</tt> that allocates memory for a single integer with a null
2517    pointer to an integer.</p>
2518
2519<div class="doc_code">
2520<pre>
2521AllocaInst* instToReplace = ...;
2522BasicBlock::iterator ii(instToReplace);
2523
2524ReplaceInstWithValue(instToReplace-&gt;getParent()-&gt;getInstList(), ii,
2525                     Constant::getNullValue(PointerType::getUnqual(Type::Int32Ty)));
2526</pre></div></li>
2527
2528  <li><tt>ReplaceInstWithInst</tt>
2529
2530    <p>This function replaces a particular instruction with another
2531    instruction, inserting the new instruction into the basic block at the
2532    location where the old instruction was, and replacing any uses of the old
2533    instruction with the new instruction. The following example illustrates
2534    the replacement of one <tt>AllocaInst</tt> with another.</p>
2535
2536<div class="doc_code">
2537<pre>
2538AllocaInst* instToReplace = ...;
2539BasicBlock::iterator ii(instToReplace);
2540
2541ReplaceInstWithInst(instToReplace-&gt;getParent()-&gt;getInstList(), ii,
2542                    new AllocaInst(Type::Int32Ty, 0, "ptrToReplacedInt"));
2543</pre></div></li>
2544</ul>
2545
2546<p><i>Replacing multiple uses of <tt>User</tt>s and <tt>Value</tt>s</i></p>
2547
2548<p>You can use <tt>Value::replaceAllUsesWith</tt> and
2549<tt>User::replaceUsesOfWith</tt> to change more than one use at a time.  See the
2550doxygen documentation for the <a href="/doxygen/classllvm_1_1Value.html">Value Class</a>
2551and <a href="/doxygen/classllvm_1_1User.html">User Class</a>, respectively, for more
2552information.</p>
2553
2554<!-- Value::replaceAllUsesWith User::replaceUsesOfWith Point out:
2555include/llvm/Transforms/Utils/ especially BasicBlockUtils.h with:
2556ReplaceInstWithValue, ReplaceInstWithInst -->
2557
2558</div>
2559
2560<!--_______________________________________________________________________-->
2561<h4>
2562  <a name="schanges_deletingGV">Deleting <tt>GlobalVariable</tt>s</a>
2563</h4>
2564
2565<div>
2566
2567<p>Deleting a global variable from a module is just as easy as deleting an
2568Instruction. First, you must have a pointer to the global variable that you wish
2569 to delete.  You use this pointer to erase it from its parent, the module.
2570 For example:</p>
2571
2572<div class="doc_code">
2573<pre>
2574<a href="#GlobalVariable">GlobalVariable</a> *GV = .. ;
2575
2576GV-&gt;eraseFromParent();
2577</pre>
2578</div>
2579
2580</div>
2581
2582</div>
2583
2584<!-- ======================================================================= -->
2585<h3>
2586  <a name="create_types">How to Create Types</a>
2587</h3>
2588
2589<div>
2590
2591<p>In generating IR, you may need some complex types.  If you know these types
2592statically, you can use <tt>TypeBuilder&lt;...&gt;::get()</tt>, defined
2593in <tt>llvm/Support/TypeBuilder.h</tt>, to retrieve them.  <tt>TypeBuilder</tt>
2594has two forms depending on whether you're building types for cross-compilation
2595or native library use.  <tt>TypeBuilder&lt;T, true&gt;</tt> requires
2596that <tt>T</tt> be independent of the host environment, meaning that it's built
2597out of types from
2598the <a href="/doxygen/namespacellvm_1_1types.html"><tt>llvm::types</tt></a>
2599namespace and pointers, functions, arrays, etc. built of
2600those.  <tt>TypeBuilder&lt;T, false&gt;</tt> additionally allows native C types
2601whose size may depend on the host compiler.  For example,</p>
2602
2603<div class="doc_code">
2604<pre>
2605FunctionType *ft = TypeBuilder&lt;types::i&lt;8&gt;(types::i&lt;32&gt;*), true&gt;::get();
2606</pre>
2607</div>
2608
2609<p>is easier to read and write than the equivalent</p>
2610
2611<div class="doc_code">
2612<pre>
2613std::vector&lt;const Type*&gt; params;
2614params.push_back(PointerType::getUnqual(Type::Int32Ty));
2615FunctionType *ft = FunctionType::get(Type::Int8Ty, params, false);
2616</pre>
2617</div>
2618
2619<p>See the <a href="/doxygen/TypeBuilder_8h-source.html#l00001">class
2620comment</a> for more details.</p>
2621
2622</div>
2623
2624</div>
2625
2626<!-- *********************************************************************** -->
2627<h2>
2628  <a name="threading">Threads and LLVM</a>
2629</h2>
2630<!-- *********************************************************************** -->
2631
2632<div>
2633<p>
2634This section describes the interaction of the LLVM APIs with multithreading,
2635both on the part of client applications, and in the JIT, in the hosted
2636application.
2637</p>
2638
2639<p>
2640Note that LLVM's support for multithreading is still relatively young.  Up
2641through version 2.5, the execution of threaded hosted applications was
2642supported, but not threaded client access to the APIs.  While this use case is
2643now supported, clients <em>must</em> adhere to the guidelines specified below to
2644ensure proper operation in multithreaded mode.
2645</p>
2646
2647<p>
2648Note that, on Unix-like platforms, LLVM requires the presence of GCC's atomic
2649intrinsics in order to support threaded operation.  If you need a
2650multhreading-capable LLVM on a platform without a suitably modern system
2651compiler, consider compiling LLVM and LLVM-GCC in single-threaded mode, and
2652using the resultant compiler to build a copy of LLVM with multithreading
2653support.
2654</p>
2655
2656<!-- ======================================================================= -->
2657<h3>
2658  <a name="startmultithreaded">Entering and Exiting Multithreaded Mode</a>
2659</h3>
2660
2661<div>
2662
2663<p>
2664In order to properly protect its internal data structures while avoiding
2665excessive locking overhead in the single-threaded case, the LLVM must intialize
2666certain data structures necessary to provide guards around its internals.  To do
2667so, the client program must invoke <tt>llvm_start_multithreaded()</tt> before
2668making any concurrent LLVM API calls.  To subsequently tear down these
2669structures, use the <tt>llvm_stop_multithreaded()</tt> call.  You can also use
2670the <tt>llvm_is_multithreaded()</tt> call to check the status of multithreaded
2671mode.
2672</p>
2673
2674<p>
2675Note that both of these calls must be made <em>in isolation</em>.  That is to
2676say that no other LLVM API calls may be executing at any time during the
2677execution of <tt>llvm_start_multithreaded()</tt> or <tt>llvm_stop_multithreaded
2678</tt>.  It's is the client's responsibility to enforce this isolation.
2679</p>
2680
2681<p>
2682The return value of <tt>llvm_start_multithreaded()</tt> indicates the success or
2683failure of the initialization.  Failure typically indicates that your copy of
2684LLVM was built without multithreading support, typically because GCC atomic
2685intrinsics were not found in your system compiler.  In this case, the LLVM API
2686will not be safe for concurrent calls.  However, it <em>will</em> be safe for
2687hosting threaded applications in the JIT, though <a href="#jitthreading">care
2688must be taken</a> to ensure that side exits and the like do not accidentally
2689result in concurrent LLVM API calls.
2690</p>
2691</div>
2692
2693<!-- ======================================================================= -->
2694<h3>
2695  <a name="shutdown">Ending Execution with <tt>llvm_shutdown()</tt></a>
2696</h3>
2697
2698<div>
2699<p>
2700When you are done using the LLVM APIs, you should call <tt>llvm_shutdown()</tt>
2701to deallocate memory used for internal structures.  This will also invoke
2702<tt>llvm_stop_multithreaded()</tt> if LLVM is operating in multithreaded mode.
2703As such, <tt>llvm_shutdown()</tt> requires the same isolation guarantees as
2704<tt>llvm_stop_multithreaded()</tt>.
2705</p>
2706
2707<p>
2708Note that, if you use scope-based shutdown, you can use the
2709<tt>llvm_shutdown_obj</tt> class, which calls <tt>llvm_shutdown()</tt> in its
2710destructor.
2711</div>
2712
2713<!-- ======================================================================= -->
2714<h3>
2715  <a name="managedstatic">Lazy Initialization with <tt>ManagedStatic</tt></a>
2716</h3>
2717
2718<div>
2719<p>
2720<tt>ManagedStatic</tt> is a utility class in LLVM used to implement static
2721initialization of static resources, such as the global type tables.  Before the
2722invocation of <tt>llvm_shutdown()</tt>, it implements a simple lazy
2723initialization scheme.  Once <tt>llvm_start_multithreaded()</tt> returns,
2724however, it uses double-checked locking to implement thread-safe lazy
2725initialization.
