blob: ec388555adadd04c43e02498a232cdfea25ccaf4 [file] [log] [blame]
Shih-wei Liaoe264f622010-02-10 11:10:31 -08001<!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<div class="doc_title">
12 LLVM Programmer's Manual
13</div>
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_fixedarrays">Fixed Size Arrays</a></li>
60 <li><a href="#dss_heaparrays">Heap Allocated Arrays</a></li>
61 <li><a href="#dss_smallvector">"llvm/ADT/SmallVector.h"</a></li>
62 <li><a href="#dss_vector">&lt;vector&gt;</a></li>
63 <li><a href="#dss_deque">&lt;deque&gt;</a></li>
64 <li><a href="#dss_list">&lt;list&gt;</a></li>
65 <li><a href="#dss_ilist">llvm/ADT/ilist.h</a></li>
66 <li><a href="#dss_other">Other Sequential Container Options</a></li>
67 </ul></li>
68 <li><a href="#ds_set">Set-Like Containers (std::set, SmallSet, SetVector, etc)</a>
69 <ul>
70 <li><a href="#dss_sortedvectorset">A sorted 'vector'</a></li>
71 <li><a href="#dss_smallset">"llvm/ADT/SmallSet.h"</a></li>
72 <li><a href="#dss_smallptrset">"llvm/ADT/SmallPtrSet.h"</a></li>
73 <li><a href="#dss_denseset">"llvm/ADT/DenseSet.h"</a></li>
74 <li><a href="#dss_FoldingSet">"llvm/ADT/FoldingSet.h"</a></li>
75 <li><a href="#dss_set">&lt;set&gt;</a></li>
76 <li><a href="#dss_setvector">"llvm/ADT/SetVector.h"</a></li>
77 <li><a href="#dss_uniquevector">"llvm/ADT/UniqueVector.h"</a></li>
78 <li><a href="#dss_otherset">Other Set-Like ContainerOptions</a></li>
79 </ul></li>
80 <li><a href="#ds_map">Map-Like Containers (std::map, DenseMap, etc)</a>
81 <ul>
82 <li><a href="#dss_sortedvectormap">A sorted 'vector'</a></li>
83 <li><a href="#dss_stringmap">"llvm/ADT/StringMap.h"</a></li>
84 <li><a href="#dss_indexedmap">"llvm/ADT/IndexedMap.h"</a></li>
85 <li><a href="#dss_densemap">"llvm/ADT/DenseMap.h"</a></li>
86 <li><a href="#dss_valuemap">"llvm/ADT/ValueMap.h"</a></li>
87 <li><a href="#dss_map">&lt;map&gt;</a></li>
88 <li><a href="#dss_othermap">Other Map-Like Container Options</a></li>
89 </ul></li>
90 <li><a href="#ds_string">String-like containers</a>
91 <!--<ul>
92 todo
93 </ul>--></li>
94 <li><a href="#ds_bit">BitVector-like containers</a>
95 <ul>
96 <li><a href="#dss_bitvector">A dense bitvector</a></li>
97 <li><a href="#dss_smallbitvector">A "small" dense bitvector</a></li>
98 <li><a href="#dss_sparsebitvector">A sparse bitvector</a></li>
99 </ul></li>
100 </ul>
101 </li>
102 <li><a href="#common">Helpful Hints for Common Operations</a>
103 <ul>
104 <li><a href="#inspection">Basic Inspection and Traversal Routines</a>
105 <ul>
106 <li><a href="#iterate_function">Iterating over the <tt>BasicBlock</tt>s
107in a <tt>Function</tt></a> </li>
108 <li><a href="#iterate_basicblock">Iterating over the <tt>Instruction</tt>s
109in a <tt>BasicBlock</tt></a> </li>
110 <li><a href="#iterate_institer">Iterating over the <tt>Instruction</tt>s
111in a <tt>Function</tt></a> </li>
112 <li><a href="#iterate_convert">Turning an iterator into a
113class pointer</a> </li>
114 <li><a href="#iterate_complex">Finding call sites: a more
115complex example</a> </li>
116 <li><a href="#calls_and_invokes">Treating calls and invokes
117the same way</a> </li>
118 <li><a href="#iterate_chains">Iterating over def-use &amp;
119use-def chains</a> </li>
120 <li><a href="#iterate_preds">Iterating over predecessors &amp;
121successors of blocks</a></li>
122 </ul>
123 </li>
124 <li><a href="#simplechanges">Making simple changes</a>
125 <ul>
126 <li><a href="#schanges_creating">Creating and inserting new
127 <tt>Instruction</tt>s</a> </li>
128 <li><a href="#schanges_deleting">Deleting <tt>Instruction</tt>s</a> </li>
129 <li><a href="#schanges_replacing">Replacing an <tt>Instruction</tt>
130with another <tt>Value</tt></a> </li>
131 <li><a href="#schanges_deletingGV">Deleting <tt>GlobalVariable</tt>s</a> </li>
132 </ul>
133 </li>
134 <li><a href="#create_types">How to Create Types</a></li>
135<!--
136 <li>Working with the Control Flow Graph
137 <ul>
138 <li>Accessing predecessors and successors of a <tt>BasicBlock</tt>
139 <li>
140 <li>
141 </ul>
142-->
143 </ul>
144 </li>
145
146 <li><a href="#threading">Threads and LLVM</a>
147 <ul>
148 <li><a href="#startmultithreaded">Entering and Exiting Multithreaded Mode
149 </a></li>
150 <li><a href="#shutdown">Ending execution with <tt>llvm_shutdown()</tt></a></li>
151 <li><a href="#managedstatic">Lazy initialization with <tt>ManagedStatic</tt></a></li>
152 <li><a href="#llvmcontext">Achieving Isolation with <tt>LLVMContext</tt></a></li>
153 <li><a href="#jitthreading">Threads and the JIT</a></li>
154 </ul>
155 </li>
156
157 <li><a href="#advanced">Advanced Topics</a>
158 <ul>
159 <li><a href="#TypeResolve">LLVM Type Resolution</a>
160 <ul>
161 <li><a href="#BuildRecType">Basic Recursive Type Construction</a></li>
162 <li><a href="#refineAbstractTypeTo">The <tt>refineAbstractTypeTo</tt> method</a></li>
163 <li><a href="#PATypeHolder">The PATypeHolder Class</a></li>
164 <li><a href="#AbstractTypeUser">The AbstractTypeUser Class</a></li>
165 </ul></li>
166
167 <li><a href="#SymbolTable">The <tt>ValueSymbolTable</tt> and <tt>TypeSymbolTable</tt> classes</a></li>
168 <li><a href="#UserLayout">The <tt>User</tt> and owned <tt>Use</tt> classes' memory layout</a></li>
169 </ul></li>
170
171 <li><a href="#coreclasses">The Core LLVM Class Hierarchy Reference</a>
172 <ul>
173 <li><a href="#Type">The <tt>Type</tt> class</a> </li>
174 <li><a href="#Module">The <tt>Module</tt> class</a></li>
175 <li><a href="#Value">The <tt>Value</tt> class</a>
176 <ul>
177 <li><a href="#User">The <tt>User</tt> class</a>
178 <ul>
179 <li><a href="#Instruction">The <tt>Instruction</tt> class</a></li>
180 <li><a href="#Constant">The <tt>Constant</tt> class</a>
181 <ul>
182 <li><a href="#GlobalValue">The <tt>GlobalValue</tt> class</a>
183 <ul>
184 <li><a href="#Function">The <tt>Function</tt> class</a></li>
185 <li><a href="#GlobalVariable">The <tt>GlobalVariable</tt> class</a></li>
186 </ul>
187 </li>
188 </ul>
189 </li>
190 </ul>
191 </li>
192 <li><a href="#BasicBlock">The <tt>BasicBlock</tt> class</a></li>
193 <li><a href="#Argument">The <tt>Argument</tt> class</a></li>
194 </ul>
195 </li>
196 </ul>
197 </li>
198</ol>
199
200<div class="doc_author">
201 <p>Written by <a href="mailto:sabre@nondot.org">Chris Lattner</a>,
202 <a href="mailto:dhurjati@cs.uiuc.edu">Dinakar Dhurjati</a>,
203 <a href="mailto:ggreif@gmail.com">Gabor Greif</a>,
204 <a href="mailto:jstanley@cs.uiuc.edu">Joel Stanley</a>,
205 <a href="mailto:rspencer@x10sys.com">Reid Spencer</a> and
206 <a href="mailto:owen@apple.com">Owen Anderson</a></p>
207</div>
208
209<!-- *********************************************************************** -->
210<div class="doc_section">
211 <a name="introduction">Introduction </a>
212</div>
213<!-- *********************************************************************** -->
214
215<div class="doc_text">
216
217<p>This document is meant to highlight some of the important classes and
218interfaces available in the LLVM source-base. This manual is not
219intended to explain what LLVM is, how it works, and what LLVM code looks
220like. It assumes that you know the basics of LLVM and are interested
221in writing transformations or otherwise analyzing or manipulating the
222code.</p>
223
224<p>This document should get you oriented so that you can find your
225way in the continuously growing source code that makes up the LLVM
226infrastructure. Note that this manual is not intended to serve as a
227replacement for reading the source code, so if you think there should be
228a method in one of these classes to do something, but it's not listed,
229check the source. Links to the <a href="/doxygen/">doxygen</a> sources
230are provided to make this as easy as possible.</p>
231
232<p>The first section of this document describes general information that is
233useful to know when working in the LLVM infrastructure, and the second describes
234the Core LLVM classes. In the future this manual will be extended with
235information describing how to use extension libraries, such as dominator
236information, CFG traversal routines, and useful utilities like the <tt><a
237href="/doxygen/InstVisitor_8h-source.html">InstVisitor</a></tt> template.</p>
238
239</div>
240
241<!-- *********************************************************************** -->
242<div class="doc_section">
243 <a name="general">General Information</a>
244</div>
245<!-- *********************************************************************** -->
246
247<div class="doc_text">
248
249<p>This section contains general information that is useful if you are working
250in the LLVM source-base, but that isn't specific to any particular API.</p>
251
252</div>
253
254<!-- ======================================================================= -->
255<div class="doc_subsection">
256 <a name="stl">The C++ Standard Template Library</a>
257</div>
258
259<div class="doc_text">
260
261<p>LLVM makes heavy use of the C++ Standard Template Library (STL),
262perhaps much more than you are used to, or have seen before. Because of
263this, you might want to do a little background reading in the
264techniques used and capabilities of the library. There are many good
265pages that discuss the STL, and several books on the subject that you
266can get, so it will not be discussed in this document.</p>
267
268<p>Here are some useful links:</p>
269
270<ol>
271
272<li><a href="http://www.dinkumware.com/refxcpp.html">Dinkumware C++ Library
273reference</a> - an excellent reference for the STL and other parts of the
274standard C++ library.</li>
275
276<li><a href="http://www.tempest-sw.com/cpp/">C++ In a Nutshell</a> - This is an
277O'Reilly book in the making. It has a decent Standard Library
278Reference that rivals Dinkumware's, and is unfortunately no longer free since the
279book has been published.</li>
280
281<li><a href="http://www.parashift.com/c++-faq-lite/">C++ Frequently Asked
282Questions</a></li>
283
284<li><a href="http://www.sgi.com/tech/stl/">SGI's STL Programmer's Guide</a> -
285Contains a useful <a
286href="http://www.sgi.com/tech/stl/stl_introduction.html">Introduction to the
287STL</a>.</li>
288
289<li><a href="http://www.research.att.com/%7Ebs/C++.html">Bjarne Stroustrup's C++
290Page</a></li>
291
292<li><a href="http://64.78.49.204/">
293Bruce Eckel's Thinking in C++, 2nd ed. Volume 2 Revision 4.0 (even better, get
294the book).</a></li>
295
296</ol>
297
298<p>You are also encouraged to take a look at the <a
299href="CodingStandards.html">LLVM Coding Standards</a> guide which focuses on how
300to write maintainable code more than where to put your curly braces.</p>
301
302</div>
303
304<!-- ======================================================================= -->
305<div class="doc_subsection">
306 <a name="stl">Other useful references</a>
307</div>
308
309<div class="doc_text">
310
311<ol>
312<li><a href="http://www.psc.edu/%7Esemke/cvs_branches.html">CVS
313Branch and Tag Primer</a></li>
314<li><a href="http://www.fortran-2000.com/ArnaudRecipes/sharedlib.html">Using
315static and shared libraries across platforms</a></li>
316</ol>
317
318</div>
319
320<!-- *********************************************************************** -->
321<div class="doc_section">
322 <a name="apis">Important and useful LLVM APIs</a>
323</div>
324<!-- *********************************************************************** -->
325
326<div class="doc_text">
327
328<p>Here we highlight some LLVM APIs that are generally useful and good to
329know about when writing transformations.</p>
330
331</div>
332
333<!-- ======================================================================= -->
334<div class="doc_subsection">
335 <a name="isa">The <tt>isa&lt;&gt;</tt>, <tt>cast&lt;&gt;</tt> and
336 <tt>dyn_cast&lt;&gt;</tt> templates</a>
337</div>
338
339<div class="doc_text">
340
341<p>The LLVM source-base makes extensive use of a custom form of RTTI.
342These templates have many similarities to the C++ <tt>dynamic_cast&lt;&gt;</tt>
343operator, but they don't have some drawbacks (primarily stemming from
344the fact that <tt>dynamic_cast&lt;&gt;</tt> only works on classes that
345have a v-table). Because they are used so often, you must know what they
346do and how they work. All of these templates are defined in the <a
347 href="/doxygen/Casting_8h-source.html"><tt>llvm/Support/Casting.h</tt></a>
348file (note that you very rarely have to include this file directly).</p>
349
350<dl>
351 <dt><tt>isa&lt;&gt;</tt>: </dt>
352
353 <dd><p>The <tt>isa&lt;&gt;</tt> operator works exactly like the Java
354 "<tt>instanceof</tt>" operator. It returns true or false depending on whether
355 a reference or pointer points to an instance of the specified class. This can
356 be very useful for constraint checking of various sorts (example below).</p>
357 </dd>
358
359 <dt><tt>cast&lt;&gt;</tt>: </dt>
360
361 <dd><p>The <tt>cast&lt;&gt;</tt> operator is a "checked cast" operation. It
362 converts a pointer or reference from a base class to a derived class, causing
363 an assertion failure if it is not really an instance of the right type. This
364 should be used in cases where you have some information that makes you believe
365 that something is of the right type. An example of the <tt>isa&lt;&gt;</tt>
366 and <tt>cast&lt;&gt;</tt> template is:</p>
367
368<div class="doc_code">
369<pre>
370static bool isLoopInvariant(const <a href="#Value">Value</a> *V, const Loop *L) {
371 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))
372 return true;
373
374 // <i>Otherwise, it must be an instruction...</i>
375 return !L-&gt;contains(cast&lt;<a href="#Instruction">Instruction</a>&gt;(V)-&gt;getParent());
376}
377</pre>
378</div>
379
380 <p>Note that you should <b>not</b> use an <tt>isa&lt;&gt;</tt> test followed
381 by a <tt>cast&lt;&gt;</tt>, for that use the <tt>dyn_cast&lt;&gt;</tt>
382 operator.</p>
383
384 </dd>
385
386 <dt><tt>dyn_cast&lt;&gt;</tt>:</dt>
387
388 <dd><p>The <tt>dyn_cast&lt;&gt;</tt> operator is a "checking cast" operation.
389 It checks to see if the operand is of the specified type, and if so, returns a
390 pointer to it (this operator does not work with references). If the operand is
391 not of the correct type, a null pointer is returned. Thus, this works very
392 much like the <tt>dynamic_cast&lt;&gt;</tt> operator in C++, and should be
393 used in the same circumstances. Typically, the <tt>dyn_cast&lt;&gt;</tt>
394 operator is used in an <tt>if</tt> statement or some other flow control
395 statement like this:</p>
396
397<div class="doc_code">
398<pre>
399if (<a href="#AllocationInst">AllocationInst</a> *AI = dyn_cast&lt;<a href="#AllocationInst">AllocationInst</a>&gt;(Val)) {
400 // <i>...</i>
401}
402</pre>
403</div>
404
405 <p>This form of the <tt>if</tt> statement effectively combines together a call
406 to <tt>isa&lt;&gt;</tt> and a call to <tt>cast&lt;&gt;</tt> into one
407 statement, which is very convenient.</p>
408
409 <p>Note that the <tt>dyn_cast&lt;&gt;</tt> operator, like C++'s
410 <tt>dynamic_cast&lt;&gt;</tt> or Java's <tt>instanceof</tt> operator, can be
411 abused. In particular, you should not use big chained <tt>if/then/else</tt>
412 blocks to check for lots of different variants of classes. If you find
413 yourself wanting to do this, it is much cleaner and more efficient to use the
414 <tt>InstVisitor</tt> class to dispatch over the instruction type directly.</p>
415
416 </dd>
417
418 <dt><tt>cast_or_null&lt;&gt;</tt>: </dt>
419
420 <dd><p>The <tt>cast_or_null&lt;&gt;</tt> operator works just like the
421 <tt>cast&lt;&gt;</tt> operator, except that it allows for a null pointer as an
422 argument (which it then propagates). This can sometimes be useful, allowing
423 you to combine several null checks into one.</p></dd>
424
425 <dt><tt>dyn_cast_or_null&lt;&gt;</tt>: </dt>
426
427 <dd><p>The <tt>dyn_cast_or_null&lt;&gt;</tt> operator works just like the
428 <tt>dyn_cast&lt;&gt;</tt> operator, except that it allows for a null pointer
429 as an argument (which it then propagates). This can sometimes be useful,
430 allowing you to combine several null checks into one.</p></dd>
431
432</dl>
433
434<p>These five templates can be used with any classes, whether they have a
435v-table or not. To add support for these templates, you simply need to add
436<tt>classof</tt> static methods to the class you are interested casting
437to. Describing this is currently outside the scope of this document, but there
438are lots of examples in the LLVM source base.</p>
439
440</div>
441
442
443<!-- ======================================================================= -->
444<div class="doc_subsection">
445 <a name="string_apis">Passing strings (the <tt>StringRef</tt>
446and <tt>Twine</tt> classes)</a>
447</div>
448
449<div class="doc_text">
450
451<p>Although LLVM generally does not do much string manipulation, we do have
452several important APIs which take strings. Two important examples are the
453Value class -- which has names for instructions, functions, etc. -- and the
454StringMap class which is used extensively in LLVM and Clang.</p>
455
456<p>These are generic classes, and they need to be able to accept strings which
457may have embedded null characters. Therefore, they cannot simply take
458a <tt>const char *</tt>, and taking a <tt>const std::string&amp;</tt> requires
459clients to perform a heap allocation which is usually unnecessary. Instead,
460many LLVM APIs use a <tt>const StringRef&amp;</tt> or a <tt>const
461Twine&amp;</tt> for passing strings efficiently.</p>
462
463</div>
464
465<!-- _______________________________________________________________________ -->
466<div class="doc_subsubsection">
467 <a name="StringRef">The <tt>StringRef</tt> class</a>
468</div>
469
470<div class="doc_text">
471
472<p>The <tt>StringRef</tt> data type represents a reference to a constant string
473(a character array and a length) and supports the common operations available
474on <tt>std:string</tt>, but does not require heap allocation.</p>
475
476<p>It can be implicitly constructed using a C style null-terminated string,
477an <tt>std::string</tt>, or explicitly with a character pointer and length.