2726</p>
2727
2728<p>
2729Note that, because no other threads are allowed to issue LLVM API calls before
2730<tt>llvm_start_multithreaded()</tt> returns, it is possible to have
2731<tt>ManagedStatic</tt>s of <tt>llvm::sys::Mutex</tt>s.
2732</p>
2733
2734<p>
2735The <tt>llvm_acquire_global_lock()</tt> and <tt>llvm_release_global_lock</tt>
2736APIs provide access to the global lock used to implement the double-checked
2737locking for lazy initialization.  These should only be used internally to LLVM,
2738and only if you know what you're doing!
2739</p>
2740</div>
2741
2742<!-- ======================================================================= -->
2743<h3>
2744  <a name="llvmcontext">Achieving Isolation with <tt>LLVMContext</tt></a>
2745</h3>
2746
2747<div>
2748<p>
2749<tt>LLVMContext</tt> is an opaque class in the LLVM API which clients can use
2750to operate multiple, isolated instances of LLVM concurrently within the same
2751address space.  For instance, in a hypothetical compile-server, the compilation
2752of an individual translation unit is conceptually independent from all the
2753others, and it would be desirable to be able to compile incoming translation
2754units concurrently on independent server threads.  Fortunately,
2755<tt>LLVMContext</tt> exists to enable just this kind of scenario!
2756</p>
2757
2758<p>
2759Conceptually, <tt>LLVMContext</tt> provides isolation.  Every LLVM entity
2760(<tt>Module</tt>s, <tt>Value</tt>s, <tt>Type</tt>s, <tt>Constant</tt>s, etc.)
2761in LLVM's in-memory IR belongs to an <tt>LLVMContext</tt>.  Entities in
2762different contexts <em>cannot</em> interact with each other: <tt>Module</tt>s in
2763different contexts cannot be linked together, <tt>Function</tt>s cannot be added
2764to <tt>Module</tt>s in different contexts, etc.  What this means is that is is
2765safe to compile on multiple threads simultaneously, as long as no two threads
2766operate on entities within the same context.
2767</p>
2768
2769<p>
2770In practice, very few places in the API require the explicit specification of a
2771<tt>LLVMContext</tt>, other than the <tt>Type</tt> creation/lookup APIs.
2772Because every <tt>Type</tt> carries a reference to its owning context, most
2773other entities can determine what context they belong to by looking at their
2774own <tt>Type</tt>.  If you are adding new entities to LLVM IR, please try to
2775maintain this interface design.
2776</p>
2777
2778<p>
2779For clients that do <em>not</em> require the benefits of isolation, LLVM
2780provides a convenience API <tt>getGlobalContext()</tt>.  This returns a global,
2781lazily initialized <tt>LLVMContext</tt> that may be used in situations where
2782isolation is not a concern.
2783</p>
2784</div>
2785
2786<!-- ======================================================================= -->
2787<h3>
2788  <a name="jitthreading">Threads and the JIT</a>
2789</h3>
2790
2791<div>
2792<p>
2793LLVM's "eager" JIT compiler is safe to use in threaded programs.  Multiple
2794threads can call <tt>ExecutionEngine::getPointerToFunction()</tt> or
2795<tt>ExecutionEngine::runFunction()</tt> concurrently, and multiple threads can
2796run code output by the JIT concurrently.  The user must still ensure that only
2797one thread accesses IR in a given <tt>LLVMContext</tt> while another thread
2798might be modifying it.  One way to do that is to always hold the JIT lock while
2799accessing IR outside the JIT (the JIT <em>modifies</em> the IR by adding
2800<tt>CallbackVH</tt>s).  Another way is to only
2801call <tt>getPointerToFunction()</tt> from the <tt>LLVMContext</tt>'s thread.
2802</p>
2803
2804<p>When the JIT is configured to compile lazily (using
2805<tt>ExecutionEngine::DisableLazyCompilation(false)</tt>), there is currently a
2806<a href="http://llvm.org/bugs/show_bug.cgi?id=5184">race condition</a> in
2807updating call sites after a function is lazily-jitted.  It's still possible to
2808use the lazy JIT in a threaded program if you ensure that only one thread at a
2809time can call any particular lazy stub and that the JIT lock guards any IR
2810access, but we suggest using only the eager JIT in threaded programs.
2811</p>
2812</div>
2813
2814</div>
2815
2816<!-- *********************************************************************** -->
2817<h2>
2818  <a name="advanced">Advanced Topics</a>
2819</h2>
2820<!-- *********************************************************************** -->
2821
2822<div>
2823<p>
2824This section describes some of the advanced or obscure API's that most clients
2825do not need to be aware of.  These API's tend manage the inner workings of the
2826LLVM system, and only need to be accessed in unusual circumstances.
2827</p>
2828
2829
2830<!-- ======================================================================= -->
2831<h3>
2832  <a name="SymbolTable">The <tt>ValueSymbolTable</tt> class</a>
2833</h3>
2834
2835<div>
2836<p>The <tt><a href="http://llvm.org/doxygen/classllvm_1_1ValueSymbolTable.html">
2837ValueSymbolTable</a></tt> class provides a symbol table that the <a
2838href="#Function"><tt>Function</tt></a> and <a href="#Module">
2839<tt>Module</tt></a> classes use for naming value definitions. The symbol table
2840can provide a name for any <a href="#Value"><tt>Value</tt></a>.
2841</p>
2842
2843<p>Note that the <tt>SymbolTable</tt> class should not be directly accessed
2844by most clients.  It should only be used when iteration over the symbol table
2845names themselves are required, which is very special purpose.  Note that not
2846all LLVM
2847<tt><a href="#Value">Value</a></tt>s have names, and those without names (i.e. they have
2848an empty name) do not exist in the symbol table.
2849</p>
2850
2851<p>Symbol tables support iteration over the values in the symbol
2852table with <tt>begin/end/iterator</tt> and supports querying to see if a
2853specific name is in the symbol table (with <tt>lookup</tt>).  The
2854<tt>ValueSymbolTable</tt> class exposes no public mutator methods, instead,
2855simply call <tt>setName</tt> on a value, which will autoinsert it into the
2856appropriate symbol table.</p>
2857
2858</div>
2859
2860
2861
2862<!-- ======================================================================= -->
2863<h3>
2864  <a name="UserLayout">The <tt>User</tt> and owned <tt>Use</tt> classes' memory layout</a>
2865</h3>
2866
2867<div>
2868<p>The <tt><a href="http://llvm.org/doxygen/classllvm_1_1User.html">
2869User</a></tt> class provides a basis for expressing the ownership of <tt>User</tt>
2870towards other <tt><a href="http://llvm.org/doxygen/classllvm_1_1Value.html">
2871Value</a></tt>s. The <tt><a href="http://llvm.org/doxygen/classllvm_1_1Use.html">
2872Use</a></tt> helper class is employed to do the bookkeeping and to facilitate <i>O(1)</i>
2873addition and removal.</p>
2874
2875<!-- ______________________________________________________________________ -->
2876<h4>
2877  <a name="Use2User">
2878    Interaction and relationship between <tt>User</tt> and <tt>Use</tt> objects
2879  </a>
2880</h4>
2881
2882<div>
2883<p>
2884A subclass of <tt>User</tt> can choose between incorporating its <tt>Use</tt> objects
2885or refer to them out-of-line by means of a pointer. A mixed variant
2886(some <tt>Use</tt>s inline others hung off) is impractical and breaks the invariant
2887that the <tt>Use</tt> objects belonging to the same <tt>User</tt> form a contiguous array.
2888</p>
2889
2890<p>
2891We have 2 different layouts in the <tt>User</tt> (sub)classes:
2892<ul>
2893<li><p>Layout a)
2894The <tt>Use</tt> object(s) are inside (resp. at fixed offset) of the <tt>User</tt>
2895object and there are a fixed number of them.</p>
2896
2897<li><p>Layout b)
2898The <tt>Use</tt> object(s) are referenced by a pointer to an
2899array from the <tt>User</tt> object and there may be a variable
2900number of them.</p>
2901</ul>
2902<p>
2903As of v2.4 each layout still possesses a direct pointer to the
2904start of the array of <tt>Use</tt>s. Though not mandatory for layout a),
2905we stick to this redundancy for the sake of simplicity.
2906The <tt>User</tt> object also stores the number of <tt>Use</tt> objects it
2907has. (Theoretically this information can also be calculated
2908given the scheme presented below.)</p>
2909<p>
2910Special forms of allocation operators (<tt>operator new</tt>)
2911enforce the following memory layouts:</p>
2912
2913<ul>
2914<li><p>Layout a) is modelled by prepending the <tt>User</tt> object by the <tt>Use[]</tt> array.</p>
2915
2916<pre>
2917...---.---.---.---.-------...
2918  | P | P | P | P | User
2919'''---'---'---'---'-------'''
2920</pre>
2921
2922<li><p>Layout b) is modelled by pointing at the <tt>Use[]</tt> array.</p>
2923<pre>
2924.-------...