478For example, the <tt>StringRef</tt> find function is declared as:</p>
479
480<div class="doc_code">
481 iterator find(const StringRef &amp;Key);
482</div>
483
484<p>and clients can call it using any one of:</p>
485
486<div class="doc_code">
487<pre>
488 Map.find("foo"); <i>// Lookup "foo"</i>
489 Map.find(std::string("bar")); <i>// Lookup "bar"</i>
490 Map.find(StringRef("\0baz", 4)); <i>// Lookup "\0baz"</i>
491</pre>
492</div>
493
494<p>Similarly, APIs which need to return a string may return a <tt>StringRef</tt>
495instance, which can be used directly or converted to an <tt>std::string</tt>
496using the <tt>str</tt> member function. See
497"<tt><a href="/doxygen/classllvm_1_1StringRef_8h-source.html">llvm/ADT/StringRef.h</a></tt>"
498for more information.</p>
499
500<p>You should rarely use the <tt>StringRef</tt> class directly, because it contains
501pointers to external memory it is not generally safe to store an instance of the
502class (unless you know that the external storage will not be freed).</p>
503
504</div>
505
506<!-- _______________________________________________________________________ -->
507<div class="doc_subsubsection">
508 <a name="Twine">The <tt>Twine</tt> class</a>
509</div>
510
511<div class="doc_text">
512
513<p>The <tt>Twine</tt> class is an efficient way for APIs to accept concatenated
514strings. For example, a common LLVM paradigm is to name one instruction based on
515the name of another instruction with a suffix, for example:</p>
516
517<div class="doc_code">
518<pre>
519 New = CmpInst::Create(<i>...</i>, SO->getName() + ".cmp");
520</pre>
521</div>
522
523<p>The <tt>Twine</tt> class is effectively a
524lightweight <a href="http://en.wikipedia.org/wiki/Rope_(computer_science)">rope</a>
525which points to temporary (stack allocated) objects. Twines can be implicitly
526constructed as the result of the plus operator applied to strings (i.e., a C
527strings, an <tt>std::string</tt>, or a <tt>StringRef</tt>). The twine delays the
528actual concatentation of strings until it is actually required, at which point
529it can be efficiently rendered directly into a character array. This avoids
530unnecessary heap allocation involved in constructing the temporary results of
531string concatenation. See
532"<tt><a href="/doxygen/classllvm_1_1Twine_8h-source.html">llvm/ADT/Twine.h</a></tt>"
533for more information.</p>
534
535<p>As with a <tt>StringRef</tt>, <tt>Twine</tt> objects point to external memory
536and should almost never be stored or mentioned directly. They are intended
537solely for use when defining a function which should be able to efficiently
538accept concatenated strings.</p>
539
540</div>
541
542
543<!-- ======================================================================= -->
544<div class="doc_subsection">
545 <a name="DEBUG">The <tt>DEBUG()</tt> macro and <tt>-debug</tt> option</a>
546</div>
547
548<div class="doc_text">
549
550<p>Often when working on your pass you will put a bunch of debugging printouts
551and other code into your pass. After you get it working, you want to remove
552it, but you may need it again in the future (to work out new bugs that you run
553across).</p>
554
555<p> Naturally, because of this, you don't want to delete the debug printouts,
556but you don't want them to always be noisy. A standard compromise is to comment
557them out, allowing you to enable them if you need them in the future.</p>
558
559<p>The "<tt><a href="/doxygen/Debug_8h-source.html">llvm/Support/Debug.h</a></tt>"
560file provides a macro named <tt>DEBUG()</tt> that is a much nicer solution to
561this problem. Basically, you can put arbitrary code into the argument of the
562<tt>DEBUG</tt> macro, and it is only executed if '<tt>opt</tt>' (or any other
563tool) is run with the '<tt>-debug</tt>' command line argument:</p>
564
565<div class="doc_code">
566<pre>
567DEBUG(errs() &lt;&lt; "I am here!\n");
568</pre>
569</div>
570
571<p>Then you can run your pass like this:</p>
572
573<div class="doc_code">
574<pre>
575$ opt &lt; a.bc &gt; /dev/null -mypass
576<i>&lt;no output&gt;</i>
577$ opt &lt; a.bc &gt; /dev/null -mypass -debug
578I am here!
579</pre>
580</div>
581
582<p>Using the <tt>DEBUG()</tt> macro instead of a home-brewed solution allows you
583to not have to create "yet another" command line option for the debug output for
584your pass. Note that <tt>DEBUG()</tt> macros are disabled for optimized builds,
585so they do not cause a performance impact at all (for the same reason, they
586should also not contain side-effects!).</p>
587
588<p>One additional nice thing about the <tt>DEBUG()</tt> macro is that you can
589enable or disable it directly in gdb. Just use "<tt>set DebugFlag=0</tt>" or
590"<tt>set DebugFlag=1</tt>" from the gdb if the program is running. If the
591program hasn't been started yet, you can always just run it with
592<tt>-debug</tt>.</p>
593
594</div>
595
596<!-- _______________________________________________________________________ -->
597<div class="doc_subsubsection">
598 <a name="DEBUG_TYPE">Fine grained debug info with <tt>DEBUG_TYPE</tt> and
599 the <tt>-debug-only</tt> option</a>
600</div>
601
602<div class="doc_text">
603
604<p>Sometimes you may find yourself in a situation where enabling <tt>-debug</tt>
605just turns on <b>too much</b> information (such as when working on the code
606generator). If you want to enable debug information with more fine-grained
607control, you define the <tt>DEBUG_TYPE</tt> macro and the <tt>-debug</tt> only
608option as follows:</p>
609
610<div class="doc_code">
611<pre>
612#undef DEBUG_TYPE
613DEBUG(errs() &lt;&lt; "No debug type\n");
614#define DEBUG_TYPE "foo"
615DEBUG(errs() &lt;&lt; "'foo' debug type\n");
616#undef DEBUG_TYPE
617#define DEBUG_TYPE "bar"
618DEBUG(errs() &lt;&lt; "'bar' debug type\n"));
619#undef DEBUG_TYPE
620#define DEBUG_TYPE ""
621DEBUG(errs() &lt;&lt; "No debug type (2)\n");
622</pre>
623</div>
624
625<p>Then you can run your pass like this:</p>
626
627<div class="doc_code">
628<pre>
629$ opt &lt; a.bc &gt; /dev/null -mypass
630<i>&lt;no output&gt;</i>
631$ opt &lt; a.bc &gt; /dev/null -mypass -debug
632No debug type
633'foo' debug type
634'bar' debug type
635No debug type (2)
636$ opt &lt; a.bc &gt; /dev/null -mypass -debug-only=foo
637'foo' debug type
638$ opt &lt; a.bc &gt; /dev/null -mypass -debug-only=bar
639'bar' debug type
640</pre>
641</div>
642
643<p>Of course, in practice, you should only set <tt>DEBUG_TYPE</tt> at the top of
644a file, to specify the debug type for the entire module (if you do this before
645you <tt>#include "llvm/Support/Debug.h"</tt>, you don't have to insert the ugly
646<tt>#undef</tt>'s). Also, you should use names more meaningful than "foo" and
647"bar", because there is no system in place to ensure that names do not
648conflict. If two different modules use the same string, they will all be turned
649on when the name is specified. This allows, for example, all debug information
650for instruction scheduling to be enabled with <tt>-debug-type=InstrSched</tt>,
651even if the source lives in multiple files.</p>
652
653<p>The <tt>DEBUG_WITH_TYPE</tt> macro is also available for situations where you
654would like to set <tt>DEBUG_TYPE</tt>, but only for one specific <tt>DEBUG</tt>
655statement. It takes an additional first parameter, which is the type to use. For
656example, the preceding example could be written as:</p>
657
658
659<div class="doc_code">
660<pre>
661DEBUG_WITH_TYPE("", errs() &lt;&lt; "No debug type\n");
662DEBUG_WITH_TYPE("foo", errs() &lt;&lt; "'foo' debug type\n");
663DEBUG_WITH_TYPE("bar", errs() &lt;&lt; "'bar' debug type\n"));
664DEBUG_WITH_TYPE("", errs() &lt;&lt; "No debug type (2)\n");
665</pre>
666</div>
667
668</div>
669
670<!-- ======================================================================= -->
671<div class="doc_subsection">
672 <a name="Statistic">The <tt>Statistic</tt> class &amp; <tt>-stats</tt>
673 option</a>
674</div>
675
676<div class="doc_text">
677
678<p>The "<tt><a
679href="/doxygen/Statistic_8h-source.html">llvm/ADT/Statistic.h</a></tt>" file
680provides a class named <tt>Statistic</tt> that is used as a unified way to
681keep track of what the LLVM compiler is doing and how effective various
682optimizations are. It is useful to see what optimizations are contributing to
683making a particular program run faster.</p>
684
685<p>Often you may run your pass on some big program, and you're interested to see
686how many times it makes a certain transformation. Although you can do this with
687hand inspection, or some ad-hoc method, this is a real pain and not very useful
688for big programs. Using the <tt>Statistic</tt> class makes it very easy to
689keep track of this information, and the calculated information is presented in a
690uniform manner with the rest of the passes being executed.</p>
691
692<p>There are many examples of <tt>Statistic</tt> uses, but the basics of using
693it are as follows:</p>
694
695<ol>
696 <li><p>Define your statistic like this:</p>
697
698<div class="doc_code">
699<pre>
700#define <a href="#DEBUG_TYPE">DEBUG_TYPE</a> "mypassname" <i>// This goes before any #includes.</i>
701STATISTIC(NumXForms, "The # of times I did stuff");
702</pre>
703</div>
704
705 <p>The <tt>STATISTIC</tt> macro defines a static variable, whose name is
706 specified by the first argument. The pass name is taken from the DEBUG_TYPE
707 macro, and the description is taken from the second argument. The variable
708 defined ("NumXForms" in this case) acts like an unsigned integer.</p></li>
709
710 <li><p>Whenever you make a transformation, bump the counter:</p>
711
712<div class="doc_code">
713<pre>
714++NumXForms; // <i>I did stuff!</i>
715</pre>
716</div>
717
718 </li>
719 </ol>
720
721 <p>That's all you have to do. To get '<tt>opt</tt>' to print out the
722 statistics gathered, use the '<tt>-stats</tt>' option:</p>
723
724<div class="doc_code">
725<pre>
726$ opt -stats -mypassname &lt; program.bc &gt; /dev/null
727<i>... statistics output ...</i>
728</pre>
729</div>
730
731 <p> When running <tt>opt</tt> on a C file from the SPEC benchmark
732suite, it gives a report that looks like this:</p>
733
734<div class="doc_code">
735<pre>
736 7646 bitcodewriter - Number of normal instructions
737 725 bitcodewriter - Number of oversized instructions
738 129996 bitcodewriter - Number of bitcode bytes written
739 2817 raise - Number of insts DCEd or constprop'd
740 3213 raise - Number of cast-of-self removed
741 5046 raise - Number of expression trees converted
742 75 raise - Number of other getelementptr's formed
743 138 raise - Number of load/store peepholes
744 42 deadtypeelim - Number of unused typenames removed from symtab
745 392 funcresolve - Number of varargs functions resolved
746 27 globaldce - Number of global variables removed
747 2 adce - Number of basic blocks removed
748 134 cee - Number of branches revectored
749 49 cee - Number of setcc instruction eliminated
750 532 gcse - Number of loads removed
751 2919 gcse - Number of instructions removed
752 86 indvars - Number of canonical indvars added
753 87 indvars - Number of aux indvars removed
754 25 instcombine - Number of dead inst eliminate
755 434 instcombine - Number of insts combined
756 248 licm - Number of load insts hoisted
757 1298 licm - Number of insts hoisted to a loop pre-header
758 3 licm - Number of insts hoisted to multiple loop preds (bad, no loop pre-header)
759 75 mem2reg - Number of alloca's promoted
760 1444 cfgsimplify - Number of blocks simplified
761</pre>
762</div>
763
764<p>Obviously, with so many optimizations, having a unified framework for this
765stuff is very nice. Making your pass fit well into the framework makes it more
766maintainable and useful.</p>
767
768</div>
769
770<!-- ======================================================================= -->
771<div class="doc_subsection">
772 <a name="ViewGraph">Viewing graphs while debugging code</a>
773</div>
774
775<div class="doc_text">
776
777<p>Several of the important data structures in LLVM are graphs: for example
778CFGs made out of LLVM <a href="#BasicBlock">BasicBlock</a>s, CFGs made out of
779LLVM <a href="CodeGenerator.html#machinebasicblock">MachineBasicBlock</a>s, and
780<a href="CodeGenerator.html#selectiondag_intro">Instruction Selection
781DAGs</a>. In many cases, while debugging various parts of the compiler, it is
782nice to instantly visualize these graphs.</p>
783
784<p>LLVM provides several callbacks that are available in a debug build to do
785exactly that. If you call the <tt>Function::viewCFG()</tt> method, for example,
786the current LLVM tool will pop up a window containing the CFG for the function
787where each basic block is a node in the graph, and each node contains the
788instructions in the block. Similarly, there also exists
789<tt>Function::viewCFGOnly()</tt> (does not include the instructions), the
790<tt>MachineFunction::viewCFG()</tt> and <tt>MachineFunction::viewCFGOnly()</tt>,
791and the <tt>SelectionDAG::viewGraph()</tt> methods. Within GDB, for example,
792you can usually use something like <tt>call DAG.viewGraph()</tt> to pop
793up a window. Alternatively, you can sprinkle calls to these functions in your
794code in places you want to debug.</p>
795
796<p>Getting this to work requires a small amount of configuration. On Unix
797systems with X11, install the <a href="http://www.graphviz.org">graphviz</a>
798toolkit, and make sure 'dot' and 'gv' are in your path. If you are running on
799Mac OS/X, download and install the Mac OS/X <a
800href="http://www.pixelglow.com/graphviz/">Graphviz program</a>, and add
801<tt>/Applications/Graphviz.app/Contents/MacOS/</tt> (or wherever you install
802it) to your path. Once in your system and path are set up, rerun the LLVM
803configure script and rebuild LLVM to enable this functionality.</p>
804
805<p><tt>SelectionDAG</tt> has been extended to make it easier to locate
806<i>interesting</i> nodes in large complex graphs. From gdb, if you
807<tt>call DAG.setGraphColor(<i>node</i>, "<i>color</i>")</tt>, then the
808next <tt>call DAG.viewGraph()</tt> would highlight the node in the
809specified color (choices of colors can be found at <a
810href="http://www.graphviz.org/doc/info/colors.html">colors</a>.) More
811complex node attributes can be provided with <tt>call
812DAG.setGraphAttrs(<i>node</i>, "<i>attributes</i>")</tt> (choices can be
813found at <a href="http://www.graphviz.org/doc/info/attrs.html">Graph
814Attributes</a>.) If you want to restart and clear all the current graph
815attributes, then you can <tt>call DAG.clearGraphAttrs()</tt>. </p>
816
817</div>
818
819<!-- *********************************************************************** -->
820<div class="doc_section">
821 <a name="datastructure">Picking the Right Data Structure for a Task</a>
822</div>
823<!-- *********************************************************************** -->
824
825<div class="doc_text">
826
827<p>LLVM has a plethora of data structures in the <tt>llvm/ADT/</tt> directory,
828 and we commonly use STL data structures. This section describes the trade-offs
829 you should consider when you pick one.</p>
830
831<p>
832The first step is a choose your own adventure: do you want a sequential
833container, a set-like container, or a map-like container? The most important
834thing when choosing a container is the algorithmic properties of how you plan to
835access the container. Based on that, you should use:</p>
836
837<ul>
838<li>a <a href="#ds_map">map-like</a> container if you need efficient look-up
839 of an value based on another value. Map-like containers also support
840 efficient queries for containment (whether a key is in the map). Map-like
841 containers generally do not support efficient reverse mapping (values to
842 keys). If you need that, use two maps. Some map-like containers also
843 support efficient iteration through the keys in sorted order. Map-like
844 containers are the most expensive sort, only use them if you need one of
845 these capabilities.</li>
846
847<li>a <a href="#ds_set">set-like</a> container if you need to put a bunch of
848 stuff into a container that automatically eliminates duplicates. Some
849 set-like containers support efficient iteration through the elements in
850 sorted order. Set-like containers are more expensive than sequential
851 containers.
852</li>
853
854<li>a <a href="#ds_sequential">sequential</a> container provides
855 the most efficient way to add elements and keeps track of the order they are
856 added to the collection. They permit duplicates and support efficient
857 iteration, but do not support efficient look-up based on a key.
858</li>
859
860<li>a <a href="#ds_string">string</a> container is a specialized sequential
861 container or reference structure that is used for character or byte
862 arrays.</li>
863
864<li>a <a href="#ds_bit">bit</a> container provides an efficient way to store and
865 perform set operations on sets of numeric id's, while automatically
866 eliminating duplicates. Bit containers require a maximum of 1 bit for each
867 identifier you want to store.
868</li>
869</ul>
870
871<p>
872Once the proper category of container is determined, you can fine tune the
873memory use, constant factors, and cache behaviors of access by intelligently
874picking a member of the category. Note that constant factors and cache behavior
875can be a big deal. If you have a vector that usually only contains a few
876elements (but could contain many), for example, it's much better to use
877<a href="#dss_smallvector">SmallVector</a> than <a href="#dss_vector">vector</a>
878. Doing so avoids (relatively) expensive malloc/free calls, which dwarf the
879cost of adding the elements to the container. </p>
880
881</div>
882
883<!-- ======================================================================= -->
884<div class="doc_subsection">
885 <a name="ds_sequential">Sequential Containers (std::vector, std::list, etc)</a>
886</div>
887
888<div class="doc_text">
889There are a variety of sequential containers available for you, based on your
890needs. Pick the first in this section that will do what you want.
891</div>
892
893<!-- _______________________________________________________________________ -->
894<div class="doc_subsubsection">
895 <a name="dss_fixedarrays">Fixed Size Arrays</a>
896</div>
897
898<div class="doc_text">
899<p>Fixed size arrays are very simple and very fast. They are good if you know
900exactly how many elements you have, or you have a (low) upper bound on how many
901you have.</p>
902</div>
903
904<!-- _______________________________________________________________________ -->
905<div class="doc_subsubsection">
906 <a name="dss_heaparrays">Heap Allocated Arrays</a>
907</div>
908
909<div class="doc_text">
910<p>Heap allocated arrays (new[] + delete[]) are also simple. They are good if
911the number of elements is variable, if you know how many elements you will need
912before the array is allocated, and if the array is usually large (if not,
913consider a <a href="#dss_smallvector">SmallVector</a>). The cost of a heap
914allocated array is the cost of the new/delete (aka malloc/free). Also note that
915if you are allocating an array of a type with a constructor, the constructor and
916destructors will be run for every element in the array (re-sizable vectors only
917construct those elements actually used).</p>
918</div>
919
920<!-- _______________________________________________________________________ -->
921<div class="doc_subsubsection">
922 <a name="dss_smallvector">"llvm/ADT/SmallVector.h"</a>
923</div>
924
925<div class="doc_text">
926<p><tt>SmallVector&lt;Type, N&gt;</tt> is a simple class that looks and smells
927just like <tt>vector&lt;Type&gt;</tt>:
928it supports efficient iteration, lays out elements in memory order (so you can
929do pointer arithmetic between elements), supports efficient push_back/pop_back
930operations, supports efficient random access to its elements, etc.</p>
931
932<p>The advantage of SmallVector is that it allocates space for
933some number of elements (N) <b>in the object itself</b>. Because of this, if
934the SmallVector is dynamically smaller than N, no malloc is performed. This can
935be a big win in cases where the malloc/free call is far more expensive than the
936code that fiddles around with the elements.</p>
937
938<p>This is good for vectors that are "usually small" (e.g. the number of
939predecessors/successors of a block is usually less than 8). On the other hand,
940this makes the size of the SmallVector itself large, so you don't want to
941allocate lots of them (doing so will waste a lot of space). As such,
942SmallVectors are most useful when on the stack.</p>
943
944<p>SmallVector also provides a nice portable and efficient replacement for
945<tt>alloca</tt>.</p>
946
947</div>
948
949<!-- _______________________________________________________________________ -->
950<div class="doc_subsubsection">
951 <a name="dss_vector">&lt;vector&gt;</a>
952</div>
953
954<div class="doc_text">
955<p>
956std::vector is well loved and respected. It is useful when SmallVector isn't:
957when the size of the vector is often large (thus the small optimization will
958rarely be a benefit) or if you will be allocating many instances of the vector
959itself (which would waste space for elements that aren't in the container).
960vector is also useful when interfacing with code that expects vectors :).
961</p>
962
963<p>One worthwhile note about std::vector: avoid code like this:</p>
964
965<div class="doc_code">
966<pre>
967for ( ... ) {
968 std::vector&lt;foo&gt; V;
969 use V;
970}
971</pre>
972</div>
973
974<p>Instead, write this as:</p>
975
976<div class="doc_code">
977<pre>
978std::vector&lt;foo&gt; V;
979for ( ... ) {
980 use V;
981 V.clear();
982}
983</pre>
984</div>
985
986<p>Doing so will save (at least) one heap allocation and free per iteration of
987the loop.</p>
988
989</div>
990
991<!-- _______________________________________________________________________ -->
992<div class="doc_subsubsection">
993 <a name="dss_deque">&lt;deque&gt;</a>
994</div>
995
996<div class="doc_text">
997<p>std::deque is, in some senses, a generalized version of std::vector. Like
998std::vector, it provides constant time random access and other similar
999properties, but it also provides efficient access to the front of the list. It
1000does not guarantee continuity of elements within memory.</p>
1001
1002<p>In exchange for this extra flexibility, std::deque has significantly higher
1003constant factor costs than std::vector. If possible, use std::vector or
1004something cheaper.</p>
1005</div>
1006
1007<!-- _______________________________________________________________________ -->
1008<div class="doc_subsubsection">
1009 <a name="dss_list">&lt;list&gt;</a>
1010</div>
1011
1012<div class="doc_text">
1013<p>std::list is an extremely inefficient class that is rarely useful.
1014It performs a heap allocation for every element inserted into it, thus having an
1015extremely high constant factor, particularly for small data types. std::list
1016also only supports bidirectional iteration, not random access iteration.</p>
1017
1018<p>In exchange for this high cost, std::list supports efficient access to both
1019ends of the list (like std::deque, but unlike std::vector or SmallVector). In
1020addition, the iterator invalidation characteristics of std::list are stronger
1021than that of a vector class: inserting or removing an element into the list does
1022not invalidate iterator or pointers to other elements in the list.</p>
1023</div>
1024
1025<!-- _______________________________________________________________________ -->
1026<div class="doc_subsubsection">
1027 <a name="dss_ilist">llvm/ADT/ilist.h</a>
1028</div>
1029
1030<div class="doc_text">
1031<p><tt>ilist&lt;T&gt;</tt> implements an 'intrusive' doubly-linked list. It is
1032intrusive, because it requires the element to store and provide access to the
1033prev/next pointers for the list.</p>
1034
1035<p><tt>ilist</tt> has the same drawbacks as <tt>std::list</tt>, and additionally
1036requires an <tt>ilist_traits</tt> implementation for the element type, but it
1037provides some novel characteristics. In particular, it can efficiently store
1038polymorphic objects, the traits class is informed when an element is inserted or
1039removed from the list, and <tt>ilist</tt>s are guaranteed to support a
1040constant-time splice operation.</p>
1041
1042<p>These properties are exactly what we want for things like
1043<tt>Instruction</tt>s and basic blocks, which is why these are implemented with
1044<tt>ilist</tt>s.</p>
1045
1046Related classes of interest are explained in the following subsections:
1047 <ul>
1048 <li><a href="#dss_ilist_traits">ilist_traits</a></li>
1049 <li><a href="#dss_iplist">iplist</a></li>
1050 <li><a href="#dss_ilist_node">llvm/ADT/ilist_node.h</a></li>
1051 <li><a href="#dss_ilist_sentinel">Sentinels</a></li>
1052 </ul>
1053</div>
1054
1055<!-- _______________________________________________________________________ -->
1056<div class="doc_subsubsection">
1057 <a name="dss_ilist_traits">ilist_traits</a>
1058</div>
1059
1060<div class="doc_text">
1061<p><tt>ilist_traits&lt;T&gt;</tt> is <tt>ilist&lt;T&gt;</tt>'s customization
1062mechanism. <tt>iplist&lt;T&gt;</tt> (and consequently <tt>ilist&lt;T&gt;</tt>)
1063publicly derive from this traits class.</p>
1064</div>
1065
1066<!-- _______________________________________________________________________ -->
1067<div class="doc_subsubsection">
1068 <a name="dss_iplist">iplist</a>
1069</div>
1070
1071<div class="doc_text">
1072<p><tt>iplist&lt;T&gt;</tt> is <tt>ilist&lt;T&gt;</tt>'s base and as such
1073supports a slightly narrower interface. Notably, inserters from
1074<tt>T&amp;</tt> are absent.</p>
1075
1076<p><tt>ilist_traits&lt;T&gt;</tt> is a public base of this class and can be
1077used for a wide variety of customizations.</p>
1078</div>
1079
1080<!-- _______________________________________________________________________ -->
1081<div class="doc_subsubsection">
1082 <a name="dss_ilist_node">llvm/ADT/ilist_node.h</a>
1083</div>
1084
1085<div class="doc_text">
1086<p><tt>ilist_node&lt;T&gt;</tt> implements a the forward and backward links
1087that are expected by the <tt>ilist&lt;T&gt;</tt> (and analogous containers)
1088in the default manner.</p>
1089
1090<p><tt>ilist_node&lt;T&gt;</tt>s are meant to be embedded in the node type
1091<tt>T</tt>, usually <tt>T</tt> publicly derives from
1092<tt>ilist_node&lt;T&gt;</tt>.</p>
1093</div>
1094
1095<!-- _______________________________________________________________________ -->
1096<div class="doc_subsubsection">
1097 <a name="dss_ilist_sentinel">Sentinels</a>
1098</div>
1099
1100<div class="doc_text">
1101<p><tt>ilist</tt>s have another speciality that must be considered. To be a good
1102citizen in the C++ ecosystem, it needs to support the standard container
1103operations, such as <tt>begin</tt> and <tt>end</tt> iterators, etc. Also, the
1104<tt>operator--</tt> must work correctly on the <tt>end</tt> iterator in the
1105case of non-empty <tt>ilist</tt>s.</p>
1106
1107<p>The only sensible solution to this problem is to allocate a so-called
1108<i>sentinel</i> along with the intrusive list, which serves as the <tt>end</tt>
1109iterator, providing the back-link to the last element. However conforming to the
1110C++ convention it is illegal to <tt>operator++</tt> beyond the sentinel and it
1111also must not be dereferenced.</p>
1112
1113<p>These constraints allow for some implementation freedom to the <tt>ilist</tt>
1114how to allocate and store the sentinel. The corresponding policy is dictated
1115by <tt>ilist_traits&lt;T&gt;</tt>. By default a <tt>T</tt> gets heap-allocated
1116whenever the need for a sentinel arises.</p>
1117
1118<p>While the default policy is sufficient in most cases, it may break down when
1119<tt>T</tt> does not provide a default constructor. Also, in the case of many
1120instances of <tt>ilist</tt>s, the memory overhead of the associated sentinels
1121is wasted. To alleviate the situation with numerous and voluminous
1122<tt>T</tt>-sentinels, sometimes a trick is employed, leading to <i>ghostly
1123sentinels</i>.</p>
1124
1125<p>Ghostly sentinels are obtained by specially-crafted <tt>ilist_traits&lt;T&gt;</tt>
1126which superpose the sentinel with the <tt>ilist</tt> instance in memory. Pointer
1127arithmetic is used to obtain the sentinel, which is relative to the
1128<tt>ilist</tt>'s <tt>this</tt> pointer. The <tt>ilist</tt> is augmented by an
1129extra pointer, which serves as the back-link of the sentinel. This is the only
1130field in the ghostly sentinel which can be legally accessed.</p>
1131</div>
1132
1133<!-- _______________________________________________________________________ -->
1134<div class="doc_subsubsection">
1135 <a name="dss_other">Other Sequential Container options</a>
1136</div>
1137
1138<div class="doc_text">
1139<p>Other STL containers are available, such as std::string.</p>
1140
1141<p>There are also various STL adapter classes such as std::queue,
1142std::priority_queue, std::stack, etc. These provide simplified access to an
1143underlying container but don't affect the cost of the container itself.</p>
1144
1145</div>
1146
1147
1148<!-- ======================================================================= -->
1149<div class="doc_subsection">
1150 <a name="ds_set">Set-Like Containers (std::set, SmallSet, SetVector, etc)</a>
1151</div>
1152
1153<div class="doc_text">
1154
1155<p>Set-like containers are useful when you need to canonicalize multiple values
1156into a single representation. There are several different choices for how to do
1157this, providing various trade-offs.</p>
1158
1159</div>
1160
1161
1162<!-- _______________________________________________________________________ -->
1163<div class="doc_subsubsection">
1164 <a name="dss_sortedvectorset">A sorted 'vector'</a>
1165</div>
1166
1167<div class="doc_text">
1168
1169<p>If you intend to insert a lot of elements, then do a lot of queries, a
1170great approach is to use a vector (or other sequential container) with
1171std::sort+std::unique to remove duplicates. This approach works really well if
1172your usage pattern has these two distinct phases (insert then query), and can be
1173coupled with a good choice of <a href="#ds_sequential">sequential container</a>.
1174</p>
1175
1176<p>
1177This combination provides the several nice properties: the result data is
1178contiguous in memory (good for cache locality), has few allocations, is easy to
1179address (iterators in the final vector are just indices or pointers), and can be
1180efficiently queried with a standard binary or radix search.</p>
1181
1182</div>
1183
1184<!-- _______________________________________________________________________ -->
1185<div class="doc_subsubsection">
1186 <a name="dss_smallset">"llvm/ADT/SmallSet.h"</a>
1187</div>
1188
1189<div class="doc_text">
1190
1191<p>If you have a set-like data structure that is usually small and whose elements
1192are reasonably small, a <tt>SmallSet&lt;Type, N&gt;</tt> is a good choice. This set
1193has space for N elements in place (thus, if the set is dynamically smaller than
1194N, no malloc traffic is required) and accesses them with a simple linear search.
1195When the set grows beyond 'N' elements, it allocates a more expensive representation that
1196guarantees efficient access (for most types, it falls back to std::set, but for
1197pointers it uses something far better, <a
1198href="#dss_smallptrset">SmallPtrSet</a>).</p>
1199
1200<p>The magic of this class is that it handles small sets extremely efficiently,
1201but gracefully handles extremely large sets without loss of efficiency. The
1202drawback is that the interface is quite small: it supports insertion, queries
1203and erasing, but does not support iteration.</p>
1204
1205</div>
1206
1207<!-- _______________________________________________________________________ -->
1208<div class="doc_subsubsection">
1209 <a name="dss_smallptrset">"llvm/ADT/SmallPtrSet.h"</a>
1210</div>
1211
1212<div class="doc_text">
1213
1214<p>SmallPtrSet has all the advantages of SmallSet (and a SmallSet of pointers is
1215transparently implemented with a SmallPtrSet), but also supports iterators. If
1216more than 'N' insertions are performed, a single quadratically
1217probed hash table is allocated and grows as needed, providing extremely
1218efficient access (constant time insertion/deleting/queries with low constant
1219factors) and is very stingy with malloc traffic.</p>
1220
1221<p>Note that, unlike std::set, the iterators of SmallPtrSet are invalidated
1222whenever an insertion occurs. Also, the values visited by the iterators are not
1223visited in sorted order.</p>
1224
1225</div>
1226
1227<!-- _______________________________________________________________________ -->
1228<div class="doc_subsubsection">
1229 <a name="dss_denseset">"llvm/ADT/DenseSet.h"</a>
1230</div>
1231
1232<div class="doc_text">
1233
1234<p>
1235DenseSet is a simple quadratically probed hash table. It excels at supporting
1236small values: it uses a single allocation to hold all of the pairs that
1237are currently inserted in the set. DenseSet is a great way to unique small
1238values that are not simple pointers (use <a
1239href="#dss_smallptrset">SmallPtrSet</a> for pointers). Note that DenseSet has
1240the same requirements for the value type that <a
1241href="#dss_densemap">DenseMap</a> has.
1242</p>
1243
1244</div>
1245
1246<!-- _______________________________________________________________________ -->
1247<div class="doc_subsubsection">
1248 <a name="dss_FoldingSet">"llvm/ADT/FoldingSet.h"</a>
1249</div>
1250
1251<div class="doc_text">
1252
1253<p>
1254FoldingSet is an aggregate class that is really good at uniquing
1255expensive-to-create or polymorphic objects. It is a combination of a chained
1256hash table with intrusive links (uniqued objects are required to inherit from
1257FoldingSetNode) that uses <a href="#dss_smallvector">SmallVector</a> as part of
1258its ID process.</p>
1259
1260<p>Consider a case where you want to implement a "getOrCreateFoo" method for
1261a complex object (for example, a node in the code generator). The client has a
1262description of *what* it wants to generate (it knows the opcode and all the
1263operands), but we don't want to 'new' a node, then try inserting it into a set
1264only to find out it already exists, at which point we would have to delete it
1265and return the node that already exists.
1266</p>
1267
1268<p>To support this style of client, FoldingSet perform a query with a
1269FoldingSetNodeID (which wraps SmallVector) that can be used to describe the
1270element that we want to query for. The query either returns the element
1271matching the ID or it returns an opaque ID that indicates where insertion should
1272take place. Construction of the ID usually does not require heap traffic.</p>
1273
1274<p>Because FoldingSet uses intrusive links, it can support polymorphic objects
1275in the set (for example, you can have SDNode instances mixed with LoadSDNodes).
1276Because the elements are individually allocated, pointers to the elements are
1277stable: inserting or removing elements does not invalidate any pointers to other
1278elements.
1279</p>
1280
1281</div>
1282
1283<!-- _______________________________________________________________________ -->
1284<div class="doc_subsubsection">
1285 <a name="dss_set">&lt;set&gt;</a>
1286</div>
1287
1288<div class="doc_text">
1289
1290<p><tt>std::set</tt> is a reasonable all-around set class, which is decent at
1291many things but great at nothing. std::set allocates memory for each element
1292inserted (thus it is very malloc intensive) and typically stores three pointers
1293per element in the set (thus adding a large amount of per-element space
1294overhead). It offers guaranteed log(n) performance, which is not particularly
1295fast from a complexity standpoint (particularly if the elements of the set are
1296expensive to compare, like strings), and has extremely high constant factors for
1297lookup, insertion and removal.</p>
1298
1299<p>The advantages of std::set are that its iterators are stable (deleting or
1300inserting an element from the set does not affect iterators or pointers to other
1301elements) and that iteration over the set is guaranteed to be in sorted order.
1302If the elements in the set are large, then the relative overhead of the pointers
1303and malloc traffic is not a big deal, but if the elements of the set are small,
1304std::set is almost never a good choice.</p>
1305
1306</div>
1307
1308<!-- _______________________________________________________________________ -->
1309<div class="doc_subsubsection">
1310 <a name="dss_setvector">"llvm/ADT/SetVector.h"</a>
1311</div>
1312
1313<div class="doc_text">
1314<p>LLVM's SetVector&lt;Type&gt; is an adapter class that combines your choice of
1315a set-like container along with a <a href="#ds_sequential">Sequential
1316Container</a>. The important property
1317that this provides is efficient insertion with uniquing (duplicate elements are
1318ignored) with iteration support. It implements this by inserting elements into
1319both a set-like container and the sequential container, using the set-like
1320container for uniquing and the sequential container for iteration.
1321</p>
1322
1323<p>The difference between SetVector and other sets is that the order of
1324iteration is guaranteed to match the order of insertion into the SetVector.
1325This property is really important for things like sets of pointers. Because
1326pointer values are non-deterministic (e.g. vary across runs of the program on
1327different machines), iterating over the pointers in the set will
1328not be in a well-defined order.</p>
1329
1330<p>
1331The drawback of SetVector is that it requires twice as much space as a normal
1332set and has the sum of constant factors from the set-like container and the
1333sequential container that it uses. Use it *only* if you need to iterate over
1334the elements in a deterministic order. SetVector is also expensive to delete
1335elements out of (linear time), unless you use it's "pop_back" method, which is
1336faster.
1337</p>
1338
1339<p>SetVector is an adapter class that defaults to using std::vector and std::set
1340for the underlying containers, so it is quite expensive. However,
1341<tt>"llvm/ADT/SetVector.h"</tt> also provides a SmallSetVector class, which
1342defaults to using a SmallVector and SmallSet of a specified size. If you use
1343this, and if your sets are dynamically smaller than N, you will save a lot of
1344heap traffic.</p>
1345
1346</div>
1347
1348<!-- _______________________________________________________________________ -->
1349<div class="doc_subsubsection">
1350 <a name="dss_uniquevector">"llvm/ADT/UniqueVector.h"</a>
1351</div>
1352
1353<div class="doc_text">
1354
1355<p>
1356UniqueVector is similar to <a href="#dss_setvector">SetVector</a>, but it
1357retains a unique ID for each element inserted into the set. It internally
1358contains a map and a vector, and it assigns a unique ID for each value inserted
1359into the set.</p>
1360
1361<p>UniqueVector is very expensive: its cost is the sum of the cost of
1362maintaining both the map and vector, it has high complexity, high constant
1363factors, and produces a lot of malloc traffic. It should be avoided.</p>
1364
1365</div>
1366
1367
1368<!-- _______________________________________________________________________ -->
1369<div class="doc_subsubsection">
1370 <a name="dss_otherset">Other Set-Like Container Options</a>
1371</div>
1372
1373<div class="doc_text">
1374
1375<p>
1376The STL provides several other options, such as std::multiset and the various
1377"hash_set" like containers (whether from C++ TR1 or from the SGI library). We
1378never use hash_set and unordered_set because they are generally very expensive
1379(each insertion requires a malloc) and very non-portable.
1380</p>
1381
1382<p>std::multiset is useful if you're not interested in elimination of
1383duplicates, but has all the drawbacks of std::set. A sorted vector (where you
1384don't delete duplicate entries) or some other approach is almost always
1385better.</p>
1386
1387</div>
1388
1389<!-- ======================================================================= -->
1390<div class="doc_subsection">
1391 <a name="ds_map">Map-Like Containers (std::map, DenseMap, etc)</a>
1392</div>
1393
1394<div class="doc_text">
1395Map-like containers are useful when you want to associate data to a key. As
1396usual, there are a lot of different ways to do this. :)
1397</div>
1398
1399<!-- _______________________________________________________________________ -->
1400<div class="doc_subsubsection">
1401 <a name="dss_sortedvectormap">A sorted 'vector'</a>
1402</div>
1403
1404<div class="doc_text">
1405
1406<p>
1407If your usage pattern follows a strict insert-then-query approach, you can
1408trivially use the same approach as <a href="#dss_sortedvectorset">sorted vectors
1409for set-like containers</a>. The only difference is that your query function
1410(which uses std::lower_bound to get efficient log(n) lookup) should only compare
1411the key, not both the key and value. This yields the same advantages as sorted
1412vectors for sets.
1413</p>
1414</div>
1415
1416<!-- _______________________________________________________________________ -->
1417<div class="doc_subsubsection">
1418 <a name="dss_stringmap">"llvm/ADT/StringMap.h"</a>
1419</div>
1420
1421<div class="doc_text">
1422
1423<p>
1424Strings are commonly used as keys in maps, and they are difficult to support
1425efficiently: they are variable length, inefficient to hash and compare when
1426long, expensive to copy, etc. StringMap is a specialized container designed to
1427cope with these issues. It supports mapping an arbitrary range of bytes to an
1428arbitrary other object.</p>
1429
1430<p>The StringMap implementation uses a quadratically-probed hash table, where
1431the buckets store a pointer to the heap allocated entries (and some other
1432stuff). The entries in the map must be heap allocated because the strings are
1433variable length. The string data (key) and the element object (value) are
1434stored in the same allocation with the string data immediately after the element
1435object. This container guarantees the "<tt>(char*)(&amp;Value+1)</tt>" points
1436to the key string for a value.</p>
1437
1438<p>The StringMap is very fast for several reasons: quadratic probing is very
1439cache efficient for lookups, the hash value of strings in buckets is not
1440recomputed when lookup up an element, StringMap rarely has to touch the
1441memory for unrelated objects when looking up a value (even when hash collisions
1442happen), hash table growth does not recompute the hash values for strings
1443already in the table, and each pair in the map is store in a single allocation
1444(the string data is stored in the same allocation as the Value of a pair).</p>
1445
1446<p>StringMap also provides query methods that take byte ranges, so it only ever
1447copies a string if a value is inserted into the table.</p>
1448</div>
1449
1450<!-- _______________________________________________________________________ -->
1451<div class="doc_subsubsection">
1452 <a name="dss_indexedmap">"llvm/ADT/IndexedMap.h"</a>
1453</div>
1454
1455<div class="doc_text">
1456<p>
1457IndexedMap is a specialized container for mapping small dense integers (or
1458values that can be mapped to small dense integers) to some other type. It is
1459internally implemented as a vector with a mapping function that maps the keys to
1460the dense integer range.