2925| User
2926'-------'''
2927    |
2928    v
2929    .---.---.---.---...
2930    | P | P | P | P |
2931    '---'---'---'---'''
2932</pre>
2933</ul>
2934<i>(In the above figures '<tt>P</tt>' stands for the <tt>Use**</tt> that
2935    is stored in each <tt>Use</tt> object in the member <tt>Use::Prev</tt>)</i>
2936
2937</div>
2938
2939<!-- ______________________________________________________________________ -->
2940<h4>
2941  <a name="Waymarking">The waymarking algorithm</a>
2942</h4>
2943
2944<div>
2945<p>
2946Since the <tt>Use</tt> objects are deprived of the direct (back)pointer to
2947their <tt>User</tt> objects, there must be a fast and exact method to
2948recover it. This is accomplished by the following scheme:</p>
2949
2950A bit-encoding in the 2 LSBits (least significant bits) of the <tt>Use::Prev</tt> allows to find the
2951start of the <tt>User</tt> object:
2952<ul>
2953<li><tt>00</tt> &mdash;&gt; binary digit 0</li>
2954<li><tt>01</tt> &mdash;&gt; binary digit 1</li>
2955<li><tt>10</tt> &mdash;&gt; stop and calculate (<tt>s</tt>)</li>
2956<li><tt>11</tt> &mdash;&gt; full stop (<tt>S</tt>)</li>
2957</ul>
2958<p>
2959Given a <tt>Use*</tt>, all we have to do is to walk till we get
2960a stop and we either have a <tt>User</tt> immediately behind or
2961we have to walk to the next stop picking up digits
2962and calculating the offset:</p>
2963<pre>
2964.---.---.---.---.---.---.---.---.---.---.---.---.---.---.---.---.----------------
2965| 1 | s | 1 | 0 | 1 | 0 | s | 1 | 1 | 0 | s | 1 | 1 | s | 1 | S | User (or User*)
2966'---'---'---'---'---'---'---'---'---'---'---'---'---'---'---'---'----------------
2967    |+15                |+10            |+6         |+3     |+1
2968    |                   |               |           |       |__>
2969    |                   |               |           |__________>
2970    |                   |               |______________________>
2971    |                   |______________________________________>
2972    |__________________________________________________________>
2973</pre>
2974<p>
2975Only the significant number of bits need to be stored between the
2976stops, so that the <i>worst case is 20 memory accesses</i> when there are
29771000 <tt>Use</tt> objects associated with a <tt>User</tt>.</p>
2978
2979</div>
2980
2981<!-- ______________________________________________________________________ -->
2982<h4>
2983  <a name="ReferenceImpl">Reference implementation</a>
2984</h4>
2985
2986<div>
2987<p>
2988The following literate Haskell fragment demonstrates the concept:</p>
2989
2990<div class="doc_code">
2991<pre>
2992> import Test.QuickCheck
2993>
2994> digits :: Int -> [Char] -> [Char]
2995> digits 0 acc = '0' : acc
2996> digits 1 acc = '1' : acc
2997> digits n acc = digits (n `div` 2) $ digits (n `mod` 2) acc
2998>
2999> dist :: Int -> [Char] -> [Char]
3000> dist 0 [] = ['S']
3001> dist 0 acc = acc
3002> dist 1 acc = let r = dist 0 acc in 's' : digits (length r) r
3003> dist n acc = dist (n - 1) $ dist 1 acc
3004>
3005> takeLast n ss = reverse $ take n $ reverse ss
3006>
3007> test = takeLast 40 $ dist 20 []
3008>
3009</pre>
3010</div>
3011<p>
3012Printing &lt;test&gt; gives: <tt>"1s100000s11010s10100s1111s1010s110s11s1S"</tt></p>
3013<p>
3014The reverse algorithm computes the length of the string just by examining
3015a certain prefix:</p>
3016
3017<div class="doc_code">
3018<pre>
3019> pref :: [Char] -> Int
3020> pref "S" = 1
3021> pref ('s':'1':rest) = decode 2 1 rest
3022> pref (_:rest) = 1 + pref rest
3023>
3024> decode walk acc ('0':rest) = decode (walk + 1) (acc * 2) rest
3025> decode walk acc ('1':rest) = decode (walk + 1) (acc * 2 + 1) rest
3026> decode walk acc _ = walk + acc
3027>
3028</pre>
3029</div>
3030<p>
3031Now, as expected, printing &lt;pref test&gt; gives <tt>40</tt>.</p>
3032<p>
3033We can <i>quickCheck</i> this with following property:</p>
3034
3035<div class="doc_code">
3036<pre>
3037> testcase = dist 2000 []
3038> testcaseLength = length testcase
3039>
3040> identityProp n = n > 0 && n <= testcaseLength ==> length arr == pref arr
3041>     where arr = takeLast n testcase
3042>
3043</pre>
3044</div>
3045<p>
3046As expected &lt;quickCheck identityProp&gt; gives:</p>
3047
3048<pre>
3049*Main> quickCheck identityProp
3050OK, passed 100 tests.
3051</pre>
3052<p>
3053Let's be a bit more exhaustive:</p>
3054
3055<div class="doc_code">
3056<pre>
3057>
3058> deepCheck p = check (defaultConfig { configMaxTest = 500 }) p
3059>
3060</pre>
3061</div>
3062<p>
3063And here is the result of &lt;deepCheck identityProp&gt;:</p>
3064
3065<pre>
3066*Main> deepCheck identityProp
3067OK, passed 500 tests.
3068</pre>
3069
3070</div>
3071
3072<!-- ______________________________________________________________________ -->
3073<h4>
3074  <a name="Tagging">Tagging considerations</a>
3075</h4>
3076
3077<div>
3078
3079<p>
3080To maintain the invariant that the 2 LSBits of each <tt>Use**</tt> in <tt>Use</tt>
3081never change after being set up, setters of <tt>Use::Prev</tt> must re-tag the
3082new <tt>Use**</tt> on every modification. Accordingly getters must strip the
3083tag bits.</p>
3084<p>
3085For layout b) instead of the <tt>User</tt> we find a pointer (<tt>User*</tt> with LSBit set).
3086Following this pointer brings us to the <tt>User</tt>. A portable trick ensures
3087that the first bytes of <tt>User</tt> (if interpreted as a pointer) never has
3088the LSBit set. (Portability is relying on the fact that all known compilers place the
3089<tt>vptr</tt> in the first word of the instances.)</p>
3090
3091</div>
3092
3093</div>
3094
3095</div>
3096
3097<!-- *********************************************************************** -->
3098<h2>
3099  <a name="coreclasses">The Core LLVM Class Hierarchy Reference </a>
3100</h2>
3101<!-- *********************************************************************** -->
3102
3103<div>
3104<p><tt>#include "<a href="/doxygen/Type_8h-source.html">llvm/Type.h</a>"</tt>
3105<br>doxygen info: <a href="/doxygen/classllvm_1_1Type.html">Type Class</a></p>
3106
3107<p>The Core LLVM classes are the primary means of representing the program
3108being inspected or transformed.  The core LLVM classes are defined in
3109header files in the <tt>include/llvm/</tt> directory, and implemented in
3110the <tt>lib/VMCore</tt> directory.</p>
3111
3112<!-- ======================================================================= -->
3113<h3>
3114  <a name="Type">The <tt>Type</tt> class and Derived Types</a>
3115</h3>
3116
3117<div>
3118
3119  <p><tt>Type</tt> is a superclass of all type classes. Every <tt>Value</tt> has
3120  a <tt>Type</tt>. <tt>Type</tt> cannot be instantiated directly but only
3121  through its subclasses. Certain primitive types (<tt>VoidType</tt>,
3122  <tt>LabelType</tt>, <tt>FloatType</tt> and <tt>DoubleType</tt>) have hidden
3123  subclasses. They are hidden because they offer no useful functionality beyond
3124  what the <tt>Type</tt> class offers except to distinguish themselves from
3125  other subclasses of <tt>Type</tt>.</p>
3126  <p>All other types are subclasses of <tt>DerivedType</tt>.  Types can be
3127  named, but this is not a requirement. There exists exactly
3128  one instance of a given shape at any one time.  This allows type equality to
3129  be performed with address equality of the Type Instance. That is, given two
3130  <tt>Type*</tt> values, the types are identical if the pointers are identical.
3131  </p>
3132
3133<!-- _______________________________________________________________________ -->
3134<h4>
3135  <a name="m_Type">Important Public Methods</a>
3136</h4>
3137
3138<div>
3139
3140<ul>
3141  <li><tt>bool isIntegerTy() const</tt>: Returns true for any integer type.</li>
3142
3143  <li><tt>bool isFloatingPointTy()</tt>: Return true if this is one of the five
3144  floating point types.</li>
3145
3146  <li><tt>bool isSized()</tt>: Return true if the type has known size. Things
3147  that don't have a size are abstract types, labels and void.</li>
3148
3149</ul>
3150</div>
3151
3152<!-- _______________________________________________________________________ -->
3153<h4>
3154  <a name="derivedtypes">Important Derived Types</a>
3155</h4>
3156<div>
3157<dl>
3158  <dt><tt>IntegerType</tt></dt>
3159  <dd>Subclass of DerivedType that represents integer types of any bit width.
3160  Any bit width between <tt>IntegerType::MIN_INT_BITS</tt> (1) and
3161  <tt>IntegerType::MAX_INT_BITS</tt> (~8 million) can be represented.
3162  <ul>
3163    <li><tt>static const IntegerType* get(unsigned NumBits)</tt>: get an integer
3164    type of a specific bit width.</li>
3165    <li><tt>unsigned getBitWidth() const</tt>: Get the bit width of an integer
3166    type.</li>
3167  </ul>
3168  </dd>
3169  <dt><tt>SequentialType</tt></dt>
3170  <dd>This is subclassed by ArrayType, PointerType and VectorType.
3171    <ul>
3172      <li><tt>const Type * getElementType() const</tt>: Returns the type of each
3173      of the elements in the sequential type. </li>
3174    </ul>
3175  </dd>
3176  <dt><tt>ArrayType</tt></dt>
3177  <dd>This is a subclass of SequentialType and defines the interface for array
3178  types.
3179    <ul>
3180      <li><tt>unsigned getNumElements() const</tt>: Returns the number of
3181      elements in the array. </li>
3182    </ul>
3183  </dd>
3184  <dt><tt>PointerType</tt></dt>
3185  <dd>Subclass of SequentialType for pointer types.</dd>
3186  <dt><tt>VectorType</tt></dt>
3187  <dd>Subclass of SequentialType for vector types. A
3188  vector type is similar to an ArrayType but is distinguished because it is
3189  a first class type whereas ArrayType is not. Vector types are used for
3190  vector operations and are usually small vectors of of an integer or floating
3191  point type.</dd>
3192  <dt><tt>StructType</tt></dt>
3193  <dd>Subclass of DerivedTypes for struct types.</dd>
3194  <dt><tt><a name="FunctionType">FunctionType</a></tt></dt>
3195  <dd>Subclass of DerivedTypes for function types.
3196    <ul>
3197      <li><tt>bool isVarArg() const</tt>: Returns true if it's a vararg
3198      function</li>
3199      <li><tt> const Type * getReturnType() const</tt>: Returns the
3200      return type of the function.</li>
3201      <li><tt>const Type * getParamType (unsigned i)</tt>: Returns
3202      the type of the ith parameter.</li>
3203      <li><tt> const unsigned getNumParams() const</tt>: Returns the
3204      number of formal parameters.</li>
3205    </ul>
3206  </dd>
3207</dl>
3208</div>
3209
3210</div>
3211
3212<!-- ======================================================================= -->
3213<h3>
3214  <a name="Module">The <tt>Module</tt> class</a>
3215</h3>
3216
3217<div>
3218
3219<p><tt>#include "<a
3220href="/doxygen/Module_8h-source.html">llvm/Module.h</a>"</tt><br> doxygen info:
3221<a href="/doxygen/classllvm_1_1Module.html">Module Class</a></p>
3222
3223<p>The <tt>Module</tt> class represents the top level structure present in LLVM
3224programs.  An LLVM module is effectively either a translation unit of the
3225original program or a combination of several translation units merged by the
3226linker.  The <tt>Module</tt> class keeps track of a list of <a
3227href="#Function"><tt>Function</tt></a>s, a list of <a
3228href="#GlobalVariable"><tt>GlobalVariable</tt></a>s, and a <a
3229href="#SymbolTable"><tt>SymbolTable</tt></a>.  Additionally, it contains a few
3230helpful member functions that try to make common operations easy.</p>
3231
3232<!-- _______________________________________________________________________ -->
3233<h4>
3234  <a name="m_Module">Important Public Members of the <tt>Module</tt> class</a>
3235</h4>
3236
3237<div>
3238
3239<ul>
3240  <li><tt>Module::Module(std::string name = "")</tt></li>
3241</ul>
3242
3243<p>Constructing a <a href="#Module">Module</a> is easy. You can optionally
3244provide a name for it (probably based on the name of the translation unit).</p>
3245
3246<ul>
3247  <li><tt>Module::iterator</tt> - Typedef for function list iterator<br>
3248    <tt>Module::const_iterator</tt> - Typedef for const_iterator.<br>
3249
3250    <tt>begin()</tt>, <tt>end()</tt>
3251    <tt>size()</tt>, <tt>empty()</tt>
3252
3253    <p>These are forwarding methods that make it easy to access the contents of
3254    a <tt>Module</tt> object's <a href="#Function"><tt>Function</tt></a>
3255    list.</p></li>
3256
3257  <li><tt>Module::FunctionListType &amp;getFunctionList()</tt>
3258
3259    <p> Returns the list of <a href="#Function"><tt>Function</tt></a>s.  This is
3260    necessary to use when you need to update the list or perform a complex
3261    action that doesn't have a forwarding method.</p>
3262
3263    <p><!--  Global Variable --></p></li>
3264</ul>
3265
3266<hr>
3267
3268<ul>
3269  <li><tt>Module::global_iterator</tt> - Typedef for global variable list iterator<br>
3270
3271    <tt>Module::const_global_iterator</tt> - Typedef for const_iterator.<br>
3272
3273    <tt>global_begin()</tt>, <tt>global_end()</tt>
3274    <tt>global_size()</tt>, <tt>global_empty()</tt>
3275
3276    <p> These are forwarding methods that make it easy to access the contents of
3277    a <tt>Module</tt> object's <a
3278    href="#GlobalVariable"><tt>GlobalVariable</tt></a> list.</p></li>
3279
3280  <li><tt>Module::GlobalListType &amp;getGlobalList()</tt>
3281
3282    <p>Returns the list of <a
3283    href="#GlobalVariable"><tt>GlobalVariable</tt></a>s.  This is necessary to
3284    use when you need to update the list or perform a complex action that
3285    doesn't have a forwarding method.</p>
3286
3287    <p><!--  Symbol table stuff --> </p></li>
3288</ul>
3289
3290<hr>
3291
3292<ul>
3293  <li><tt><a href="#SymbolTable">SymbolTable</a> *getSymbolTable()</tt>
3294
3295    <p>Return a reference to the <a href="#SymbolTable"><tt>SymbolTable</tt></a>
3296    for this <tt>Module</tt>.</p>
3297
3298    <p><!--  Convenience methods --></p></li>
3299</ul>
3300
3301<hr>
3302
3303<ul>
3304  <li><tt><a href="#Function">Function</a> *getFunction(const std::string
3305  &amp;Name, const <a href="#FunctionType">FunctionType</a> *Ty)</tt>
3306
3307    <p>Look up the specified function in the <tt>Module</tt> <a
3308    href="#SymbolTable"><tt>SymbolTable</tt></a>. If it does not exist, return
3309    <tt>null</tt>.</p></li>
3310
3311  <li><tt><a href="#Function">Function</a> *getOrInsertFunction(const
3312  std::string &amp;Name, const <a href="#FunctionType">FunctionType</a> *T)</tt>
3313
3314    <p>Look up the specified function in the <tt>Module</tt> <a
3315    href="#SymbolTable"><tt>SymbolTable</tt></a>. If it does not exist, add an
3316    external declaration for the function and return it.</p></li>
3317
3318  <li><tt>std::string getTypeName(const <a href="#Type">Type</a> *Ty)</tt>
3319
3320    <p>If there is at least one entry in the <a
3321    href="#SymbolTable"><tt>SymbolTable</tt></a> for the specified <a
3322    href="#Type"><tt>Type</tt></a>, return it.  Otherwise return the empty
3323    string.</p></li>
3324
3325  <li><tt>bool addTypeName(const std::string &amp;Name, const <a
3326  href="#Type">Type</a> *Ty)</tt>
3327
3328    <p>Insert an entry in the <a href="#SymbolTable"><tt>SymbolTable</tt></a>
3329    mapping <tt>Name</tt> to <tt>Ty</tt>. If there is already an entry for this
3330    name, true is returned and the <a
3331    href="#SymbolTable"><tt>SymbolTable</tt></a> is not modified.</p></li>
3332</ul>
3333
3334</div>