1461</p>
1462
1463<p>
1464This is useful for cases like virtual registers in the LLVM code generator: they
1465have a dense mapping that is offset by a compile-time constant (the first
1466virtual register ID).</p>
1467
1468</div>
1469
1470<!-- _______________________________________________________________________ -->
1471<div class="doc_subsubsection">
1472 <a name="dss_densemap">"llvm/ADT/DenseMap.h"</a>
1473</div>
1474
1475<div class="doc_text">
1476
1477<p>
1478DenseMap is a simple quadratically probed hash table. It excels at supporting
1479small keys and values: it uses a single allocation to hold all of the pairs that
1480are currently inserted in the map. DenseMap is a great way to map pointers to
1481pointers, or map other small types to each other.
1482</p>
1483
1484<p>
1485There are several aspects of DenseMap that you should be aware of, however. The
1486iterators in a densemap are invalidated whenever an insertion occurs, unlike
1487map. Also, because DenseMap allocates space for a large number of key/value
1488pairs (it starts with 64 by default), it will waste a lot of space if your keys
1489or values are large. Finally, you must implement a partial specialization of
1490DenseMapInfo for the key that you want, if it isn't already supported. This
1491is required to tell DenseMap about two special marker values (which can never be
1492inserted into the map) that it needs internally.</p>
1493
1494</div>
1495
1496<!-- _______________________________________________________________________ -->
1497<div class="doc_subsubsection">
1498 <a name="dss_valuemap">"llvm/ADT/ValueMap.h"</a>
1499</div>
1500
1501<div class="doc_text">
1502
1503<p>
1504ValueMap is a wrapper around a <a href="#dss_densemap">DenseMap</a> mapping
1505Value*s (or subclasses) to another type. When a Value is deleted or RAUW'ed,
1506ValueMap will update itself so the new version of the key is mapped to the same
1507value, just as if the key were a WeakVH. You can configure exactly how this
1508happens, and what else happens on these two events, by passing
1509a <code>Config</code> parameter to the ValueMap template.</p>
1510
1511</div>
1512
1513<!-- _______________________________________________________________________ -->
1514<div class="doc_subsubsection">
1515 <a name="dss_map">&lt;map&gt;</a>
1516</div>
1517
1518<div class="doc_text">
1519
1520<p>
1521std::map has similar characteristics to <a href="#dss_set">std::set</a>: it uses
1522a single allocation per pair inserted into the map, it offers log(n) lookup with
1523an extremely large constant factor, imposes a space penalty of 3 pointers per
1524pair in the map, etc.</p>
1525
1526<p>std::map is most useful when your keys or values are very large, if you need
1527to iterate over the collection in sorted order, or if you need stable iterators
1528into the map (i.e. they don't get invalidated if an insertion or deletion of
1529another element takes place).</p>
1530
1531</div>
1532
1533<!-- _______________________________________________________________________ -->
1534<div class="doc_subsubsection">
1535 <a name="dss_othermap">Other Map-Like Container Options</a>
1536</div>
1537
1538<div class="doc_text">
1539
1540<p>
1541The STL provides several other options, such as std::multimap and the various
1542"hash_map" like containers (whether from C++ TR1 or from the SGI library). We
1543never use hash_set and unordered_set because they are generally very expensive
1544(each insertion requires a malloc) and very non-portable.</p>
1545
1546<p>std::multimap is useful if you want to map a key to multiple values, but has
1547all the drawbacks of std::map. A sorted vector or some other approach is almost
1548always better.</p>
1549
1550</div>
1551
1552<!-- ======================================================================= -->
1553<div class="doc_subsection">
1554 <a name="ds_string">String-like containers</a>
1555</div>
1556
1557<div class="doc_text">
1558
1559<p>
1560TODO: const char* vs stringref vs smallstring vs std::string. Describe twine,
1561xref to #string_apis.
1562</p>
1563
1564</div>
1565
1566<!-- ======================================================================= -->
1567<div class="doc_subsection">
1568 <a name="ds_bit">Bit storage containers (BitVector, SparseBitVector)</a>
1569</div>
1570
1571<div class="doc_text">
1572<p>Unlike the other containers, there are only two bit storage containers, and
1573choosing when to use each is relatively straightforward.</p>
1574
1575<p>One additional option is
1576<tt>std::vector&lt;bool&gt;</tt>: we discourage its use for two reasons 1) the
1577implementation in many common compilers (e.g. commonly available versions of
1578GCC) is extremely inefficient and 2) the C++ standards committee is likely to
1579deprecate this container and/or change it significantly somehow. In any case,
1580please don't use it.</p>
1581</div>
1582
1583<!-- _______________________________________________________________________ -->
1584<div class="doc_subsubsection">
1585 <a name="dss_bitvector">BitVector</a>
1586</div>
1587
1588<div class="doc_text">
1589<p> The BitVector container provides a dynamic size set of bits for manipulation.
1590It supports individual bit setting/testing, as well as set operations. The set
1591operations take time O(size of bitvector), but operations are performed one word
1592at a time, instead of one bit at a time. This makes the BitVector very fast for
1593set operations compared to other containers. Use the BitVector when you expect
1594the number of set bits to be high (IE a dense set).
1595</p>
1596</div>
1597
1598<!-- _______________________________________________________________________ -->
1599<div class="doc_subsubsection">
1600 <a name="dss_smallbitvector">SmallBitVector</a>
1601</div>
1602
1603<div class="doc_text">
1604<p> The SmallBitVector container provides the same interface as BitVector, but
1605it is optimized for the case where only a small number of bits, less than
160625 or so, are needed. It also transparently supports larger bit counts, but
1607slightly less efficiently than a plain BitVector, so SmallBitVector should
1608only be used when larger counts are rare.
1609</p>
1610
1611<p>
1612At this time, SmallBitVector does not support set operations (and, or, xor),
1613and its operator[] does not provide an assignable lvalue.
1614</p>
1615</div>
1616
1617<!-- _______________________________________________________________________ -->
1618<div class="doc_subsubsection">
1619 <a name="dss_sparsebitvector">SparseBitVector</a>
1620</div>
1621
1622<div class="doc_text">
1623<p> The SparseBitVector container is much like BitVector, with one major
1624difference: Only the bits that are set, are stored. This makes the
1625SparseBitVector much more space efficient than BitVector when the set is sparse,
1626as well as making set operations O(number of set bits) instead of O(size of
1627universe). 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
1628(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).
1629</p>
1630</div>
1631
1632<!-- *********************************************************************** -->
1633<div class="doc_section">
1634 <a name="common">Helpful Hints for Common Operations</a>
1635</div>
1636<!-- *********************************************************************** -->
1637
1638<div class="doc_text">
1639
1640<p>This section describes how to perform some very simple transformations of
1641LLVM code. This is meant to give examples of common idioms used, showing the
1642practical side of LLVM transformations. <p> Because this is a "how-to" section,
1643you should also read about the main classes that you will be working with. The
1644<a href="#coreclasses">Core LLVM Class Hierarchy Reference</a> contains details
1645and descriptions of the main classes that you should know about.</p>
1646
1647</div>
1648
1649<!-- NOTE: this section should be heavy on example code -->
1650<!-- ======================================================================= -->
1651<div class="doc_subsection">
1652 <a name="inspection">Basic Inspection and Traversal Routines</a>
1653</div>
1654
1655<div class="doc_text">
1656
1657<p>The LLVM compiler infrastructure have many different data structures that may
1658be traversed. Following the example of the C++ standard template library, the
1659techniques used to traverse these various data structures are all basically the
1660same. For a enumerable sequence of values, the <tt>XXXbegin()</tt> function (or
1661method) returns an iterator to the start of the sequence, the <tt>XXXend()</tt>
1662function returns an iterator pointing to one past the last valid element of the
1663sequence, and there is some <tt>XXXiterator</tt> data type that is common
1664between the two operations.</p>
1665
1666<p>Because the pattern for iteration is common across many different aspects of
1667the program representation, the standard template library algorithms may be used
1668on them, and it is easier to remember how to iterate. First we show a few common
1669examples of the data structures that need to be traversed. Other data
1670structures are traversed in very similar ways.</p>
1671
1672</div>
1673
1674<!-- _______________________________________________________________________ -->
1675<div class="doc_subsubsection">
1676 <a name="iterate_function">Iterating over the </a><a
1677 href="#BasicBlock"><tt>BasicBlock</tt></a>s in a <a
1678 href="#Function"><tt>Function</tt></a>
1679</div>
1680
1681<div class="doc_text">
1682
1683<p>It's quite common to have a <tt>Function</tt> instance that you'd like to
1684transform in some way; in particular, you'd like to manipulate its
1685<tt>BasicBlock</tt>s. To facilitate this, you'll need to iterate over all of
1686the <tt>BasicBlock</tt>s that constitute the <tt>Function</tt>. The following is
1687an example that prints the name of a <tt>BasicBlock</tt> and the number of
1688<tt>Instruction</tt>s it contains:</p>
1689
1690<div class="doc_code">
1691<pre>
1692// <i>func is a pointer to a Function instance</i>
1693for (Function::iterator i = func-&gt;begin(), e = func-&gt;end(); i != e; ++i)
1694 // <i>Print out the name of the basic block if it has one, and then the</i>
1695 // <i>number of instructions that it contains</i>
1696 errs() &lt;&lt; "Basic block (name=" &lt;&lt; i-&gt;getName() &lt;&lt; ") has "
1697 &lt;&lt; i-&gt;size() &lt;&lt; " instructions.\n";
1698</pre>
1699</div>
1700
1701<p>Note that i can be used as if it were a pointer for the purposes of
1702invoking member functions of the <tt>Instruction</tt> class. This is
1703because the indirection operator is overloaded for the iterator
1704classes. In the above code, the expression <tt>i-&gt;size()</tt> is
1705exactly equivalent to <tt>(*i).size()</tt> just like you'd expect.</p>
1706
1707</div>
1708
1709<!-- _______________________________________________________________________ -->
1710<div class="doc_subsubsection">
1711 <a name="iterate_basicblock">Iterating over the </a><a
1712 href="#Instruction"><tt>Instruction</tt></a>s in a <a
1713 href="#BasicBlock"><tt>BasicBlock</tt></a>
1714</div>
1715
1716<div class="doc_text">
1717
1718<p>Just like when dealing with <tt>BasicBlock</tt>s in <tt>Function</tt>s, it's
1719easy to iterate over the individual instructions that make up
1720<tt>BasicBlock</tt>s. Here's a code snippet that prints out each instruction in
1721a <tt>BasicBlock</tt>:</p>
1722
1723<div class="doc_code">
1724<pre>
1725// <i>blk is a pointer to a BasicBlock instance</i>
1726for (BasicBlock::iterator i = blk-&gt;begin(), e = blk-&gt;end(); i != e; ++i)
1727 // <i>The next statement works since operator&lt;&lt;(ostream&amp;,...)</i>
1728 // <i>is overloaded for Instruction&amp;</i>
1729 errs() &lt;&lt; *i &lt;&lt; "\n";
1730</pre>
1731</div>
1732
1733<p>However, this isn't really the best way to print out the contents of a
1734<tt>BasicBlock</tt>! Since the ostream operators are overloaded for virtually
1735anything you'll care about, you could have just invoked the print routine on the
1736basic block itself: <tt>errs() &lt;&lt; *blk &lt;&lt; "\n";</tt>.</p>
1737
1738</div>
1739
1740<!-- _______________________________________________________________________ -->
1741<div class="doc_subsubsection">
1742 <a name="iterate_institer">Iterating over the </a><a
1743 href="#Instruction"><tt>Instruction</tt></a>s in a <a
1744 href="#Function"><tt>Function</tt></a>
1745</div>
1746
1747<div class="doc_text">
1748
1749<p>If you're finding that you commonly iterate over a <tt>Function</tt>'s
1750<tt>BasicBlock</tt>s and then that <tt>BasicBlock</tt>'s <tt>Instruction</tt>s,
1751<tt>InstIterator</tt> should be used instead. You'll need to include <a
1752href="/doxygen/InstIterator_8h-source.html"><tt>llvm/Support/InstIterator.h</tt></a>,
1753and then instantiate <tt>InstIterator</tt>s explicitly in your code. Here's a
1754small example that shows how to dump all instructions in a function to the standard error stream:<p>
1755
1756<div class="doc_code">
1757<pre>
1758#include "<a href="/doxygen/InstIterator_8h-source.html">llvm/Support/InstIterator.h</a>"
1759
1760// <i>F is a pointer to a Function instance</i>
1761for (inst_iterator I = inst_begin(F), E = inst_end(F); I != E; ++I)
1762 errs() &lt;&lt; *I &lt;&lt; "\n";
1763</pre>
1764</div>
1765
1766<p>Easy, isn't it? You can also use <tt>InstIterator</tt>s to fill a
1767work list with its initial contents. For example, if you wanted to
1768initialize a work list to contain all instructions in a <tt>Function</tt>
1769F, all you would need to do is something like:</p>
1770
1771<div class="doc_code">
1772<pre>
1773std::set&lt;Instruction*&gt; worklist;
1774// or better yet, SmallPtrSet&lt;Instruction*, 64&gt; worklist;
1775
1776for (inst_iterator I = inst_begin(F), E = inst_end(F); I != E; ++I)
1777 worklist.insert(&amp;*I);
1778</pre>
1779</div>
1780
1781<p>The STL set <tt>worklist</tt> would now contain all instructions in the
1782<tt>Function</tt> pointed to by F.</p>
1783
1784</div>
1785
1786<!-- _______________________________________________________________________ -->
1787<div class="doc_subsubsection">
1788 <a name="iterate_convert">Turning an iterator into a class pointer (and
1789 vice-versa)</a>
1790</div>
1791
1792<div class="doc_text">
1793
1794<p>Sometimes, it'll be useful to grab a reference (or pointer) to a class
1795instance when all you've got at hand is an iterator. Well, extracting
1796a reference or a pointer from an iterator is very straight-forward.
1797Assuming that <tt>i</tt> is a <tt>BasicBlock::iterator</tt> and <tt>j</tt>
1798is a <tt>BasicBlock::const_iterator</tt>:</p>
1799
1800<div class="doc_code">
1801<pre>
1802Instruction&amp; inst = *i; // <i>Grab reference to instruction reference</i>
1803Instruction* pinst = &amp;*i; // <i>Grab pointer to instruction reference</i>
1804const Instruction&amp; inst = *j;
1805</pre>
1806</div>
1807
1808<p>However, the iterators you'll be working with in the LLVM framework are
1809special: they will automatically convert to a ptr-to-instance type whenever they
1810need to. Instead of dereferencing the iterator and then taking the address of
1811the result, you can simply assign the iterator to the proper pointer type and
1812you get the dereference and address-of operation as a result of the assignment
1813(behind the scenes, this is a result of overloading casting mechanisms). Thus
1814the last line of the last example,</p>
1815
1816<div class="doc_code">
1817<pre>
1818Instruction *pinst = &amp;*i;
1819</pre>
1820</div>
1821
1822<p>is semantically equivalent to</p>
1823
1824<div class="doc_code">
1825<pre>
1826Instruction *pinst = i;
1827</pre>
1828</div>
1829
1830<p>It's also possible to turn a class pointer into the corresponding iterator,
1831and this is a constant time operation (very efficient). The following code
1832snippet illustrates use of the conversion constructors provided by LLVM
1833iterators. By using these, you can explicitly grab the iterator of something
1834without actually obtaining it via iteration over some structure:</p>
1835
1836<div class="doc_code">
1837<pre>
1838void printNextInstruction(Instruction* inst) {
1839 BasicBlock::iterator it(inst);
1840 ++it; // <i>After this line, it refers to the instruction after *inst</i>
1841 if (it != inst-&gt;getParent()-&gt;end()) errs() &lt;&lt; *it &lt;&lt; "\n";
1842}
1843</pre>
1844</div>
1845
1846</div>
1847
1848<!--_______________________________________________________________________-->
1849<div class="doc_subsubsection">
1850 <a name="iterate_complex">Finding call sites: a slightly more complex
1851 example</a>
1852</div>
1853
1854<div class="doc_text">
1855
1856<p>Say that you're writing a FunctionPass and would like to count all the
1857locations in the entire module (that is, across every <tt>Function</tt>) where a
1858certain function (i.e., some <tt>Function</tt>*) is already in scope. As you'll
1859learn later, you may want to use an <tt>InstVisitor</tt> to accomplish this in a
1860much more straight-forward manner, but this example will allow us to explore how
1861you'd do it if you didn't have <tt>InstVisitor</tt> around. In pseudo-code, this
1862is what we want to do:</p>
1863
1864<div class="doc_code">
1865<pre>
1866initialize callCounter to zero
1867for each Function f in the Module
1868 for each BasicBlock b in f
1869 for each Instruction i in b
1870 if (i is a CallInst and calls the given function)
1871 increment callCounter
1872</pre>
1873</div>
1874
1875<p>And the actual code is (remember, because we're writing a
1876<tt>FunctionPass</tt>, our <tt>FunctionPass</tt>-derived class simply has to
1877override the <tt>runOnFunction</tt> method):</p>
1878
1879<div class="doc_code">
1880<pre>
1881Function* targetFunc = ...;
1882
1883class OurFunctionPass : public FunctionPass {
1884 public:
1885 OurFunctionPass(): callCounter(0) { }
1886
1887 virtual runOnFunction(Function&amp; F) {
1888 for (Function::iterator b = F.begin(), be = F.end(); b != be; ++b) {
1889 for (BasicBlock::iterator i = b-&gt;begin(), ie = b-&gt;end(); i != ie; ++i) {
1890 if (<a href="#CallInst">CallInst</a>* callInst = <a href="#isa">dyn_cast</a>&lt;<a
1891 href="#CallInst">CallInst</a>&gt;(&amp;*i)) {
1892 // <i>We know we've encountered a call instruction, so we</i>
1893 // <i>need to determine if it's a call to the</i>
1894 // <i>function pointed to by m_func or not.</i>
1895 if (callInst-&gt;getCalledFunction() == targetFunc)
1896 ++callCounter;
1897 }
1898 }
1899 }
1900 }
1901
1902 private:
1903 unsigned callCounter;
1904};
1905</pre>
1906</div>
1907
1908</div>
1909
1910<!--_______________________________________________________________________-->
1911<div class="doc_subsubsection">
1912 <a name="calls_and_invokes">Treating calls and invokes the same way</a>
1913</div>
1914
1915<div class="doc_text">
1916
1917<p>You may have noticed that the previous example was a bit oversimplified in
1918that it did not deal with call sites generated by 'invoke' instructions. In
1919this, and in other situations, you may find that you want to treat
1920<tt>CallInst</tt>s and <tt>InvokeInst</tt>s the same way, even though their
1921most-specific common base class is <tt>Instruction</tt>, which includes lots of
1922less closely-related things. For these cases, LLVM provides a handy wrapper
1923class called <a
1924href="http://llvm.org/doxygen/classllvm_1_1CallSite.html"><tt>CallSite</tt></a>.