3335
3336</div>
3337
3338<!-- ======================================================================= -->
3339<h3>
3340  <a name="Value">The <tt>Value</tt> class</a>
3341</h3>
3342
3343<div>
3344
3345<p><tt>#include "<a href="/doxygen/Value_8h-source.html">llvm/Value.h</a>"</tt>
3346<br>
3347doxygen info: <a href="/doxygen/classllvm_1_1Value.html">Value Class</a></p>
3348
3349<p>The <tt>Value</tt> class is the most important class in the LLVM Source
3350base.  It represents a typed value that may be used (among other things) as an
3351operand to an instruction.  There are many different types of <tt>Value</tt>s,
3352such as <a href="#Constant"><tt>Constant</tt></a>s,<a
3353href="#Argument"><tt>Argument</tt></a>s. Even <a
3354href="#Instruction"><tt>Instruction</tt></a>s and <a
3355href="#Function"><tt>Function</tt></a>s are <tt>Value</tt>s.</p>
3356
3357<p>A particular <tt>Value</tt> may be used many times in the LLVM representation
3358for a program.  For example, an incoming argument to a function (represented
3359with an instance of the <a href="#Argument">Argument</a> class) is "used" by
3360every instruction in the function that references the argument.  To keep track
3361of this relationship, the <tt>Value</tt> class keeps a list of all of the <a
3362href="#User"><tt>User</tt></a>s that is using it (the <a
3363href="#User"><tt>User</tt></a> class is a base class for all nodes in the LLVM
3364graph that can refer to <tt>Value</tt>s).  This use list is how LLVM represents
3365def-use information in the program, and is accessible through the <tt>use_</tt>*
3366methods, shown below.</p>
3367
3368<p>Because LLVM is a typed representation, every LLVM <tt>Value</tt> is typed,
3369and this <a href="#Type">Type</a> is available through the <tt>getType()</tt>
3370method. In addition, all LLVM values can be named.  The "name" of the
3371<tt>Value</tt> is a symbolic string printed in the LLVM code:</p>
3372
3373<div class="doc_code">
3374<pre>
3375%<b>foo</b> = add i32 1, 2
3376</pre>
3377</div>
3378
3379<p><a name="nameWarning">The name of this instruction is "foo".</a> <b>NOTE</b>
3380that the name of any value may be missing (an empty string), so names should
3381<b>ONLY</b> be used for debugging (making the source code easier to read,
3382debugging printouts), they should not be used to keep track of values or map
3383between them.  For this purpose, use a <tt>std::map</tt> of pointers to the
3384<tt>Value</tt> itself instead.</p>
3385
3386<p>One important aspect of LLVM is that there is no distinction between an SSA
3387variable and the operation that produces it.  Because of this, any reference to
3388the value produced by an instruction (or the value available as an incoming
3389argument, for example) is represented as a direct pointer to the instance of
3390the class that
3391represents this value.  Although this may take some getting used to, it
3392simplifies the representation and makes it easier to manipulate.</p>
3393
3394<!-- _______________________________________________________________________ -->
3395<h4>
3396  <a name="m_Value">Important Public Members of the <tt>Value</tt> class</a>
3397</h4>
3398
3399<div>
3400
3401<ul>
3402  <li><tt>Value::use_iterator</tt> - Typedef for iterator over the
3403use-list<br>
3404    <tt>Value::const_use_iterator</tt> - Typedef for const_iterator over
3405the use-list<br>
3406    <tt>unsigned use_size()</tt> - Returns the number of users of the
3407value.<br>
3408    <tt>bool use_empty()</tt> - Returns true if there are no users.<br>
3409    <tt>use_iterator use_begin()</tt> - Get an iterator to the start of
3410the use-list.<br>
3411    <tt>use_iterator use_end()</tt> - Get an iterator to the end of the
3412use-list.<br>
3413    <tt><a href="#User">User</a> *use_back()</tt> - Returns the last
3414element in the list.
3415    <p> These methods are the interface to access the def-use
3416information in LLVM.  As with all other iterators in LLVM, the naming
3417conventions follow the conventions defined by the <a href="#stl">STL</a>.</p>
3418  </li>
3419  <li><tt><a href="#Type">Type</a> *getType() const</tt>
3420    <p>This method returns the Type of the Value.</p>
3421  </li>
3422  <li><tt>bool hasName() const</tt><br>
3423    <tt>std::string getName() const</tt><br>
3424    <tt>void setName(const std::string &amp;Name)</tt>
3425    <p> This family of methods is used to access and assign a name to a <tt>Value</tt>,
3426be aware of the <a href="#nameWarning">precaution above</a>.</p>
3427  </li>
3428  <li><tt>void replaceAllUsesWith(Value *V)</tt>
3429
3430    <p>This method traverses the use list of a <tt>Value</tt> changing all <a
3431    href="#User"><tt>User</tt>s</a> of the current value to refer to
3432    "<tt>V</tt>" instead.  For example, if you detect that an instruction always
3433    produces a constant value (for example through constant folding), you can
3434    replace all uses of the instruction with the constant like this:</p>
3435
3436<div class="doc_code">
3437<pre>
3438Inst-&gt;replaceAllUsesWith(ConstVal);
3439</pre>
3440</div>
3441
3442</ul>
3443
3444</div>
3445
3446</div>
3447
3448<!-- ======================================================================= -->
3449<h3>
3450  <a name="User">The <tt>User</tt> class</a>
3451</h3>
3452
3453<div>
3454
3455<p>
3456<tt>#include "<a href="/doxygen/User_8h-source.html">llvm/User.h</a>"</tt><br>
3457doxygen info: <a href="/doxygen/classllvm_1_1User.html">User Class</a><br>
3458Superclass: <a href="#Value"><tt>Value</tt></a></p>
3459
3460<p>The <tt>User</tt> class is the common base class of all LLVM nodes that may
3461refer to <a href="#Value"><tt>Value</tt></a>s.  It exposes a list of "Operands"
3462that are all of the <a href="#Value"><tt>Value</tt></a>s that the User is
3463referring to.  The <tt>User</tt> class itself is a subclass of
3464<tt>Value</tt>.</p>
3465
3466<p>The operands of a <tt>User</tt> point directly to the LLVM <a
3467href="#Value"><tt>Value</tt></a> that it refers to.  Because LLVM uses Static
3468Single Assignment (SSA) form, there can only be one definition referred to,
3469allowing this direct connection.  This connection provides the use-def
3470information in LLVM.</p>
3471
3472<!-- _______________________________________________________________________ -->
3473<h4>
3474  <a name="m_User">Important Public Members of the <tt>User</tt> class</a>
3475</h4>
3476
3477<div>
3478
3479<p>The <tt>User</tt> class exposes the operand list in two ways: through
3480an index access interface and through an iterator based interface.</p>
3481
3482<ul>
3483  <li><tt>Value *getOperand(unsigned i)</tt><br>
3484    <tt>unsigned getNumOperands()</tt>
3485    <p> These two methods expose the operands of the <tt>User</tt> in a
3486convenient form for direct access.</p></li>
3487
3488  <li><tt>User::op_iterator</tt> - Typedef for iterator over the operand
3489list<br>
3490    <tt>op_iterator op_begin()</tt> - Get an iterator to the start of
3491the operand list.<br>
3492    <tt>op_iterator op_end()</tt> - Get an iterator to the end of the
3493operand list.