1925It is essentially a wrapper around an <tt>Instruction</tt> pointer, with some
1926methods that provide functionality common to <tt>CallInst</tt>s and
1927<tt>InvokeInst</tt>s.</p>
1928
1929<p>This class has "value semantics": it should be passed by value, not by
1930reference and it should not be dynamically allocated or deallocated using
1931<tt>operator new</tt> or <tt>operator delete</tt>. It is efficiently copyable,
1932assignable and constructable, with costs equivalents to that of a bare pointer.
1933If you look at its definition, it has only a single pointer member.</p>
1934
1935</div>
1936
1937<!--_______________________________________________________________________-->
1938<div class="doc_subsubsection">
1939 <a name="iterate_chains">Iterating over def-use &amp; use-def chains</a>
1940</div>
1941
1942<div class="doc_text">
1943
1944<p>Frequently, we might have an instance of the <a
1945href="/doxygen/classllvm_1_1Value.html">Value Class</a> and we want to
1946determine which <tt>User</tt>s use the <tt>Value</tt>. The list of all
1947<tt>User</tt>s of a particular <tt>Value</tt> is called a <i>def-use</i> chain.
1948For example, let's say we have a <tt>Function*</tt> named <tt>F</tt> to a
1949particular function <tt>foo</tt>. Finding all of the instructions that
1950<i>use</i> <tt>foo</tt> is as simple as iterating over the <i>def-use</i> chain
1951of <tt>F</tt>:</p>
1952
1953<div class="doc_code">
1954<pre>
1955Function *F = ...;
1956
1957for (Value::use_iterator i = F-&gt;use_begin(), e = F-&gt;use_end(); i != e; ++i)
1958 if (Instruction *Inst = dyn_cast&lt;Instruction&gt;(*i)) {
1959 errs() &lt;&lt; "F is used in instruction:\n";
1960 errs() &lt;&lt; *Inst &lt;&lt; "\n";
1961 }
1962</pre>
1963</div>
1964
1965<p>Alternately, it's common to have an instance of the <a
1966href="/doxygen/classllvm_1_1User.html">User Class</a> and need to know what
1967<tt>Value</tt>s are used by it. The list of all <tt>Value</tt>s used by a
1968<tt>User</tt> is known as a <i>use-def</i> chain. Instances of class
1969<tt>Instruction</tt> are common <tt>User</tt>s, so we might want to iterate over
1970all of the values that a particular instruction uses (that is, the operands of
1971the particular <tt>Instruction</tt>):</p>
1972
1973<div class="doc_code">
1974<pre>
1975Instruction *pi = ...;
1976
1977for (User::op_iterator i = pi-&gt;op_begin(), e = pi-&gt;op_end(); i != e; ++i) {
1978 Value *v = *i;
1979 // <i>...</i>
1980}
1981</pre>
1982</div>
1983
1984<!--
1985 def-use chains ("finding all users of"): Value::use_begin/use_end
1986 use-def chains ("finding all values used"): User::op_begin/op_end [op=operand]
1987-->
1988
1989</div>
1990
1991<!--_______________________________________________________________________-->
1992<div class="doc_subsubsection">
1993 <a name="iterate_preds">Iterating over predecessors &amp;
1994successors of blocks</a>
1995</div>
1996
1997<div class="doc_text">
1998
1999<p>Iterating over the predecessors and successors of a block is quite easy
2000with the routines defined in <tt>"llvm/Support/CFG.h"</tt>. Just use code like
2001this to iterate over all predecessors of BB:</p>
2002
2003<div class="doc_code">
2004<pre>
2005#include "llvm/Support/CFG.h"
2006BasicBlock *BB = ...;
2007
2008for (pred_iterator PI = pred_begin(BB), E = pred_end(BB); PI != E; ++PI) {
2009 BasicBlock *Pred = *PI;
2010 // <i>...</i>
2011}
2012</pre>
2013</div>
2014
2015<p>Similarly, to iterate over successors use
2016succ_iterator/succ_begin/succ_end.</p>
2017
2018</div>
2019
2020
2021<!-- ======================================================================= -->
2022<div class="doc_subsection">
2023 <a name="simplechanges">Making simple changes</a>
2024</div>
2025
2026<div class="doc_text">
2027
2028<p>There are some primitive transformation operations present in the LLVM
2029infrastructure that are worth knowing about. When performing
2030transformations, it's fairly common to manipulate the contents of basic
2031blocks. This section describes some of the common methods for doing so
2032and gives example code.</p>
2033
2034</div>
2035
2036<!--_______________________________________________________________________-->
2037<div class="doc_subsubsection">
2038 <a name="schanges_creating">Creating and inserting new
2039 <tt>Instruction</tt>s</a>
2040</div>
2041
2042<div class="doc_text">
2043
2044<p><i>Instantiating Instructions</i></p>
2045
2046<p>Creation of <tt>Instruction</tt>s is straight-forward: simply call the
2047constructor for the kind of instruction to instantiate and provide the necessary
2048parameters. For example, an <tt>AllocaInst</tt> only <i>requires</i> a
2049(const-ptr-to) <tt>Type</tt>. Thus:</p>
2050
2051<div class="doc_code">
2052<pre>
2053AllocaInst* ai = new AllocaInst(Type::Int32Ty);
2054</pre>
2055</div>
2056
2057<p>will create an <tt>AllocaInst</tt> instance that represents the allocation of
2058one integer in the current stack frame, at run time. Each <tt>Instruction</tt>
2059subclass is likely to have varying default parameters which change the semantics
2060of the instruction, so refer to the <a
2061href="/doxygen/classllvm_1_1Instruction.html">doxygen documentation for the subclass of
2062Instruction</a> that you're interested in instantiating.</p>
2063
2064<p><i>Naming values</i></p>
2065
2066<p>It is very useful to name the values of instructions when you're able to, as
2067this facilitates the debugging of your transformations. If you end up looking
2068at generated LLVM machine code, you definitely want to have logical names
2069associated with the results of instructions! By supplying a value for the
2070<tt>Name</tt> (default) parameter of the <tt>Instruction</tt> constructor, you
2071associate a logical name with the result of the instruction's execution at
2072run time. For example, say that I'm writing a transformation that dynamically
2073allocates space for an integer on the stack, and that integer is going to be
2074used as some kind of index by some other code. To accomplish this, I place an
2075<tt>AllocaInst</tt> at the first point in the first <tt>BasicBlock</tt> of some
2076<tt>Function</tt>, and I'm intending to use it within the same
2077<tt>Function</tt>. I might do:</p>
2078
2079<div class="doc_code">
2080<pre>
2081AllocaInst* pa = new AllocaInst(Type::Int32Ty, 0, "indexLoc");
2082</pre>
2083</div>
2084
2085<p>where <tt>indexLoc</tt> is now the logical name of the instruction's
2086execution value, which is a pointer to an integer on the run time stack.</p>
2087
2088<p><i>Inserting instructions</i></p>
2089
2090<p>There are essentially two ways to insert an <tt>Instruction</tt>
2091into an existing sequence of instructions that form a <tt>BasicBlock</tt>:</p>
2092
2093<ul>
2094 <li>Insertion into an explicit instruction list
2095
2096 <p>Given a <tt>BasicBlock* pb</tt>, an <tt>Instruction* pi</tt> within that
2097 <tt>BasicBlock</tt>, and a newly-created instruction we wish to insert
2098 before <tt>*pi</tt>, we do the following: </p>
2099
2100<div class="doc_code">
2101<pre>
2102BasicBlock *pb = ...;
2103Instruction *pi = ...;
2104Instruction *newInst = new Instruction(...);
2105
2106pb-&gt;getInstList().insert(pi, newInst); // <i>Inserts newInst before pi in pb</i>
2107</pre>
2108</div>
2109
2110 <p>Appending to the end of a <tt>BasicBlock</tt> is so common that
2111 the <tt>Instruction</tt> class and <tt>Instruction</tt>-derived
2112 classes provide constructors which take a pointer to a
2113 <tt>BasicBlock</tt> to be appended to. For example code that
2114 looked like: </p>
2115
2116<div class="doc_code">
2117<pre>
2118BasicBlock *pb = ...;
2119Instruction *newInst = new Instruction(...);
2120
2121pb-&gt;getInstList().push_back(newInst); // <i>Appends newInst to pb</i>
2122</pre>
2123</div>
2124
2125 <p>becomes: </p>
2126
2127<div class="doc_code">
2128<pre>
2129BasicBlock *pb = ...;
2130Instruction *newInst = new Instruction(..., pb);
2131</pre>
2132</div>
2133
2134 <p>which is much cleaner, especially if you are creating
2135 long instruction streams.</p></li>
2136
2137 <li>Insertion into an implicit instruction list
2138
2139 <p><tt>Instruction</tt> instances that are already in <tt>BasicBlock</tt>s
2140 are implicitly associated with an existing instruction list: the instruction
2141 list of the enclosing basic block. Thus, we could have accomplished the same
2142 thing as the above code without being given a <tt>BasicBlock</tt> by doing:
2143 </p>
2144
2145<div class="doc_code">
2146<pre>
2147Instruction *pi = ...;
2148Instruction *newInst = new Instruction(...);
2149
2150pi-&gt;getParent()-&gt;getInstList().insert(pi, newInst);
2151</pre>
2152</div>
2153
2154 <p>In fact, this sequence of steps occurs so frequently that the
2155 <tt>Instruction</tt> class and <tt>Instruction</tt>-derived classes provide
2156 constructors which take (as a default parameter) a pointer to an
2157 <tt>Instruction</tt> which the newly-created <tt>Instruction</tt> should
2158 precede. That is, <tt>Instruction</tt> constructors are capable of
2159 inserting the newly-created instance into the <tt>BasicBlock</tt> of a
2160 provided instruction, immediately before that instruction. Using an
2161 <tt>Instruction</tt> constructor with a <tt>insertBefore</tt> (default)
2162 parameter, the above code becomes:</p>
2163
2164<div class="doc_code">
2165<pre>
2166Instruction* pi = ...;
2167Instruction* newInst = new Instruction(..., pi);
2168</pre>
2169</div>
2170
2171 <p>which is much cleaner, especially if you're creating a lot of
2172 instructions and adding them to <tt>BasicBlock</tt>s.</p></li>
2173</ul>
2174
2175</div>
2176
2177<!--_______________________________________________________________________-->
2178<div class="doc_subsubsection">
2179 <a name="schanges_deleting">Deleting <tt>Instruction</tt>s</a>
2180</div>
2181
2182<div class="doc_text">
2183
2184<p>Deleting an instruction from an existing sequence of instructions that form a
2185<a href="#BasicBlock"><tt>BasicBlock</tt></a> is very straight-forward. First,
2186you must have a pointer to the instruction that you wish to delete. Second, you
2187need to obtain the pointer to that instruction's basic block. You use the
2188pointer to the basic block to get its list of instructions and then use the
2189erase function to remove your instruction. For example:</p>
2190
2191<div class="doc_code">
2192<pre>
2193<a href="#Instruction">Instruction</a> *I = .. ;
2194I-&gt;eraseFromParent();
2195</pre>
2196</div>
2197
2198</div>
2199
2200<!--_______________________________________________________________________-->
2201<div class="doc_subsubsection">
2202 <a name="schanges_replacing">Replacing an <tt>Instruction</tt> with another
2203 <tt>Value</tt></a>
2204</div>
2205
2206<div class="doc_text">
2207
2208<p><i>Replacing individual instructions</i></p>
2209
2210<p>Including "<a href="/doxygen/BasicBlockUtils_8h-source.html">llvm/Transforms/Utils/BasicBlockUtils.h</a>"
2211permits use of two very useful replace functions: <tt>ReplaceInstWithValue</tt>
2212and <tt>ReplaceInstWithInst</tt>.</p>
2213
2214<h4><a name="schanges_deleting">Deleting <tt>Instruction</tt>s</a></h4>
2215
2216<ul>
2217 <li><tt>ReplaceInstWithValue</tt>
2218
2219 <p>This function replaces all uses of a given instruction with a value,
2220 and then removes the original instruction. The following example
2221 illustrates the replacement of the result of a particular
2222 <tt>AllocaInst</tt> that allocates memory for a single integer with a null
2223 pointer to an integer.</p>
2224
2225<div class="doc_code">
2226<pre>
2227AllocaInst* instToReplace = ...;
2228BasicBlock::iterator ii(instToReplace);
2229
2230ReplaceInstWithValue(instToReplace-&gt;getParent()-&gt;getInstList(), ii,
2231 Constant::getNullValue(PointerType::getUnqual(Type::Int32Ty)));
2232</pre></div></li>
2233
2234 <li><tt>ReplaceInstWithInst</tt>
2235
2236 <p>This function replaces a particular instruction with another
2237 instruction, inserting the new instruction into the basic block at the
2238 location where the old instruction was, and replacing any uses of the old
2239 instruction with the new instruction. The following example illustrates
2240 the replacement of one <tt>AllocaInst</tt> with another.</p>
2241
2242<div class="doc_code">
2243<pre>
2244AllocaInst* instToReplace = ...;
2245BasicBlock::iterator ii(instToReplace);
2246
2247ReplaceInstWithInst(instToReplace-&gt;getParent()-&gt;getInstList(), ii,
2248 new AllocaInst(Type::Int32Ty, 0, "ptrToReplacedInt"));
2249</pre></div></li>
2250</ul>
2251
2252<p><i>Replacing multiple uses of <tt>User</tt>s and <tt>Value</tt>s</i></p>
2253
2254<p>You can use <tt>Value::replaceAllUsesWith</tt> and
2255<tt>User::replaceUsesOfWith</tt> to change more than one use at a time. See the
2256doxygen documentation for the <a href="/doxygen/classllvm_1_1Value.html">Value Class</a>
2257and <a href="/doxygen/classllvm_1_1User.html">User Class</a>, respectively, for more
2258information.</p>
2259
2260<!-- Value::replaceAllUsesWith User::replaceUsesOfWith Point out:
2261include/llvm/Transforms/Utils/ especially BasicBlockUtils.h with:
2262ReplaceInstWithValue, ReplaceInstWithInst -->
2263
2264</div>
2265
2266<!--_______________________________________________________________________-->
2267<div class="doc_subsubsection">
2268 <a name="schanges_deletingGV">Deleting <tt>GlobalVariable</tt>s</a>
2269</div>
2270
2271<div class="doc_text">
2272
2273<p>Deleting a global variable from a module is just as easy as deleting an
2274Instruction. First, you must have a pointer to the global variable that you wish
2275 to delete. You use this pointer to erase it from its parent, the module.
2276 For example:</p>
2277
2278<div class="doc_code">
2279<pre>
2280<a href="#GlobalVariable">GlobalVariable</a> *GV = .. ;
2281
2282GV-&gt;eraseFromParent();
2283</pre>
2284</div>
2285
2286</div>
2287
2288<!-- ======================================================================= -->
2289<div class="doc_subsection">
2290 <a name="create_types">How to Create Types</a>
2291</div>
2292
2293<div class="doc_text">
2294
2295<p>In generating IR, you may need some complex types. If you know these types
2296statically, you can use <tt>TypeBuilder&lt;...&gt;::get()</tt>, defined
2297in <tt>llvm/Support/TypeBuilder.h</tt>, to retrieve them. <tt>TypeBuilder</tt>
2298has two forms depending on whether you're building types for cross-compilation
2299or native library use. <tt>TypeBuilder&lt;T, true&gt;</tt> requires
2300that <tt>T</tt> be independent of the host environment, meaning that it's built
2301out of types from
2302the <a href="/doxygen/namespacellvm_1_1types.html"><tt>llvm::types</tt></a>
2303namespace and pointers, functions, arrays, etc. built of
2304those. <tt>TypeBuilder&lt;T, false&gt;</tt> additionally allows native C types
2305whose size may depend on the host compiler. For example,</p>
2306
2307<div class="doc_code">
2308<pre>
2309FunctionType *ft = TypeBuilder&lt;types::i&lt;8&gt;(types::i&lt;32&gt;*), true&gt;::get();
2310</pre>
2311</div>
2312
2313<p>is easier to read and write than the equivalent</p>
2314
2315<div class="doc_code">
2316<pre>
2317std::vector&lt;const Type*&gt; params;
2318params.push_back(PointerType::getUnqual(Type::Int32Ty));
2319FunctionType *ft = FunctionType::get(Type::Int8Ty, params, false);
2320</pre>
2321</div>
2322
2323<p>See the <a href="/doxygen/TypeBuilder_8h-source.html#l00001">class
2324comment</a> for more details.</p>
2325
2326</div>
2327
2328<!-- *********************************************************************** -->
2329<div class="doc_section">
2330 <a name="threading">Threads and LLVM</a>
2331</div>
2332<!-- *********************************************************************** -->
2333
2334<div class="doc_text">
2335<p>
2336This section describes the interaction of the LLVM APIs with multithreading,
2337both on the part of client applications, and in the JIT, in the hosted
2338application.
2339</p>
2340
2341<p>
2342Note that LLVM's support for multithreading is still relatively young. Up
2343through version 2.5, the execution of threaded hosted applications was
2344supported, but not threaded client access to the APIs. While this use case is
2345now supported, clients <em>must</em> adhere to the guidelines specified below to
2346ensure proper operation in multithreaded mode.
2347</p>
2348
2349<p>
2350Note that, on Unix-like platforms, LLVM requires the presence of GCC's atomic
2351intrinsics in order to support threaded operation. If you need a
2352multhreading-capable LLVM on a platform without a suitably modern system
2353compiler, consider compiling LLVM and LLVM-GCC in single-threaded mode, and
2354using the resultant compiler to build a copy of LLVM with multithreading
2355support.
2356</p>
2357</div>
2358
2359<!-- ======================================================================= -->
2360<div class="doc_subsection">
2361 <a name="startmultithreaded">Entering and Exiting Multithreaded Mode</a>
2362</div>
2363
2364<div class="doc_text">
2365
2366<p>
2367In order to properly protect its internal data structures while avoiding
2368excessive locking overhead in the single-threaded case, the LLVM must intialize
2369certain data structures necessary to provide guards around its internals. To do
2370so, the client program must invoke <tt>llvm_start_multithreaded()</tt> before
2371making any concurrent LLVM API calls. To subsequently tear down these
2372structures, use the <tt>llvm_stop_multithreaded()</tt> call. You can also use
2373the <tt>llvm_is_multithreaded()</tt> call to check the status of multithreaded
2374mode.
2375</p>
2376
2377<p>
2378Note that both of these calls must be made <em>in isolation</em>. That is to
2379say that no other LLVM API calls may be executing at any time during the
2380execution of <tt>llvm_start_multithreaded()</tt> or <tt>llvm_stop_multithreaded
2381</tt>. It's is the client's responsibility to enforce this isolation.
2382</p>
2383
2384<p>
2385The return value of <tt>llvm_start_multithreaded()</tt> indicates the success or
2386failure of the initialization. Failure typically indicates that your copy of
2387LLVM was built without multithreading support, typically because GCC atomic
2388intrinsics were not found in your system compiler. In this case, the LLVM API
2389will not be safe for concurrent calls. However, it <em>will</em> be safe for
2390hosting threaded applications in the JIT, though <a href="#jitthreading">care
2391must be taken</a> to ensure that side exits and the like do not accidentally
2392result in concurrent LLVM API calls.
2393</p>
2394</div>
2395
2396<!-- ======================================================================= -->
2397<div class="doc_subsection">
2398 <a name="shutdown">Ending Execution with <tt>llvm_shutdown()</tt></a>
2399</div>
2400
2401<div class="doc_text">
2402<p>
2403When you are done using the LLVM APIs, you should call <tt>llvm_shutdown()</tt>
2404to deallocate memory used for internal structures. This will also invoke
2405<tt>llvm_stop_multithreaded()</tt> if LLVM is operating in multithreaded mode.
2406As such, <tt>llvm_shutdown()</tt> requires the same isolation guarantees as
2407<tt>llvm_stop_multithreaded()</tt>.
2408</p>
2409
2410<p>
2411Note that, if you use scope-based shutdown, you can use the
2412<tt>llvm_shutdown_obj</tt> class, which calls <tt>llvm_shutdown()</tt> in its
2413destructor.