3494    <p> Together, these methods make up the iterator based interface to
3495the operands of a <tt>User</tt>.</p></li>
3496</ul>
3497
3498</div>
3499
3500</div>
3501
3502<!-- ======================================================================= -->
3503<h3>
3504  <a name="Instruction">The <tt>Instruction</tt> class</a>
3505</h3>
3506
3507<div>
3508
3509<p><tt>#include "</tt><tt><a
3510href="/doxygen/Instruction_8h-source.html">llvm/Instruction.h</a>"</tt><br>
3511doxygen info: <a href="/doxygen/classllvm_1_1Instruction.html">Instruction Class</a><br>
3512Superclasses: <a href="#User"><tt>User</tt></a>, <a
3513href="#Value"><tt>Value</tt></a></p>
3514
3515<p>The <tt>Instruction</tt> class is the common base class for all LLVM
3516instructions.  It provides only a few methods, but is a very commonly used
3517class.  The primary data tracked by the <tt>Instruction</tt> class itself is the
3518opcode (instruction type) and the parent <a
3519href="#BasicBlock"><tt>BasicBlock</tt></a> the <tt>Instruction</tt> is embedded
3520into.  To represent a specific type of instruction, one of many subclasses of
3521<tt>Instruction</tt> are used.</p>
3522
3523<p> Because the <tt>Instruction</tt> class subclasses the <a
3524href="#User"><tt>User</tt></a> class, its operands can be accessed in the same
3525way as for other <a href="#User"><tt>User</tt></a>s (with the
3526<tt>getOperand()</tt>/<tt>getNumOperands()</tt> and
3527<tt>op_begin()</tt>/<tt>op_end()</tt> methods).</p> <p> An important file for
3528the <tt>Instruction</tt> class is the <tt>llvm/Instruction.def</tt> file. This
3529file contains some meta-data about the various different types of instructions
3530in LLVM.  It describes the enum values that are used as opcodes (for example
3531<tt>Instruction::Add</tt> and <tt>Instruction::ICmp</tt>), as well as the
3532concrete sub-classes of <tt>Instruction</tt> that implement the instruction (for
3533example <tt><a href="#BinaryOperator">BinaryOperator</a></tt> and <tt><a
3534href="#CmpInst">CmpInst</a></tt>).  Unfortunately, the use of macros in
3535this file confuses doxygen, so these enum values don't show up correctly in the
3536<a href="/doxygen/classllvm_1_1Instruction.html">doxygen output</a>.</p>
3537
3538<!-- _______________________________________________________________________ -->
3539<h4>
3540  <a name="s_Instruction">
3541    Important Subclasses of the <tt>Instruction</tt> class
3542  </a>
3543</h4>
3544<div>
3545  <ul>
3546    <li><tt><a name="BinaryOperator">BinaryOperator</a></tt>
3547    <p>This subclasses represents all two operand instructions whose operands
3548    must be the same type, except for the comparison instructions.</p></li>
3549    <li><tt><a name="CastInst">CastInst</a></tt>
3550    <p>This subclass is the parent of the 12 casting instructions. It provides
3551    common operations on cast instructions.</p>
3552    <li><tt><a name="CmpInst">CmpInst</a></tt>
3553    <p>This subclass respresents the two comparison instructions,
3554    <a href="LangRef.html#i_icmp">ICmpInst</a> (integer opreands), and
3555    <a href="LangRef.html#i_fcmp">FCmpInst</a> (floating point operands).</p>
3556    <li><tt><a name="TerminatorInst">TerminatorInst</a></tt>
3557    <p>This subclass is the parent of all terminator instructions (those which
3558    can terminate a block).</p>
3559  </ul>
3560  </div>
3561
3562<!-- _______________________________________________________________________ -->
3563<h4>
3564  <a name="m_Instruction">
3565    Important Public Members of the <tt>Instruction</tt> class
3566  </a>
3567</h4>
3568
3569<div>
3570
3571<ul>
3572  <li><tt><a href="#BasicBlock">BasicBlock</a> *getParent()</tt>
3573    <p>Returns the <a href="#BasicBlock"><tt>BasicBlock</tt></a> that
3574this  <tt>Instruction</tt> is embedded into.</p></li>
3575  <li><tt>bool mayWriteToMemory()</tt>
3576    <p>Returns true if the instruction writes to memory, i.e. it is a
3577      <tt>call</tt>,<tt>free</tt>,<tt>invoke</tt>, or <tt>store</tt>.</p></li>
3578  <li><tt>unsigned getOpcode()</tt>
3579    <p>Returns the opcode for the <tt>Instruction</tt>.</p></li>
3580  <li><tt><a href="#Instruction">Instruction</a> *clone() const</tt>
3581    <p>Returns another instance of the specified instruction, identical
3582in all ways to the original except that the instruction has no parent
3583(ie it's not embedded into a <a href="#BasicBlock"><tt>BasicBlock</tt></a>),
3584and it has no name</p></li>
3585</ul>
3586
3587</div>
3588
3589</div>
3590
3591<!-- ======================================================================= -->
3592<h3>
3593  <a name="Constant">The <tt>Constant</tt> class and subclasses</a>
3594</h3>
3595
3596<div>
3597
3598<p>Constant represents a base class for different types of constants. It
3599is subclassed by ConstantInt, ConstantArray, etc. for representing
3600the various types of Constants.  <a href="#GlobalValue">GlobalValue</a> is also
3601a subclass, which represents the address of a global variable or function.
3602</p>
3603
3604<!-- _______________________________________________________________________ -->
3605<h4>Important Subclasses of Constant</h4>
3606<div>
3607<ul>
3608  <li>ConstantInt : This subclass of Constant represents an integer constant of
3609  any width.
3610    <ul>
3611      <li><tt>const APInt&amp; getValue() const</tt>: Returns the underlying
3612      value of this constant, an APInt value.</li>
3613      <li><tt>int64_t getSExtValue() const</tt>: Converts the underlying APInt
3614      value to an int64_t via sign extension. If the value (not the bit width)
3615      of the APInt is too large to fit in an int64_t, an assertion will result.
3616      For this reason, use of this method is discouraged.</li>
3617      <li><tt>uint64_t getZExtValue() const</tt>: Converts the underlying APInt
3618      value to a uint64_t via zero extension. IF the value (not the bit width)
3619      of the APInt is too large to fit in a uint64_t, an assertion will result.
3620      For this reason, use of this method is discouraged.</li>
3621      <li><tt>static ConstantInt* get(const APInt&amp; Val)</tt>: Returns the
3622      ConstantInt object that represents the value provided by <tt>Val</tt>.
3623      The type is implied as the IntegerType that corresponds to the bit width
3624      of <tt>Val</tt>.</li>
3625      <li><tt>static ConstantInt* get(const Type *Ty, uint64_t Val)</tt>:
3626      Returns the ConstantInt object that represents the value provided by
3627      <tt>Val</tt> for integer type <tt>Ty</tt>.</li>
3628    </ul>
3629  </li>
3630  <li>ConstantFP : This class represents a floating point constant.
3631    <ul>
3632      <li><tt>double getValue() const</tt>: Returns the underlying value of
3633      this constant. </li>
3634    </ul>
3635  </li>
3636  <li>ConstantArray : This represents a constant array.
3637    <ul>
3638      <li><tt>const std::vector&lt;Use&gt; &amp;getValues() const</tt>: Returns
3639      a vector of component constants that makeup this array. </li>
3640    </ul>
3641  </li>
3642  <li>ConstantStruct : This represents a constant struct.
3643    <ul>
3644      <li><tt>const std::vector&lt;Use&gt; &amp;getValues() const</tt>: Returns
3645      a vector of component constants that makeup this array. </li>
3646    </ul>
3647  </li>
3648  <li>GlobalValue : This represents either a global variable or a function. In
3649  either case, the value is a constant fixed address (after linking).
3650  </li>
3651</ul>
3652</div>
3653
3654</div>
3655
3656<!-- ======================================================================= -->
3657<h3>
3658  <a name="GlobalValue">The <tt>GlobalValue</tt> class</a>
3659</h3>
3660
3661<div>
3662
3663<p><tt>#include "<a
3664href="/doxygen/GlobalValue_8h-source.html">llvm/GlobalValue.h</a>"</tt><br>
3665doxygen info: <a href="/doxygen/classllvm_1_1GlobalValue.html">GlobalValue
3666Class</a><br>
3667Superclasses: <a href="#Constant"><tt>Constant</tt></a>,
3668<a href="#User"><tt>User</tt></a>, <a href="#Value"><tt>Value</tt></a></p>
3669
3670<p>Global values (<a href="#GlobalVariable"><tt>GlobalVariable</tt></a>s or <a
3671href="#Function"><tt>Function</tt></a>s) are the only LLVM values that are
3672visible in the bodies of all <a href="#Function"><tt>Function</tt></a>s.