2414</div>
2415
2416<!-- ======================================================================= -->
2417<div class="doc_subsection">
2418 <a name="managedstatic">Lazy Initialization with <tt>ManagedStatic</tt></a>
2419</div>
2420
2421<div class="doc_text">
2422<p>
2423<tt>ManagedStatic</tt> is a utility class in LLVM used to implement static
2424initialization of static resources, such as the global type tables. Before the
2425invocation of <tt>llvm_shutdown()</tt>, it implements a simple lazy
2426initialization scheme. Once <tt>llvm_start_multithreaded()</tt> returns,
2427however, it uses double-checked locking to implement thread-safe lazy
2428initialization.
2429</p>
2430
2431<p>
2432Note that, because no other threads are allowed to issue LLVM API calls before
2433<tt>llvm_start_multithreaded()</tt> returns, it is possible to have
2434<tt>ManagedStatic</tt>s of <tt>llvm::sys::Mutex</tt>s.
2435</p>
2436
2437<p>
2438The <tt>llvm_acquire_global_lock()</tt> and <tt>llvm_release_global_lock</tt>
2439APIs provide access to the global lock used to implement the double-checked
2440locking for lazy initialization. These should only be used internally to LLVM,
2441and only if you know what you're doing!
2442</p>
2443</div>
2444
2445<!-- ======================================================================= -->
2446<div class="doc_subsection">
2447 <a name="llvmcontext">Achieving Isolation with <tt>LLVMContext</tt></a>
2448</div>
2449
2450<div class="doc_text">
2451<p>
2452<tt>LLVMContext</tt> is an opaque class in the LLVM API which clients can use
2453to operate multiple, isolated instances of LLVM concurrently within the same
2454address space. For instance, in a hypothetical compile-server, the compilation
2455of an individual translation unit is conceptually independent from all the
2456others, and it would be desirable to be able to compile incoming translation
2457units concurrently on independent server threads. Fortunately,
2458<tt>LLVMContext</tt> exists to enable just this kind of scenario!
2459</p>
2460
2461<p>
2462Conceptually, <tt>LLVMContext</tt> provides isolation. Every LLVM entity
2463(<tt>Module</tt>s, <tt>Value</tt>s, <tt>Type</tt>s, <tt>Constant</tt>s, etc.)
2464in LLVM's in-memory IR belongs to an <tt>LLVMContext</tt>. Entities in
2465different contexts <em>cannot</em> interact with each other: <tt>Module</tt>s in
2466different contexts cannot be linked together, <tt>Function</tt>s cannot be added
2467to <tt>Module</tt>s in different contexts, etc. What this means is that is is
2468safe to compile on multiple threads simultaneously, as long as no two threads
2469operate on entities within the same context.
2470</p>
2471
2472<p>
2473In practice, very few places in the API require the explicit specification of a
2474<tt>LLVMContext</tt>, other than the <tt>Type</tt> creation/lookup APIs.
2475Because every <tt>Type</tt> carries a reference to its owning context, most
2476other entities can determine what context they belong to by looking at their
2477own <tt>Type</tt>. If you are adding new entities to LLVM IR, please try to
2478maintain this interface design.
2479</p>
2480
2481<p>
2482For clients that do <em>not</em> require the benefits of isolation, LLVM
2483provides a convenience API <tt>getGlobalContext()</tt>. This returns a global,
2484lazily initialized <tt>LLVMContext</tt> that may be used in situations where
2485isolation is not a concern.
2486</p>
2487</div>
2488
2489<!-- ======================================================================= -->
2490<div class="doc_subsection">
2491 <a name="jitthreading">Threads and the JIT</a>
2492</div>
2493
2494<div class="doc_text">
2495<p>
2496LLVM's "eager" JIT compiler is safe to use in threaded programs. Multiple
2497threads can call <tt>ExecutionEngine::getPointerToFunction()</tt> or
2498<tt>ExecutionEngine::runFunction()</tt> concurrently, and multiple threads can
2499run code output by the JIT concurrently. The user must still ensure that only
2500one thread accesses IR in a given <tt>LLVMContext</tt> while another thread
2501might be modifying it. One way to do that is to always hold the JIT lock while
2502accessing IR outside the JIT (the JIT <em>modifies</em> the IR by adding
2503<tt>CallbackVH</tt>s). Another way is to only
2504call <tt>getPointerToFunction()</tt> from the <tt>LLVMContext</tt>'s thread.
2505</p>
2506
2507<p>When the JIT is configured to compile lazily (using
2508<tt>ExecutionEngine::DisableLazyCompilation(false)</tt>), there is currently a
2509<a href="http://llvm.org/bugs/show_bug.cgi?id=5184">race condition</a> in
2510updating call sites after a function is lazily-jitted. It's still possible to
2511use the lazy JIT in a threaded program if you ensure that only one thread at a
2512time can call any particular lazy stub and that the JIT lock guards any IR
2513access, but we suggest using only the eager JIT in threaded programs.
2514</p>
2515</div>
2516
2517<!-- *********************************************************************** -->
2518<div class="doc_section">
2519 <a name="advanced">Advanced Topics</a>
2520</div>
2521<!-- *********************************************************************** -->
2522
2523<div class="doc_text">
2524<p>
2525This section describes some of the advanced or obscure API's that most clients
2526do not need to be aware of. These API's tend manage the inner workings of the
2527LLVM system, and only need to be accessed in unusual circumstances.
2528</p>
2529</div>
2530
2531<!-- ======================================================================= -->
2532<div class="doc_subsection">
2533 <a name="TypeResolve">LLVM Type Resolution</a>
2534</div>
2535
2536<div class="doc_text">
2537
2538<p>
2539The LLVM type system has a very simple goal: allow clients to compare types for
2540structural equality with a simple pointer comparison (aka a shallow compare).
2541This goal makes clients much simpler and faster, and is used throughout the LLVM
2542system.
2543</p>
2544
2545<p>
2546Unfortunately achieving this goal is not a simple matter. In particular,
2547recursive types and late resolution of opaque types makes the situation very
2548difficult to handle. Fortunately, for the most part, our implementation makes
2549most clients able to be completely unaware of the nasty internal details. The
2550primary case where clients are exposed to the inner workings of it are when
2551building a recursive type. In addition to this case, the LLVM bitcode reader,
2552assembly parser, and linker also have to be aware of the inner workings of this
2553system.
2554</p>
2555
2556<p>
2557For our purposes below, we need three concepts. First, an "Opaque Type" is
2558exactly as defined in the <a href="LangRef.html#t_opaque">language
2559reference</a>. Second an "Abstract Type" is any type which includes an
2560opaque type as part of its type graph (for example "<tt>{ opaque, i32 }</tt>").
2561Third, a concrete type is a type that is not an abstract type (e.g. "<tt>{ i32,
2562float }</tt>").
2563</p>
2564
2565</div>
2566
2567<!-- ______________________________________________________________________ -->
2568<div class="doc_subsubsection">
2569 <a name="BuildRecType">Basic Recursive Type Construction</a>
2570</div>
2571
2572<div class="doc_text">
2573
2574<p>
2575Because the most common question is "how do I build a recursive type with LLVM",
2576we answer it now and explain it as we go. Here we include enough to cause this
2577to be emitted to an output .ll file:
2578</p>
2579
2580<div class="doc_code">
2581<pre>
2582%mylist = type { %mylist*, i32 }
2583</pre>
2584</div>
2585
2586<p>
2587To build this, use the following LLVM APIs:
2588</p>
2589
2590<div class="doc_code">
2591<pre>
2592// <i>Create the initial outer struct</i>
2593<a href="#PATypeHolder">PATypeHolder</a> StructTy = OpaqueType::get();
2594std::vector&lt;const Type*&gt; Elts;
2595Elts.push_back(PointerType::getUnqual(StructTy));
2596Elts.push_back(Type::Int32Ty);
2597StructType *NewSTy = StructType::get(Elts);
2598
2599// <i>At this point, NewSTy = "{ opaque*, i32 }". Tell VMCore that</i>
2600// <i>the struct and the opaque type are actually the same.</i>
2601cast&lt;OpaqueType&gt;(StructTy.get())-&gt;<a href="#refineAbstractTypeTo">refineAbstractTypeTo</a>(NewSTy);
2602
2603// <i>NewSTy is potentially invalidated, but StructTy (a <a href="#PATypeHolder">PATypeHolder</a>) is</i>
2604// <i>kept up-to-date</i>
2605NewSTy = cast&lt;StructType&gt;(StructTy.get());
2606
2607// <i>Add a name for the type to the module symbol table (optional)</i>
2608MyModule-&gt;addTypeName("mylist", NewSTy);
2609</pre>
2610</div>
2611
2612<p>
2613This code shows the basic approach used to build recursive types: build a
2614non-recursive type using 'opaque', then use type unification to close the cycle.
2615The type unification step is performed by the <tt><a
2616href="#refineAbstractTypeTo">refineAbstractTypeTo</a></tt> method, which is
2617described next. After that, we describe the <a
2618href="#PATypeHolder">PATypeHolder class</a>.
2619</p>
2620
2621</div>
2622
2623<!-- ______________________________________________________________________ -->
2624<div class="doc_subsubsection">
2625 <a name="refineAbstractTypeTo">The <tt>refineAbstractTypeTo</tt> method</a>
2626</div>
2627
2628<div class="doc_text">
2629<p>
2630The <tt>refineAbstractTypeTo</tt> method starts the type unification process.
2631While this method is actually a member of the DerivedType class, it is most
2632often used on OpaqueType instances. Type unification is actually a recursive
2633process. After unification, types can become structurally isomorphic to
2634existing types, and all duplicates are deleted (to preserve pointer equality).
2635</p>
2636
2637<p>
2638In the example above, the OpaqueType object is definitely deleted.
2639Additionally, if there is an "{ \2*, i32}" type already created in the system,
2640the pointer and struct type created are <b>also</b> deleted. Obviously whenever
2641a type is deleted, any "Type*" pointers in the program are invalidated. As
2642such, it is safest to avoid having <i>any</i> "Type*" pointers to abstract types
2643live across a call to <tt>refineAbstractTypeTo</tt> (note that non-abstract
2644types can never move or be deleted). To deal with this, the <a
2645href="#PATypeHolder">PATypeHolder</a> class is used to maintain a stable
2646reference to a possibly refined type, and the <a
2647href="#AbstractTypeUser">AbstractTypeUser</a> class is used to update more
2648complex datastructures.
2649</p>
2650
2651</div>
2652
2653<!-- ______________________________________________________________________ -->
2654<div class="doc_subsubsection">
2655 <a name="PATypeHolder">The PATypeHolder Class</a>
2656</div>
2657
2658<div class="doc_text">
2659<p>
2660PATypeHolder is a form of a "smart pointer" for Type objects. When VMCore
2661happily goes about nuking types that become isomorphic to existing types, it
2662automatically updates all PATypeHolder objects to point to the new type. In the
2663example above, this allows the code to maintain a pointer to the resultant
2664resolved recursive type, even though the Type*'s are potentially invalidated.
2665</p>
2666
2667<p>
2668PATypeHolder is an extremely light-weight object that uses a lazy union-find
2669implementation to update pointers. For example the pointer from a Value to its
2670Type is maintained by PATypeHolder objects.
2671</p>
2672
2673</div>
2674
2675<!-- ______________________________________________________________________ -->
2676<div class="doc_subsubsection">
2677 <a name="AbstractTypeUser">The AbstractTypeUser Class</a>
2678</div>
2679
2680<div class="doc_text">
2681
2682<p>
2683Some data structures need more to perform more complex updates when types get
2684resolved. To support this, a class can derive from the AbstractTypeUser class.
2685This class
2686allows it to get callbacks when certain types are resolved. To register to get
2687callbacks for a particular type, the DerivedType::{add/remove}AbstractTypeUser
2688methods can be called on a type. Note that these methods only work for <i>
2689 abstract</i> types. Concrete types (those that do not include any opaque
2690objects) can never be refined.
2691</p>
2692</div>
2693
2694
2695<!-- ======================================================================= -->
2696<div class="doc_subsection">
2697 <a name="SymbolTable">The <tt>ValueSymbolTable</tt> and
2698 <tt>TypeSymbolTable</tt> classes</a>
2699</div>
2700
2701<div class="doc_text">
2702<p>The <tt><a href="http://llvm.org/doxygen/classllvm_1_1ValueSymbolTable.html">
2703ValueSymbolTable</a></tt> class provides a symbol table that the <a
2704href="#Function"><tt>Function</tt></a> and <a href="#Module">
2705<tt>Module</tt></a> classes use for naming value definitions. The symbol table
2706can provide a name for any <a href="#Value"><tt>Value</tt></a>.
2707The <tt><a href="http://llvm.org/doxygen/classllvm_1_1TypeSymbolTable.html">
2708TypeSymbolTable</a></tt> class is used by the <tt>Module</tt> class to store
2709names for types.</p>
2710
2711<p>Note that the <tt>SymbolTable</tt> class should not be directly accessed
2712by most clients. It should only be used when iteration over the symbol table
2713names themselves are required, which is very special purpose. Note that not
2714all LLVM
2715<tt><a href="#Value">Value</a></tt>s have names, and those without names (i.e. they have
2716an empty name) do not exist in the symbol table.
2717</p>
2718
2719<p>These symbol tables support iteration over the values/types in the symbol
2720table with <tt>begin/end/iterator</tt> and supports querying to see if a
2721specific name is in the symbol table (with <tt>lookup</tt>). The
2722<tt>ValueSymbolTable</tt> class exposes no public mutator methods, instead,
2723simply call <tt>setName</tt> on a value, which will autoinsert it into the
2724appropriate symbol table. For types, use the Module::addTypeName method to
2725insert entries into the symbol table.</p>
2726
2727</div>
2728
2729
2730
2731<!-- ======================================================================= -->
2732<div class="doc_subsection">
2733 <a name="UserLayout">The <tt>User</tt> and owned <tt>Use</tt> classes' memory layout</a>
2734</div>
2735
2736<div class="doc_text">
2737<p>The <tt><a href="http://llvm.org/doxygen/classllvm_1_1User.html">
2738User</a></tt> class provides a basis for expressing the ownership of <tt>User</tt>
2739towards other <tt><a href="http://llvm.org/doxygen/classllvm_1_1Value.html">
2740Value</a></tt>s. The <tt><a href="http://llvm.org/doxygen/classllvm_1_1Use.html">
2741Use</a></tt> helper class is employed to do the bookkeeping and to facilitate <i>O(1)</i>
2742addition and removal.</p>
2743
2744<!-- ______________________________________________________________________ -->
2745<div class="doc_subsubsection">
2746 <a name="Use2User">Interaction and relationship between <tt>User</tt> and <tt>Use</tt> objects</a>
2747</div>
2748
2749<div class="doc_text">
2750<p>
2751A subclass of <tt>User</tt> can choose between incorporating its <tt>Use</tt> objects
2752or refer to them out-of-line by means of a pointer. A mixed variant
2753(some <tt>Use</tt>s inline others hung off) is impractical and breaks the invariant
2754that the <tt>Use</tt> objects belonging to the same <tt>User</tt> form a contiguous array.
2755</p>
2756</div>
2757
2758<p>
2759We have 2 different layouts in the <tt>User</tt> (sub)classes:
2760<ul>
2761<li><p>Layout a)
2762The <tt>Use</tt> object(s) are inside (resp. at fixed offset) of the <tt>User</tt>
2763object and there are a fixed number of them.</p>
2764
2765<li><p>Layout b)
2766The <tt>Use</tt> object(s) are referenced by a pointer to an
2767array from the <tt>User</tt> object and there may be a variable
2768number of them.</p>
2769</ul>
2770<p>
2771As of v2.4 each layout still possesses a direct pointer to the
2772start of the array of <tt>Use</tt>s. Though not mandatory for layout a),
2773we stick to this redundancy for the sake of simplicity.
2774The <tt>User</tt> object also stores the number of <tt>Use</tt> objects it
2775has. (Theoretically this information can also be calculated
2776given the scheme presented below.)</p>
2777<p>
2778Special forms of allocation operators (<tt>operator new</tt>)
2779enforce the following memory layouts:</p>
2780
2781<ul>
2782<li><p>Layout a) is modelled by prepending the <tt>User</tt> object by the <tt>Use[]</tt> array.</p>
2783
2784<pre>
2785...---.---.---.---.-------...
2786 | P | P | P | P | User
2787'''---'---'---'---'-------'''
2788</pre>
2789
2790<li><p>Layout b) is modelled by pointing at the <tt>Use[]</tt> array.</p>
2791<pre>
2792.-------...
2793| User
2794'-------'''
2795 |
2796 v
2797 .---.---.---.---...
2798 | P | P | P | P |
2799 '---'---'---'---'''
2800</pre>
2801</ul>
2802<i>(In the above figures '<tt>P</tt>' stands for the <tt>Use**</tt> that
2803 is stored in each <tt>Use</tt> object in the member <tt>Use::Prev</tt>)</i>
2804
2805<!-- ______________________________________________________________________ -->
2806<div class="doc_subsubsection">
2807 <a name="Waymarking">The waymarking algorithm</a>
2808</div>
2809
2810<div class="doc_text">
2811<p>
2812Since the <tt>Use</tt> objects are deprived of the direct (back)pointer to
2813their <tt>User</tt> objects, there must be a fast and exact method to
2814recover it. This is accomplished by the following scheme:</p>
2815</div>
2816
2817A bit-encoding in the 2 LSBits (least significant bits) of the <tt>Use::Prev</tt> allows to find the
2818start of the <tt>User</tt> object:
2819<ul>
2820<li><tt>00</tt> &mdash;&gt; binary digit 0</li>
2821<li><tt>01</tt> &mdash;&gt; binary digit 1</li>
2822<li><tt>10</tt> &mdash;&gt; stop and calculate (<tt>s</tt>)</li>
2823<li><tt>11</tt> &mdash;&gt; full stop (<tt>S</tt>)</li>
2824</ul>
2825<p>
2826Given a <tt>Use*</tt>, all we have to do is to walk till we get
2827a stop and we either have a <tt>User</tt> immediately behind or
2828we have to walk to the next stop picking up digits
2829and calculating the offset:</p>
2830<pre>
2831.---.---.---.---.---.---.---.---.---.---.---.---.---.---.---.---.----------------
2832| 1 | s | 1 | 0 | 1 | 0 | s | 1 | 1 | 0 | s | 1 | 1 | s | 1 | S | User (or User*)
2833'---'---'---'---'---'---'---'---'---'---'---'---'---'---'---'---'----------------
2834 |+15 |+10 |+6 |+3 |+1
2835 | | | | |__>
2836 | | | |__________>
2837 | | |______________________>
2838 | |______________________________________>
2839 |__________________________________________________________>
2840</pre>
2841<p>
2842Only the significant number of bits need to be stored between the
2843stops, so that the <i>worst case is 20 memory accesses</i> when there are
28441000 <tt>Use</tt> objects associated with a <tt>User</tt>.</p>
2845
2846<!-- ______________________________________________________________________ -->
2847<div class="doc_subsubsection">
2848 <a name="ReferenceImpl">Reference implementation</a>
2849</div>
2850
2851<div class="doc_text">
2852<p>
2853The following literate Haskell fragment demonstrates the concept:</p>
2854</div>
2855
2856<div class="doc_code">
2857<pre>
2858> import Test.QuickCheck
2859>
2860> digits :: Int -> [Char] -> [Char]
2861> digits 0 acc = '0' : acc
2862> digits 1 acc = '1' : acc
2863> digits n acc = digits (n `div` 2) $ digits (n `mod` 2) acc
2864>
2865> dist :: Int -> [Char] -> [Char]
2866> dist 0 [] = ['S']
2867> dist 0 acc = acc
2868> dist 1 acc = let r = dist 0 acc in 's' : digits (length r) r
2869> dist n acc = dist (n - 1) $ dist 1 acc
2870>
2871> takeLast n ss = reverse $ take n $ reverse ss
2872>
2873> test = takeLast 40 $ dist 20 []
2874>
2875</pre>
2876</div>
2877<p>
2878Printing &lt;test&gt; gives: <tt>"1s100000s11010s10100s1111s1010s110s11s1S"</tt></p>
2879<p>
2880The reverse algorithm computes the length of the string just by examining
2881a certain prefix:</p>
2882
2883<div class="doc_code">
2884<pre>
2885> pref :: [Char] -> Int
2886> pref "S" = 1
2887> pref ('s':'1':rest) = decode 2 1 rest
2888> pref (_:rest) = 1 + pref rest
2889>
2890> decode walk acc ('0':rest) = decode (walk + 1) (acc * 2) rest
2891> decode walk acc ('1':rest) = decode (walk + 1) (acc * 2 + 1) rest
2892> decode walk acc _ = walk + acc
2893>
2894</pre>
2895</div>
2896<p>
2897Now, as expected, printing &lt;pref test&gt; gives <tt>40</tt>.</p>
2898<p>
2899We can <i>quickCheck</i> this with following property:</p>
2900
2901<div class="doc_code">
2902<pre>
2903> testcase = dist 2000 []
2904> testcaseLength = length testcase
2905>
2906> identityProp n = n > 0 && n <= testcaseLength ==> length arr == pref arr
2907> where arr = takeLast n testcase
2908>
2909</pre>
2910</div>
2911<p>
2912As expected &lt;quickCheck identityProp&gt; gives:</p>
2913
2914<pre>
2915*Main> quickCheck identityProp
2916OK, passed 100 tests.
2917</pre>
2918<p>
2919Let's be a bit more exhaustive:</p>
2920
2921<div class="doc_code">
2922<pre>
2923>
2924> deepCheck p = check (defaultConfig { configMaxTest = 500 }) p
2925>
2926</pre>
2927</div>
2928<p>
2929And here is the result of &lt;deepCheck identityProp&gt;:</p>
2930
2931<pre>
2932*Main> deepCheck identityProp
2933OK, passed 500 tests.
2934</pre>
2935
2936<!-- ______________________________________________________________________ -->
2937<div class="doc_subsubsection">
2938 <a name="Tagging">Tagging considerations</a>
2939</div>
2940
2941<p>
2942To maintain the invariant that the 2 LSBits of each <tt>Use**</tt> in <tt>Use</tt>
2943never change after being set up, setters of <tt>Use::Prev</tt> must re-tag the
2944new <tt>Use**</tt> on every modification. Accordingly getters must strip the
2945tag bits.</p>
2946<p>
2947For layout b) instead of the <tt>User</tt> we find a pointer (<tt>User*</tt> with LSBit set).
2948Following this pointer brings us to the <tt>User</tt>. A portable trick ensures
2949that the first bytes of <tt>User</tt> (if interpreted as a pointer) never has
2950the LSBit set. (Portability is relying on the fact that all known compilers place the
2951<tt>vptr</tt> in the first word of the instances.)</p>
2952
2953</div>
2954
2955 <!-- *********************************************************************** -->
2956<div class="doc_section">
2957 <a name="coreclasses">The Core LLVM Class Hierarchy Reference </a>
2958</div>
2959<!-- *********************************************************************** -->
2960
2961<div class="doc_text">
2962<p><tt>#include "<a href="/doxygen/Type_8h-source.html">llvm/Type.h</a>"</tt>
2963<br>doxygen info: <a href="/doxygen/classllvm_1_1Type.html">Type Class</a></p>
2964
2965<p>The Core LLVM classes are the primary means of representing the program
2966being inspected or transformed. The core LLVM classes are defined in
2967header files in the <tt>include/llvm/</tt> directory, and implemented in
2968the <tt>lib/VMCore</tt> directory.</p>
2969
2970</div>
2971
2972<!-- ======================================================================= -->
2973<div class="doc_subsection">
2974 <a name="Type">The <tt>Type</tt> class and Derived Types</a>
2975</div>
2976
2977<div class="doc_text">
2978
2979 <p><tt>Type</tt> is a superclass of all type classes. Every <tt>Value</tt> has
2980 a <tt>Type</tt>. <tt>Type</tt> cannot be instantiated directly but only
2981 through its subclasses. Certain primitive types (<tt>VoidType</tt>,
2982 <tt>LabelType</tt>, <tt>FloatType</tt> and <tt>DoubleType</tt>) have hidden
2983 subclasses. They are hidden because they offer no useful functionality beyond
2984 what the <tt>Type</tt> class offers except to distinguish themselves from
2985 other subclasses of <tt>Type</tt>.</p>
2986 <p>All other types are subclasses of <tt>DerivedType</tt>. Types can be
2987 named, but this is not a requirement. There exists exactly
2988 one instance of a given shape at any one time. This allows type equality to
2989 be performed with address equality of the Type Instance. That is, given two
2990 <tt>Type*</tt> values, the types are identical if the pointers are identical.
2991 </p>
2992</div>
2993
2994<!-- _______________________________________________________________________ -->
2995<div class="doc_subsubsection">
2996 <a name="m_Type">Important Public Methods</a>
2997</div>
2998
2999<div class="doc_text">
3000
3001<ul>
3002 <li><tt>bool isInteger() const</tt>: Returns true for any integer type.</li>
3003
3004 <li><tt>bool isFloatingPoint()</tt>: Return true if this is one of the two
3005 floating point types.</li>
3006
3007 <li><tt>bool isAbstract()</tt>: Return true if the type is abstract (contains
3008 an OpaqueType anywhere in its definition).</li>
3009
3010 <li><tt>bool isSized()</tt>: Return true if the type has known size. Things
3011 that don't have a size are abstract types, labels and void.</li>
3012
3013</ul>
3014</div>
3015
3016<!-- _______________________________________________________________________ -->
3017<div class="doc_subsubsection">
3018 <a name="derivedtypes">Important Derived Types</a>
3019</div>
3020<div class="doc_text">
3021<dl>
3022 <dt><tt>IntegerType</tt></dt>
3023 <dd>Subclass of DerivedType that represents integer types of any bit width.
3024 Any bit width between <tt>IntegerType::MIN_INT_BITS</tt> (1) and
3025 <tt>IntegerType::MAX_INT_BITS</tt> (~8 million) can be represented.
3026 <ul>
3027 <li><tt>static const IntegerType* get(unsigned NumBits)</tt>: get an integer
3028 type of a specific bit width.</li>
3029 <li><tt>unsigned getBitWidth() const</tt>: Get the bit width of an integer
3030 type.</li>
3031 </ul>
3032 </dd>
3033 <dt><tt>SequentialType</tt></dt>
3034 <dd>This is subclassed by ArrayType and PointerType
3035 <ul>
3036 <li><tt>const Type * getElementType() const</tt>: Returns the type of each
3037 of the elements in the sequential type. </li>
3038 </ul>
3039 </dd>
3040 <dt><tt>ArrayType</tt></dt>
3041 <dd>This is a subclass of SequentialType and defines the interface for array
3042 types.
3043 <ul>
3044 <li><tt>unsigned getNumElements() const</tt>: Returns the number of
3045 elements in the array. </li>
3046 </ul>
3047 </dd>
3048 <dt><tt>PointerType</tt></dt>
3049 <dd>Subclass of SequentialType for pointer types.</dd>
3050 <dt><tt>VectorType</tt></dt>
3051 <dd>Subclass of SequentialType for vector types. A
3052 vector type is similar to an ArrayType but is distinguished because it is
3053 a first class type whereas ArrayType is not. Vector types are used for
3054 vector operations and are usually small vectors of of an integer or floating
3055 point type.</dd>
3056 <dt><tt>StructType</tt></dt>
3057 <dd>Subclass of DerivedTypes for struct types.</dd>
3058 <dt><tt><a name="FunctionType">FunctionType</a></tt></dt>
3059 <dd>Subclass of DerivedTypes for function types.
3060 <ul>
3061 <li><tt>bool isVarArg() const</tt>: Returns true if its a vararg
3062 function</li>
3063 <li><tt> const Type * getReturnType() const</tt>: Returns the
3064 return type of the function.</li>
3065 <li><tt>const Type * getParamType (unsigned i)</tt>: Returns
3066 the type of the ith parameter.</li>
3067 <li><tt> const unsigned getNumParams() const</tt>: Returns the
3068 number of formal parameters.</li>
3069 </ul>
3070 </dd>
3071 <dt><tt>OpaqueType</tt></dt>
3072 <dd>Sublcass of DerivedType for abstract types. This class
3073 defines no content and is used as a placeholder for some other type. Note
3074 that OpaqueType is used (temporarily) during type resolution for forward
3075 references of types. Once the referenced type is resolved, the OpaqueType
3076 is replaced with the actual type. OpaqueType can also be used for data
3077 abstraction. At link time opaque types can be resolved to actual types
3078 of the same name.</dd>
3079</dl>
3080</div>
3081
3082
3083
3084<!-- ======================================================================= -->
3085<div class="doc_subsection">
3086 <a name="Module">The <tt>Module</tt> class</a>
3087</div>
3088
3089<div class="doc_text">
3090
3091<p><tt>#include "<a
3092href="/doxygen/Module_8h-source.html">llvm/Module.h</a>"</tt><br> doxygen info:
3093<a href="/doxygen/classllvm_1_1Module.html">Module Class</a></p>
3094
3095<p>The <tt>Module</tt> class represents the top level structure present in LLVM
3096programs. An LLVM module is effectively either a translation unit of the
3097original program or a combination of several translation units merged by the
3098linker. The <tt>Module</tt> class keeps track of a list of <a
3099href="#Function"><tt>Function</tt></a>s, a list of <a
3100href="#GlobalVariable"><tt>GlobalVariable</tt></a>s, and a <a
3101href="#SymbolTable"><tt>SymbolTable</tt></a>. Additionally, it contains a few
3102helpful member functions that try to make common operations easy.</p>
3103
3104</div>
3105
3106<!-- _______________________________________________________________________ -->
3107<div class="doc_subsubsection">
3108 <a name="m_Module">Important Public Members of the <tt>Module</tt> class</a>
3109</div>
3110
3111<div class="doc_text">
3112
3113<ul>
3114 <li><tt>Module::Module(std::string name = "")</tt></li>
3115</ul>
3116
3117<p>Constructing a <a href="#Module">Module</a> is easy. You can optionally
3118provide a name for it (probably based on the name of the translation unit).</p>
3119
3120<ul>
3121 <li><tt>Module::iterator</tt> - Typedef for function list iterator<br>
3122 <tt>Module::const_iterator</tt> - Typedef for const_iterator.<br>
3123
3124 <tt>begin()</tt>, <tt>end()</tt>
3125 <tt>size()</tt>, <tt>empty()</tt>
3126
3127 <p>These are forwarding methods that make it easy to access the contents of
3128 a <tt>Module</tt> object's <a href="#Function"><tt>Function</tt></a>
3129 list.</p></li>
3130
3131 <li><tt>Module::FunctionListType &amp;getFunctionList()</tt>
3132
3133 <p> Returns the list of <a href="#Function"><tt>Function</tt></a>s. This is
3134 necessary to use when you need to update the list or perform a complex
3135 action that doesn't have a forwarding method.</p>
3136
3137 <p><!-- Global Variable --></p></li>
3138</ul>
3139
3140<hr>
3141
3142<ul>
3143 <li><tt>Module::global_iterator</tt> - Typedef for global variable list iterator<br>
3144
3145 <tt>Module::const_global_iterator</tt> - Typedef for const_iterator.<br>
3146
3147 <tt>global_begin()</tt>, <tt>global_end()</tt>
3148 <tt>global_size()</tt>, <tt>global_empty()</tt>
3149
3150 <p> These are forwarding methods that make it easy to access the contents of
3151 a <tt>Module</tt> object's <a
3152 href="#GlobalVariable"><tt>GlobalVariable</tt></a> list.</p></li>
3153
3154 <li><tt>Module::GlobalListType &amp;getGlobalList()</tt>
3155
3156 <p>Returns the list of <a
3157 href="#GlobalVariable"><tt>GlobalVariable</tt></a>s. This is necessary to
3158 use when you need to update the list or perform a complex action that
3159 doesn't have a forwarding method.</p>
3160
3161 <p><!-- Symbol table stuff --> </p></li>
3162</ul>
3163
3164<hr>
3165
3166<ul>
3167 <li><tt><a href="#SymbolTable">SymbolTable</a> *getSymbolTable()</tt>
3168
3169 <p>Return a reference to the <a href="#SymbolTable"><tt>SymbolTable</tt></a>
3170 for this <tt>Module</tt>.</p>
3171
3172 <p><!-- Convenience methods --></p></li>
3173</ul>
3174
3175<hr>
3176
3177<ul>
3178 <li><tt><a href="#Function">Function</a> *getFunction(const std::string
3179 &amp;Name, const <a href="#FunctionType">FunctionType</a> *Ty)</tt>
3180
3181 <p>Look up the specified function in the <tt>Module</tt> <a
3182 href="#SymbolTable"><tt>SymbolTable</tt></a>. If it does not exist, return
3183 <tt>null</tt>.</p></li>
3184
3185 <li><tt><a href="#Function">Function</a> *getOrInsertFunction(const
3186 std::string &amp;Name, const <a href="#FunctionType">FunctionType</a> *T)</tt>
3187
3188 <p>Look up the specified function in the <tt>Module</tt> <a
3189 href="#SymbolTable"><tt>SymbolTable</tt></a>. If it does not exist, add an
3190 external declaration for the function and return it.</p></li>
3191
3192 <li><tt>std::string getTypeName(const <a href="#Type">Type</a> *Ty)</tt>
3193
3194 <p>If there is at least one entry in the <a
3195 href="#SymbolTable"><tt>SymbolTable</tt></a> for the specified <a
3196 href="#Type"><tt>Type</tt></a>, return it. Otherwise return the empty
3197 string.</p></li>
3198
3199 <li><tt>bool addTypeName(const std::string &amp;Name, const <a
3200 href="#Type">Type</a> *Ty)</tt>
3201
3202 <p>Insert an entry in the <a href="#SymbolTable"><tt>SymbolTable</tt></a>
3203 mapping <tt>Name</tt> to <tt>Ty</tt>. If there is already an entry for this
3204 name, true is returned and the <a
3205 href="#SymbolTable"><tt>SymbolTable</tt></a> is not modified.</p></li>
3206</ul>
3207
3208</div>
3209
3210
3211<!-- ======================================================================= -->
3212<div class="doc_subsection">
3213 <a name="Value">The <tt>Value</tt> class</a>
3214</div>
3215
3216<div class="doc_text">
3217
3218<p><tt>#include "<a href="/doxygen/Value_8h-source.html">llvm/Value.h</a>"</tt>
3219<br>
3220doxygen info: <a href="/doxygen/classllvm_1_1Value.html">Value Class</a></p>
3221
3222<p>The <tt>Value</tt> class is the most important class in the LLVM Source
3223base. It represents a typed value that may be used (among other things) as an
3224operand to an instruction. There are many different types of <tt>Value</tt>s,
3225such as <a href="#Constant"><tt>Constant</tt></a>s,<a
3226href="#Argument"><tt>Argument</tt></a>s. Even <a
3227href="#Instruction"><tt>Instruction</tt></a>s and <a
3228href="#Function"><tt>Function</tt></a>s are <tt>Value</tt>s.</p>
3229
3230<p>A particular <tt>Value</tt> may be used many times in the LLVM representation
3231for a program. For example, an incoming argument to a function (represented
3232with an instance of the <a href="#Argument">Argument</a> class) is "used" by
3233every instruction in the function that references the argument. To keep track
3234of this relationship, the <tt>Value</tt> class keeps a list of all of the <a
3235href="#User"><tt>User</tt></a>s that is using it (the <a
3236href="#User"><tt>User</tt></a> class is a base class for all nodes in the LLVM
3237graph that can refer to <tt>Value</tt>s). This use list is how LLVM represents
3238def-use information in the program, and is accessible through the <tt>use_</tt>*
3239methods, shown below.</p>
3240
3241<p>Because LLVM is a typed representation, every LLVM <tt>Value</tt> is typed,
3242and this <a href="#Type">Type</a> is available through the <tt>getType()</tt>
3243method. In addition, all LLVM values can be named. The "name" of the
3244<tt>Value</tt> is a symbolic string printed in the LLVM code:</p>
3245
3246<div class="doc_code">
3247<pre>
3248%<b>foo</b> = add i32 1, 2
3249</pre>
3250</div>
3251
3252<p><a name="nameWarning">The name of this instruction is "foo".</a> <b>NOTE</b>
3253that the name of any value may be missing (an empty string), so names should
3254<b>ONLY</b> be used for debugging (making the source code easier to read,
3255debugging printouts), they should not be used to keep track of values or map
3256between them. For this purpose, use a <tt>std::map</tt> of pointers to the
3257<tt>Value</tt> itself instead.</p>
3258
3259<p>One important aspect of LLVM is that there is no distinction between an SSA
3260variable and the operation that produces it. Because of this, any reference to
3261the value produced by an instruction (or the value available as an incoming
3262argument, for example) is represented as a direct pointer to the instance of
3263the class that
3264represents this value. Although this may take some getting used to, it
3265simplifies the representation and makes it easier to manipulate.</p>
3266
3267</div>
3268
3269<!-- _______________________________________________________________________ -->
3270<div class="doc_subsubsection">
3271 <a name="m_Value">Important Public Members of the <tt>Value</tt> class</a>
3272</div>
3273
3274<div class="doc_text">
3275
3276<ul>
3277 <li><tt>Value::use_iterator</tt> - Typedef for iterator over the
3278use-list<br>
3279 <tt>Value::use_const_iterator</tt> - Typedef for const_iterator over
3280the use-list<br>
3281 <tt>unsigned use_size()</tt> - Returns the number of users of the
3282value.<br>
3283 <tt>bool use_empty()</tt> - Returns true if there are no users.<br>
3284 <tt>use_iterator use_begin()</tt> - Get an iterator to the start of
3285the use-list.<br>
3286 <tt>use_iterator use_end()</tt> - Get an iterator to the end of the
3287use-list.<br>
3288 <tt><a href="#User">User</a> *use_back()</tt> - Returns the last
3289element in the list.
3290 <p> These methods are the interface to access the def-use
3291information in LLVM. As with all other iterators in LLVM, the naming
3292conventions follow the conventions defined by the <a href="#stl">STL</a>.</p>
3293 </li>
3294 <li><tt><a href="#Type">Type</a> *getType() const</tt>
3295 <p>This method returns the Type of the Value.</p>
3296 </li>
3297 <li><tt>bool hasName() const</tt><br>
3298 <tt>std::string getName() const</tt><br>
3299 <tt>void setName(const std::string &amp;Name)</tt>
3300 <p> This family of methods is used to access and assign a name to a <tt>Value</tt>,
3301be aware of the <a href="#nameWarning">precaution above</a>.</p>
3302 </li>
3303 <li><tt>void replaceAllUsesWith(Value *V)</tt>
3304
3305 <p>This method traverses the use list of a <tt>Value</tt> changing all <a
3306 href="#User"><tt>User</tt>s</a> of the current value to refer to
3307 "<tt>V</tt>" instead. For example, if you detect that an instruction always
3308 produces a constant value (for example through constant folding), you can
3309 replace all uses of the instruction with the constant like this:</p>
3310
3311<div class="doc_code">
3312<pre>
3313Inst-&gt;replaceAllUsesWith(ConstVal);
3314</pre>
3315</div>
3316
3317</ul>
3318
3319</div>
3320
3321<!-- ======================================================================= -->
3322<div class="doc_subsection">
3323 <a name="User">The <tt>User</tt> class</a>
3324</div>
3325
3326<div class="doc_text">
3327
3328<p>
3329<tt>#include "<a href="/doxygen/User_8h-source.html">llvm/User.h</a>"</tt><br>
3330doxygen info: <a href="/doxygen/classllvm_1_1User.html">User Class</a><br>
3331Superclass: <a href="#Value"><tt>Value</tt></a></p>
3332
3333<p>The <tt>User</tt> class is the common base class of all LLVM nodes that may
3334refer to <a href="#Value"><tt>Value</tt></a>s. It exposes a list of "Operands"
3335that are all of the <a href="#Value"><tt>Value</tt></a>s that the User is
3336referring to. The <tt>User</tt> class itself is a subclass of
3337<tt>Value</tt>.</p>
3338
3339<p>The operands of a <tt>User</tt> point directly to the LLVM <a
3340href="#Value"><tt>Value</tt></a> that it refers to. Because LLVM uses Static
3341Single Assignment (SSA) form, there can only be one definition referred to,
3342allowing this direct connection. This connection provides the use-def
3343information in LLVM.</p>
3344
3345</div>
3346
3347<!-- _______________________________________________________________________ -->
3348<div class="doc_subsubsection">
3349 <a name="m_User">Important Public Members of the <tt>User</tt> class</a>
3350</div>
3351
3352<div class="doc_text">
3353
3354<p>The <tt>User</tt> class exposes the operand list in two ways: through
3355an index access interface and through an iterator based interface.</p>
3356
3357<ul>
3358 <li><tt>Value *getOperand(unsigned i)</tt><br>
3359 <tt>unsigned getNumOperands()</tt>
3360 <p> These two methods expose the operands of the <tt>User</tt> in a
3361convenient form for direct access.</p></li>
3362
3363 <li><tt>User::op_iterator</tt> - Typedef for iterator over the operand
3364list<br>
3365 <tt>op_iterator op_begin()</tt> - Get an iterator to the start of
3366the operand list.<br>
3367 <tt>op_iterator op_end()</tt> - Get an iterator to the end of the
3368operand list.