3673Because they are visible at global scope, they are also subject to linking with
3674other globals defined in different translation units.  To control the linking
3675process, <tt>GlobalValue</tt>s know their linkage rules. Specifically,
3676<tt>GlobalValue</tt>s know whether they have internal or external linkage, as
3677defined by the <tt>LinkageTypes</tt> enumeration.</p>
3678
3679<p>If a <tt>GlobalValue</tt> has internal linkage (equivalent to being
3680<tt>static</tt> in C), it is not visible to code outside the current translation
3681unit, and does not participate in linking.  If it has external linkage, it is
3682visible to external code, and does participate in linking.  In addition to
3683linkage information, <tt>GlobalValue</tt>s keep track of which <a
3684href="#Module"><tt>Module</tt></a> they are currently part of.</p>
3685
3686<p>Because <tt>GlobalValue</tt>s are memory objects, they are always referred to
3687by their <b>address</b>. As such, the <a href="#Type"><tt>Type</tt></a> of a
3688global is always a pointer to its contents. It is important to remember this
3689when using the <tt>GetElementPtrInst</tt> instruction because this pointer must
3690be dereferenced first. For example, if you have a <tt>GlobalVariable</tt> (a
3691subclass of <tt>GlobalValue)</tt> that is an array of 24 ints, type <tt>[24 x
3692i32]</tt>, then the <tt>GlobalVariable</tt> is a pointer to that array. Although
3693the address of the first element of this array and the value of the
3694<tt>GlobalVariable</tt> are the same, they have different types. The
3695<tt>GlobalVariable</tt>'s type is <tt>[24 x i32]</tt>. The first element's type
3696is <tt>i32.</tt> Because of this, accessing a global value requires you to
3697dereference the pointer with <tt>GetElementPtrInst</tt> first, then its elements
3698can be accessed. This is explained in the <a href="LangRef.html#globalvars">LLVM
3699Language Reference Manual</a>.</p>
3700
3701<!-- _______________________________________________________________________ -->
3702<h4>
3703  <a name="m_GlobalValue">
3704    Important Public Members of the <tt>GlobalValue</tt> class
3705  </a>
3706</h4>
3707
3708<div>
3709
3710<ul>
3711  <li><tt>bool hasInternalLinkage() const</tt><br>
3712    <tt>bool hasExternalLinkage() const</tt><br>
3713    <tt>void setInternalLinkage(bool HasInternalLinkage)</tt>
3714    <p> These methods manipulate the linkage characteristics of the <tt>GlobalValue</tt>.</p>
3715    <p> </p>
3716  </li>
3717  <li><tt><a href="#Module">Module</a> *getParent()</tt>
3718    <p> This returns the <a href="#Module"><tt>Module</tt></a> that the
3719GlobalValue is currently embedded into.</p></li>
3720</ul>
3721
3722</div>
3723
3724</div>
3725
3726<!-- ======================================================================= -->
3727<h3>
3728  <a name="Function">The <tt>Function</tt> class</a>
3729</h3>
3730
3731<div>
3732
3733<p><tt>#include "<a
3734href="/doxygen/Function_8h-source.html">llvm/Function.h</a>"</tt><br> doxygen
3735info: <a href="/doxygen/classllvm_1_1Function.html">Function Class</a><br>
3736Superclasses: <a href="#GlobalValue"><tt>GlobalValue</tt></a>,
3737<a href="#Constant"><tt>Constant</tt></a>,
3738<a href="#User"><tt>User</tt></a>,
3739<a href="#Value"><tt>Value</tt></a></p>
3740
3741<p>The <tt>Function</tt> class represents a single procedure in LLVM.  It is
3742actually one of the more complex classes in the LLVM hierarchy because it must
3743keep track of a large amount of data.  The <tt>Function</tt> class keeps track
3744of a list of <a href="#BasicBlock"><tt>BasicBlock</tt></a>s, a list of formal
3745<a href="#Argument"><tt>Argument</tt></a>s, and a
3746<a href="#SymbolTable"><tt>SymbolTable</tt></a>.</p>
3747
3748<p>The list of <a href="#BasicBlock"><tt>BasicBlock</tt></a>s is the most
3749commonly used part of <tt>Function</tt> objects.  The list imposes an implicit
3750ordering of the blocks in the function, which indicate how the code will be
3751laid out by the backend.  Additionally, the first <a
3752href="#BasicBlock"><tt>BasicBlock</tt></a> is the implicit entry node for the
3753<tt>Function</tt>.  It is not legal in LLVM to explicitly branch to this initial
3754block.  There are no implicit exit nodes, and in fact there may be multiple exit
3755nodes from a single <tt>Function</tt>.  If the <a
3756href="#BasicBlock"><tt>BasicBlock</tt></a> list is empty, this indicates that
3757the <tt>Function</tt> is actually a function declaration: the actual body of the
3758function hasn't been linked in yet.</p>
3759
3760<p>In addition to a list of <a href="#BasicBlock"><tt>BasicBlock</tt></a>s, the
3761<tt>Function</tt> class also keeps track of the list of formal <a
3762href="#Argument"><tt>Argument</tt></a>s that the function receives.  This
3763container manages the lifetime of the <a href="#Argument"><tt>Argument</tt></a>
3764nodes, just like the <a href="#BasicBlock"><tt>BasicBlock</tt></a> list does for
3765the <a href="#BasicBlock"><tt>BasicBlock</tt></a>s.</p>
3766
3767<p>The <a href="#SymbolTable"><tt>SymbolTable</tt></a> is a very rarely used
3768LLVM feature that is only used when you have to look up a value by name.  Aside
3769from that, the <a href="#SymbolTable"><tt>SymbolTable</tt></a> is used
3770internally to make sure that there are not conflicts between the names of <a
3771href="#Instruction"><tt>Instruction</tt></a>s, <a
3772href="#BasicBlock"><tt>BasicBlock</tt></a>s, or <a
3773href="#Argument"><tt>Argument</tt></a>s in the function body.</p>
3774
3775<p>Note that <tt>Function</tt> is a <a href="#GlobalValue">GlobalValue</a>
3776and therefore also a <a href="#Constant">Constant</a>. The value of the function
3777is its address (after linking) which is guaranteed to be constant.</p>
3778
3779<!-- _______________________________________________________________________ -->
3780<h4>
3781  <a name="m_Function">
3782    Important Public Members of the <tt>Function</tt> class
3783  </a>
3784</h4>
3785
3786<div>
3787
3788<ul>
3789  <li><tt>Function(const </tt><tt><a href="#FunctionType">FunctionType</a>
3790  *Ty, LinkageTypes Linkage, const std::string &amp;N = "", Module* Parent = 0)</tt>
3791
3792    <p>Constructor used when you need to create new <tt>Function</tt>s to add
3793    the the program.  The constructor must specify the type of the function to
3794    create and what type of linkage the function should have. The <a
3795    href="#FunctionType"><tt>FunctionType</tt></a> argument
3796    specifies the formal arguments and return value for the function. The same
3797    <a href="#FunctionType"><tt>FunctionType</tt></a> value can be used to
3798    create multiple functions. The <tt>Parent</tt> argument specifies the Module
3799    in which the function is defined. If this argument is provided, the function
3800    will automatically be inserted into that module's list of
3801    functions.</p></li>
3802
3803  <li><tt>bool isDeclaration()</tt>
3804
3805    <p>Return whether or not the <tt>Function</tt> has a body defined.  If the
3806    function is "external", it does not have a body, and thus must be resolved
3807    by linking with a function defined in a different translation unit.</p></li>
3808
3809  <li><tt>Function::iterator</tt> - Typedef for basic block list iterator<br>
3810    <tt>Function::const_iterator</tt> - Typedef for const_iterator.<br>
3811
3812    <tt>begin()</tt>, <tt>end()</tt>
3813    <tt>size()</tt>, <tt>empty()</tt>
3814
3815    <p>These are forwarding methods that make it easy to access the contents of
3816    a <tt>Function</tt> object's <a href="#BasicBlock"><tt>BasicBlock</tt></a>
3817    list.</p></li>
3818
3819  <li><tt>Function::BasicBlockListType &amp;getBasicBlockList()</tt>
3820
3821    <p>Returns the list of <a href="#BasicBlock"><tt>BasicBlock</tt></a>s.  This
3822    is necessary to use when you need to update the list or perform a complex
3823    action that doesn't have a forwarding method.</p></li>
3824
3825  <li><tt>Function::arg_iterator</tt> - Typedef for the argument list
3826iterator<br>
3827    <tt>Function::const_arg_iterator</tt> - Typedef for const_iterator.<br>
3828
3829    <tt>arg_begin()</tt>, <tt>arg_end()</tt>
3830    <tt>arg_size()</tt>, <tt>arg_empty()</tt>
3831
3832    <p>These are forwarding methods that make it easy to access the contents of
3833    a <tt>Function</tt> object's <a href="#Argument"><tt>Argument</tt></a>
3834    list.</p></li>
3835
3836  <li><tt>Function::ArgumentListType &amp;getArgumentList()</tt>
3837
3838    <p>Returns the list of <a href="#Argument"><tt>Argument</tt></a>s.  This is
3839    necessary to use when you need to update the list or perform a complex
3840    action that doesn't have a forwarding method.</p></li>
3841
3842  <li><tt><a href="#BasicBlock">BasicBlock</a> &amp;getEntryBlock()</tt>
3843
3844    <p>Returns the entry <a href="#BasicBlock"><tt>BasicBlock</tt></a> for the
3845    function.  Because the entry block for the function is always the first
3846    block, this returns the first block of the <tt>Function</tt>.</p></li>
3847
3848  <li><tt><a href="#Type">Type</a> *getReturnType()</tt><br>
3849    <tt><a href="#FunctionType">FunctionType</a> *getFunctionType()</tt>
3850
3851    <p>This traverses the <a href="#Type"><tt>Type</tt></a> of the
3852    <tt>Function</tt> and returns the return type of the function, or the <a
3853    href="#FunctionType"><tt>FunctionType</tt></a> of the actual
3854    function.</p></li>
3855
3856  <li><tt><a href="#SymbolTable">SymbolTable</a> *getSymbolTable()</tt>
3857
3858    <p> Return a pointer to the <a href="#SymbolTable"><tt>SymbolTable</tt></a>
3859    for this <tt>Function</tt>.</p></li>
3860</ul>
3861
3862</div>
3863
3864</div>
3865
3866<!-- ======================================================================= -->
3867<h3>
3868  <a name="GlobalVariable">The <tt>GlobalVariable</tt> class</a>
3869</h3>
3870
3871<div>
3872
3873<p><tt>#include "<a
3874href="/doxygen/GlobalVariable_8h-source.html">llvm/GlobalVariable.h</a>"</tt>
3875<br>
3876doxygen info: <a href="/doxygen/classllvm_1_1GlobalVariable.html">GlobalVariable
3877 Class</a><br>
3878Superclasses: <a href="#GlobalValue"><tt>GlobalValue</tt></a>,
3879<a href="#Constant"><tt>Constant</tt></a>,
3880<a href="#User"><tt>User</tt></a>,
3881<a href="#Value"><tt>Value</tt></a></p>
3882
3883<p>Global variables are represented with the (surprise surprise)
3884<tt>GlobalVariable</tt> class. Like functions, <tt>GlobalVariable</tt>s are also
3885subclasses of <a href="#GlobalValue"><tt>GlobalValue</tt></a>, and as such are
3886always referenced by their address (global values must live in memory, so their
3887"name" refers to their constant address). See
3888<a href="#GlobalValue"><tt>GlobalValue</tt></a> for more on this.  Global
3889variables may have an initial value (which must be a
3890<a href="#Constant"><tt>Constant</tt></a>), and if they have an initializer,
3891they may be marked as "constant" themselves (indicating that their contents
3892never change at runtime).</p>
3893
3894<!-- _______________________________________________________________________ -->
3895<h4>
3896  <a name="m_GlobalVariable">
3897    Important Public Members of the <tt>GlobalVariable</tt> class
3898  </a>
3899</h4>
3900
3901<div>
3902
3903<ul>
3904  <li><tt>GlobalVariable(const </tt><tt><a href="#Type">Type</a> *Ty, bool
3905  isConstant, LinkageTypes&amp; Linkage, <a href="#Constant">Constant</a>
3906  *Initializer = 0, const std::string &amp;Name = "", Module* Parent = 0)</tt>
3907
3908    <p>Create a new global variable of the specified type. If
3909    <tt>isConstant</tt> is true then the global variable will be marked as
3910    unchanging for the program. The Linkage parameter specifies the type of
3911    linkage (internal, external, weak, linkonce, appending) for the variable.
3912    If the linkage is InternalLinkage, WeakAnyLinkage, WeakODRLinkage,
3913    LinkOnceAnyLinkage or LinkOnceODRLinkage,&nbsp; then the resultant
3914    global variable will have internal linkage.  AppendingLinkage concatenates
3915    together all instances (in different translation units) of the variable
3916    into a single variable but is only applicable to arrays.  &nbsp;See
3917    the <a href="LangRef.html#modulestructure">LLVM Language Reference</a> for
3918    further details on linkage types. Optionally an initializer, a name, and the
3919    module to put the variable into may be specified for the global variable as
3920    well.</p></li>
3921
3922  <li><tt>bool isConstant() const</tt>
3923
3924    <p>Returns true if this is a global variable that is known not to
3925    be modified at runtime.</p></li>
3926
3927  <li><tt>bool hasInitializer()</tt>
3928
3929    <p>Returns true if this <tt>GlobalVariable</tt> has an intializer.</p></li>
3930
3931  <li><tt><a href="#Constant">Constant</a> *getInitializer()</tt>
3932
3933    <p>Returns the initial value for a <tt>GlobalVariable</tt>.  It is not legal
3934    to call this method if there is no initializer.</p></li>
3935</ul>
3936
3937</div>
3938
3939</div>
3940
3941<!-- ======================================================================= -->
3942<h3>
3943  <a name="BasicBlock">The <tt>BasicBlock</tt> class</a>
3944</h3>
3945
3946<div>
3947
3948<p><tt>#include "<a
3949href="/doxygen/BasicBlock_8h-source.html">llvm/BasicBlock.h</a>"</tt><br>
3950doxygen info: <a href="/doxygen/classllvm_1_1BasicBlock.html">BasicBlock
3951Class</a><br>
3952Superclass: <a href="#Value"><tt>Value</tt></a></p>
3953
3954<p>This class represents a single entry single exit section of the code,
3955commonly known as a basic block by the compiler community.  The
3956<tt>BasicBlock</tt> class maintains a list of <a
3957href="#Instruction"><tt>Instruction</tt></a>s, which form the body of the block.
3958Matching the language definition, the last element of this list of instructions
3959is always a terminator instruction (a subclass of the <a
3960href="#TerminatorInst"><tt>TerminatorInst</tt></a> class).</p>
3961
3962<p>In addition to tracking the list of instructions that make up the block, the
3963<tt>BasicBlock</tt> class also keeps track of the <a
3964href="#Function"><tt>Function</tt></a> that it is embedded into.</p>
3965
3966<p>Note that <tt>BasicBlock</tt>s themselves are <a
3967href="#Value"><tt>Value</tt></a>s, because they are referenced by instructions
3968like branches and can go in the switch tables. <tt>BasicBlock</tt>s have type
3969<tt>label</tt>.</p>
3970
3971<!-- _______________________________________________________________________ -->
3972<h4>
3973  <a name="m_BasicBlock">
3974    Important Public Members of the <tt>BasicBlock</tt> class
3975  </a>
3976</h4>
3977
3978<div>
3979<ul>
3980
3981<li><tt>BasicBlock(const std::string &amp;Name = "", </tt><tt><a
3982 href="#Function">Function</a> *Parent = 0)</tt>
3983
3984<p>The <tt>BasicBlock</tt> constructor is used to create new basic blocks for
3985insertion into a function.  The constructor optionally takes a name for the new
3986block, and a <a href="#Function"><tt>Function</tt></a> to insert it into.  If
3987the <tt>Parent</tt> parameter is specified, the new <tt>BasicBlock</tt> is
3988automatically inserted at the end of the specified <a
3989href="#Function"><tt>Function</tt></a>, if not specified, the BasicBlock must be
3990manually inserted into the <a href="#Function"><tt>Function</tt></a>.</p></li>
3991
3992<li><tt>BasicBlock::iterator</tt> - Typedef for instruction list iterator<br>
3993<tt>BasicBlock::const_iterator</tt> - Typedef for const_iterator.<br>
3994<tt>begin()</tt>, <tt>end()</tt>, <tt>front()</tt>, <tt>back()</tt>,
3995<tt>size()</tt>, <tt>empty()</tt>
3996STL-style functions for accessing the instruction list.
3997
3998<p>These methods and typedefs are forwarding functions that have the same
3999semantics as the standard library methods of the same names.  These methods
4000expose the underlying instruction list of a basic block in a way that is easy to
4001manipulate.  To get the full complement of container operations (including
4002operations to update the list), you must use the <tt>getInstList()</tt>
4003method.</p></li>
4004
4005<li><tt>BasicBlock::InstListType &amp;getInstList()</tt>
4006
4007<p>This method is used to get access to the underlying container that actually
4008holds the Instructions.  This method must be used when there isn't a forwarding
4009function in the <tt>BasicBlock</tt> class for the operation that you would like
4010to perform.  Because there are no forwarding functions for "updating"
4011operations, you need to use this if you want to update the contents of a
4012<tt>BasicBlock</tt>.</p></li>
4013
4014<li><tt><a href="#Function">Function</a> *getParent()</tt>
4015
4016<p> Returns a pointer to <a href="#Function"><tt>Function</tt></a> the block is
4017embedded into, or a null pointer if it is homeless.</p></li>
4018
4019<li><tt><a href="#TerminatorInst">TerminatorInst</a> *getTerminator()</tt>
4020
4021<p> Returns a pointer to the terminator instruction that appears at the end of
4022the <tt>BasicBlock</tt>.  If there is no terminator instruction, or if the last
4023instruction in the block is not a terminator, then a null pointer is
4024returned.</p></li>
4025
4026</ul>
4027
4028</div>
4029
4030</div>
4031
4032<!-- ======================================================================= -->
4033<h3>
4034  <a name="Argument">The <tt>Argument</tt> class</a>
4035</h3>
4036
4037<div>
4038
4039<p>This subclass of Value defines the interface for incoming formal
4040arguments to a function. A Function maintains a list of its formal
4041arguments. An argument has a pointer to the parent Function.</p>
4042
4043</div>
4044
4045</div>
4046
4047<!-- *********************************************************************** -->
4048<hr>
4049<address>
4050  <a href="http://jigsaw.w3.org/css-validator/check/referer"><img
4051  src="http://jigsaw.w3.org/css-validator/images/vcss-blue" alt="Valid CSS"></a>
4052  <a href="http://validator.w3.org/check/referer"><img
4053  src="http://www.w3.org/Icons/valid-html401" alt="Valid HTML 4.01 Strict"></a>
4054
4055  <a href="mailto:dhurjati@cs.uiuc.edu">Dinakar Dhurjati</a> and
4056  <a href="mailto:sabre@nondot.org">Chris Lattner</a><br>
4057  <a href="http://llvm.org/">The LLVM Compiler Infrastructure</a><br>
4058  Last modified: $Date: 2011-10-11 02:33:56 -0400 (Tue, 11 Oct 2011) $
4059</address>
4060
4061</body>
4062</html>
4063