3369 <p> Together, these methods make up the iterator based interface to
3370the operands of a <tt>User</tt>.</p></li>
3371</ul>
3372
3373</div>
3374
3375<!-- ======================================================================= -->
3376<div class="doc_subsection">
3377 <a name="Instruction">The <tt>Instruction</tt> class</a>
3378</div>
3379
3380<div class="doc_text">
3381
3382<p><tt>#include "</tt><tt><a
3383href="/doxygen/Instruction_8h-source.html">llvm/Instruction.h</a>"</tt><br>
3384doxygen info: <a href="/doxygen/classllvm_1_1Instruction.html">Instruction Class</a><br>
3385Superclasses: <a href="#User"><tt>User</tt></a>, <a
3386href="#Value"><tt>Value</tt></a></p>
3387
3388<p>The <tt>Instruction</tt> class is the common base class for all LLVM
3389instructions. It provides only a few methods, but is a very commonly used
3390class. The primary data tracked by the <tt>Instruction</tt> class itself is the
3391opcode (instruction type) and the parent <a
3392href="#BasicBlock"><tt>BasicBlock</tt></a> the <tt>Instruction</tt> is embedded
3393into. To represent a specific type of instruction, one of many subclasses of
3394<tt>Instruction</tt> are used.</p>
3395
3396<p> Because the <tt>Instruction</tt> class subclasses the <a
3397href="#User"><tt>User</tt></a> class, its operands can be accessed in the same
3398way as for other <a href="#User"><tt>User</tt></a>s (with the
3399<tt>getOperand()</tt>/<tt>getNumOperands()</tt> and
3400<tt>op_begin()</tt>/<tt>op_end()</tt> methods).</p> <p> An important file for
3401the <tt>Instruction</tt> class is the <tt>llvm/Instruction.def</tt> file. This
3402file contains some meta-data about the various different types of instructions
3403in LLVM. It describes the enum values that are used as opcodes (for example
3404<tt>Instruction::Add</tt> and <tt>Instruction::ICmp</tt>), as well as the
3405concrete sub-classes of <tt>Instruction</tt> that implement the instruction (for
3406example <tt><a href="#BinaryOperator">BinaryOperator</a></tt> and <tt><a
3407href="#CmpInst">CmpInst</a></tt>). Unfortunately, the use of macros in
3408this file confuses doxygen, so these enum values don't show up correctly in the
3409<a href="/doxygen/classllvm_1_1Instruction.html">doxygen output</a>.</p>
3410
3411</div>
3412
3413<!-- _______________________________________________________________________ -->
3414<div class="doc_subsubsection">
3415 <a name="s_Instruction">Important Subclasses of the <tt>Instruction</tt>
3416 class</a>
3417</div>
3418<div class="doc_text">
3419 <ul>
3420 <li><tt><a name="BinaryOperator">BinaryOperator</a></tt>
3421 <p>This subclasses represents all two operand instructions whose operands
3422 must be the same type, except for the comparison instructions.</p></li>
3423 <li><tt><a name="CastInst">CastInst</a></tt>
3424 <p>This subclass is the parent of the 12 casting instructions. It provides
3425 common operations on cast instructions.</p>
3426 <li><tt><a name="CmpInst">CmpInst</a></tt>
3427 <p>This subclass respresents the two comparison instructions,
3428 <a href="LangRef.html#i_icmp">ICmpInst</a> (integer opreands), and
3429 <a href="LangRef.html#i_fcmp">FCmpInst</a> (floating point operands).</p>
3430 <li><tt><a name="TerminatorInst">TerminatorInst</a></tt>
3431 <p>This subclass is the parent of all terminator instructions (those which
3432 can terminate a block).</p>
3433 </ul>
3434 </div>
3435
3436<!-- _______________________________________________________________________ -->
3437<div class="doc_subsubsection">
3438 <a name="m_Instruction">Important Public Members of the <tt>Instruction</tt>
3439 class</a>
3440</div>
3441
3442<div class="doc_text">
3443
3444<ul>
3445 <li><tt><a href="#BasicBlock">BasicBlock</a> *getParent()</tt>
3446 <p>Returns the <a href="#BasicBlock"><tt>BasicBlock</tt></a> that
3447this <tt>Instruction</tt> is embedded into.</p></li>
3448 <li><tt>bool mayWriteToMemory()</tt>
3449 <p>Returns true if the instruction writes to memory, i.e. it is a
3450 <tt>call</tt>,<tt>free</tt>,<tt>invoke</tt>, or <tt>store</tt>.</p></li>
3451 <li><tt>unsigned getOpcode()</tt>
3452 <p>Returns the opcode for the <tt>Instruction</tt>.</p></li>
3453 <li><tt><a href="#Instruction">Instruction</a> *clone() const</tt>
3454 <p>Returns another instance of the specified instruction, identical
3455in all ways to the original except that the instruction has no parent
3456(ie it's not embedded into a <a href="#BasicBlock"><tt>BasicBlock</tt></a>),
3457and it has no name</p></li>
3458</ul>
3459
3460</div>
3461
3462<!-- ======================================================================= -->
3463<div class="doc_subsection">
3464 <a name="Constant">The <tt>Constant</tt> class and subclasses</a>
3465</div>
3466
3467<div class="doc_text">
3468
3469<p>Constant represents a base class for different types of constants. It
3470is subclassed by ConstantInt, ConstantArray, etc. for representing
3471the various types of Constants. <a href="#GlobalValue">GlobalValue</a> is also
3472a subclass, which represents the address of a global variable or function.
3473</p>
3474
3475</div>
3476
3477<!-- _______________________________________________________________________ -->
3478<div class="doc_subsubsection">Important Subclasses of Constant </div>
3479<div class="doc_text">
3480<ul>
3481 <li>ConstantInt : This subclass of Constant represents an integer constant of
3482 any width.
3483 <ul>
3484 <li><tt>const APInt&amp; getValue() const</tt>: Returns the underlying
3485 value of this constant, an APInt value.</li>
3486 <li><tt>int64_t getSExtValue() const</tt>: Converts the underlying APInt
3487 value to an int64_t via sign extension. If the value (not the bit width)
3488 of the APInt is too large to fit in an int64_t, an assertion will result.
3489 For this reason, use of this method is discouraged.</li>
3490 <li><tt>uint64_t getZExtValue() const</tt>: Converts the underlying APInt
3491 value to a uint64_t via zero extension. IF the value (not the bit width)
3492 of the APInt is too large to fit in a uint64_t, an assertion will result.
3493 For this reason, use of this method is discouraged.</li>
3494 <li><tt>static ConstantInt* get(const APInt&amp; Val)</tt>: Returns the
3495 ConstantInt object that represents the value provided by <tt>Val</tt>.
3496 The type is implied as the IntegerType that corresponds to the bit width
3497 of <tt>Val</tt>.</li>
3498 <li><tt>static ConstantInt* get(const Type *Ty, uint64_t Val)</tt>:
3499 Returns the ConstantInt object that represents the value provided by
3500 <tt>Val</tt> for integer type <tt>Ty</tt>.</li>
3501 </ul>
3502 </li>
3503 <li>ConstantFP : This class represents a floating point constant.
3504 <ul>
3505 <li><tt>double getValue() const</tt>: Returns the underlying value of
3506 this constant. </li>
3507 </ul>
3508 </li>
3509 <li>ConstantArray : This represents a constant array.
3510 <ul>
3511 <li><tt>const std::vector&lt;Use&gt; &amp;getValues() const</tt>: Returns
3512 a vector of component constants that makeup this array. </li>
3513 </ul>
3514 </li>
3515 <li>ConstantStruct : This represents a constant struct.
3516 <ul>
3517 <li><tt>const std::vector&lt;Use&gt; &amp;getValues() const</tt>: Returns
3518 a vector of component constants that makeup this array. </li>
3519 </ul>
3520 </li>
3521 <li>GlobalValue : This represents either a global variable or a function. In
3522 either case, the value is a constant fixed address (after linking).
3523 </li>
3524</ul>
3525</div>
3526
3527
3528<!-- ======================================================================= -->
3529<div class="doc_subsection">
3530 <a name="GlobalValue">The <tt>GlobalValue</tt> class</a>
3531</div>
3532
3533<div class="doc_text">
3534
3535<p><tt>#include "<a
3536href="/doxygen/GlobalValue_8h-source.html">llvm/GlobalValue.h</a>"</tt><br>
3537doxygen info: <a href="/doxygen/classllvm_1_1GlobalValue.html">GlobalValue
3538Class</a><br>
3539Superclasses: <a href="#Constant"><tt>Constant</tt></a>,
3540<a href="#User"><tt>User</tt></a>, <a href="#Value"><tt>Value</tt></a></p>
3541
3542<p>Global values (<a href="#GlobalVariable"><tt>GlobalVariable</tt></a>s or <a
3543href="#Function"><tt>Function</tt></a>s) are the only LLVM values that are
3544visible in the bodies of all <a href="#Function"><tt>Function</tt></a>s.
3545Because they are visible at global scope, they are also subject to linking with
3546other globals defined in different translation units. To control the linking
3547process, <tt>GlobalValue</tt>s know their linkage rules. Specifically,
3548<tt>GlobalValue</tt>s know whether they have internal or external linkage, as
3549defined by the <tt>LinkageTypes</tt> enumeration.</p>
3550
3551<p>If a <tt>GlobalValue</tt> has internal linkage (equivalent to being
3552<tt>static</tt> in C), it is not visible to code outside the current translation
3553unit, and does not participate in linking. If it has external linkage, it is
3554visible to external code, and does participate in linking. In addition to
3555linkage information, <tt>GlobalValue</tt>s keep track of which <a
3556href="#Module"><tt>Module</tt></a> they are currently part of.</p>
3557
3558<p>Because <tt>GlobalValue</tt>s are memory objects, they are always referred to
3559by their <b>address</b>. As such, the <a href="#Type"><tt>Type</tt></a> of a
3560global is always a pointer to its contents. It is important to remember this
3561when using the <tt>GetElementPtrInst</tt> instruction because this pointer must
3562be dereferenced first. For example, if you have a <tt>GlobalVariable</tt> (a
3563subclass of <tt>GlobalValue)</tt> that is an array of 24 ints, type <tt>[24 x
3564i32]</tt>, then the <tt>GlobalVariable</tt> is a pointer to that array. Although
3565the address of the first element of this array and the value of the
3566<tt>GlobalVariable</tt> are the same, they have different types. The
3567<tt>GlobalVariable</tt>'s type is <tt>[24 x i32]</tt>. The first element's type
3568is <tt>i32.</tt> Because of this, accessing a global value requires you to
3569dereference the pointer with <tt>GetElementPtrInst</tt> first, then its elements
3570can be accessed. This is explained in the <a href="LangRef.html#globalvars">LLVM
3571Language Reference Manual</a>.</p>
3572
3573</div>
3574
3575<!-- _______________________________________________________________________ -->
3576<div class="doc_subsubsection">
3577 <a name="m_GlobalValue">Important Public Members of the <tt>GlobalValue</tt>
3578 class</a>
3579</div>
3580
3581<div class="doc_text">
3582
3583<ul>
3584 <li><tt>bool hasInternalLinkage() const</tt><br>
3585 <tt>bool hasExternalLinkage() const</tt><br>
3586 <tt>void setInternalLinkage(bool HasInternalLinkage)</tt>
3587 <p> These methods manipulate the linkage characteristics of the <tt>GlobalValue</tt>.</p>
3588 <p> </p>
3589 </li>
3590 <li><tt><a href="#Module">Module</a> *getParent()</tt>
3591 <p> This returns the <a href="#Module"><tt>Module</tt></a> that the
3592GlobalValue is currently embedded into.</p></li>
3593</ul>
3594
3595</div>
3596
3597<!-- ======================================================================= -->
3598<div class="doc_subsection">
3599 <a name="Function">The <tt>Function</tt> class</a>
3600</div>
3601
3602<div class="doc_text">
3603
3604<p><tt>#include "<a
3605href="/doxygen/Function_8h-source.html">llvm/Function.h</a>"</tt><br> doxygen
3606info: <a href="/doxygen/classllvm_1_1Function.html">Function Class</a><br>
3607Superclasses: <a href="#GlobalValue"><tt>GlobalValue</tt></a>,
3608<a href="#Constant"><tt>Constant</tt></a>,
3609<a href="#User"><tt>User</tt></a>,
3610<a href="#Value"><tt>Value</tt></a></p>
3611
3612<p>The <tt>Function</tt> class represents a single procedure in LLVM. It is
3613actually one of the more complex classes in the LLVM hierarchy because it must
3614keep track of a large amount of data. The <tt>Function</tt> class keeps track
3615of a list of <a href="#BasicBlock"><tt>BasicBlock</tt></a>s, a list of formal
3616<a href="#Argument"><tt>Argument</tt></a>s, and a
3617<a href="#SymbolTable"><tt>SymbolTable</tt></a>.</p>
3618
3619<p>The list of <a href="#BasicBlock"><tt>BasicBlock</tt></a>s is the most
3620commonly used part of <tt>Function</tt> objects. The list imposes an implicit
3621ordering of the blocks in the function, which indicate how the code will be
3622laid out by the backend. Additionally, the first <a
3623href="#BasicBlock"><tt>BasicBlock</tt></a> is the implicit entry node for the
3624<tt>Function</tt>. It is not legal in LLVM to explicitly branch to this initial
3625block. There are no implicit exit nodes, and in fact there may be multiple exit
3626nodes from a single <tt>Function</tt>. If the <a
3627href="#BasicBlock"><tt>BasicBlock</tt></a> list is empty, this indicates that
3628the <tt>Function</tt> is actually a function declaration: the actual body of the
3629function hasn't been linked in yet.</p>
3630
3631<p>In addition to a list of <a href="#BasicBlock"><tt>BasicBlock</tt></a>s, the
3632<tt>Function</tt> class also keeps track of the list of formal <a
3633href="#Argument"><tt>Argument</tt></a>s that the function receives. This
3634container manages the lifetime of the <a href="#Argument"><tt>Argument</tt></a>
3635nodes, just like the <a href="#BasicBlock"><tt>BasicBlock</tt></a> list does for
3636the <a href="#BasicBlock"><tt>BasicBlock</tt></a>s.</p>
3637
3638<p>The <a href="#SymbolTable"><tt>SymbolTable</tt></a> is a very rarely used
3639LLVM feature that is only used when you have to look up a value by name. Aside
3640from that, the <a href="#SymbolTable"><tt>SymbolTable</tt></a> is used
3641internally to make sure that there are not conflicts between the names of <a
3642href="#Instruction"><tt>Instruction</tt></a>s, <a
3643href="#BasicBlock"><tt>BasicBlock</tt></a>s, or <a
3644href="#Argument"><tt>Argument</tt></a>s in the function body.</p>
3645
3646<p>Note that <tt>Function</tt> is a <a href="#GlobalValue">GlobalValue</a>
3647and therefore also a <a href="#Constant">Constant</a>. The value of the function
3648is its address (after linking) which is guaranteed to be constant.</p>
3649</div>
3650
3651<!-- _______________________________________________________________________ -->
3652<div class="doc_subsubsection">
3653 <a name="m_Function">Important Public Members of the <tt>Function</tt>
3654 class</a>
3655</div>
3656
3657<div class="doc_text">
3658
3659<ul>
3660 <li><tt>Function(const </tt><tt><a href="#FunctionType">FunctionType</a>
3661 *Ty, LinkageTypes Linkage, const std::string &amp;N = "", Module* Parent = 0)</tt>
3662
3663 <p>Constructor used when you need to create new <tt>Function</tt>s to add
3664 the the program. The constructor must specify the type of the function to
3665 create and what type of linkage the function should have. The <a
3666 href="#FunctionType"><tt>FunctionType</tt></a> argument
3667 specifies the formal arguments and return value for the function. The same
3668 <a href="#FunctionType"><tt>FunctionType</tt></a> value can be used to
3669 create multiple functions. The <tt>Parent</tt> argument specifies the Module
3670 in which the function is defined. If this argument is provided, the function
3671 will automatically be inserted into that module's list of
3672 functions.</p></li>
3673
3674 <li><tt>bool isDeclaration()</tt>
3675
3676 <p>Return whether or not the <tt>Function</tt> has a body defined. If the
3677 function is "external", it does not have a body, and thus must be resolved
3678 by linking with a function defined in a different translation unit.</p></li>
3679
3680 <li><tt>Function::iterator</tt> - Typedef for basic block list iterator<br>
3681 <tt>Function::const_iterator</tt> - Typedef for const_iterator.<br>
3682
3683 <tt>begin()</tt>, <tt>end()</tt>
3684 <tt>size()</tt>, <tt>empty()</tt>
3685
3686 <p>These are forwarding methods that make it easy to access the contents of
3687 a <tt>Function</tt> object's <a href="#BasicBlock"><tt>BasicBlock</tt></a>
3688 list.</p></li>
3689
3690 <li><tt>Function::BasicBlockListType &amp;getBasicBlockList()</tt>
3691
3692 <p>Returns the list of <a href="#BasicBlock"><tt>BasicBlock</tt></a>s. This
3693 is necessary to use when you need to update the list or perform a complex
3694 action that doesn't have a forwarding method.</p></li>
3695
3696 <li><tt>Function::arg_iterator</tt> - Typedef for the argument list
3697iterator<br>
3698 <tt>Function::const_arg_iterator</tt> - Typedef for const_iterator.<br>
3699
3700 <tt>arg_begin()</tt>, <tt>arg_end()</tt>
3701 <tt>arg_size()</tt>, <tt>arg_empty()</tt>
3702
3703 <p>These are forwarding methods that make it easy to access the contents of
3704 a <tt>Function</tt> object's <a href="#Argument"><tt>Argument</tt></a>
3705 list.</p></li>
3706
3707 <li><tt>Function::ArgumentListType &amp;getArgumentList()</tt>
3708
3709 <p>Returns the list of <a href="#Argument"><tt>Argument</tt></a>s. This is
3710 necessary to use when you need to update the list or perform a complex
3711 action that doesn't have a forwarding method.</p></li>
3712
3713 <li><tt><a href="#BasicBlock">BasicBlock</a> &amp;getEntryBlock()</tt>
3714
3715 <p>Returns the entry <a href="#BasicBlock"><tt>BasicBlock</tt></a> for the
3716 function. Because the entry block for the function is always the first
3717 block, this returns the first block of the <tt>Function</tt>.</p></li>
3718
3719 <li><tt><a href="#Type">Type</a> *getReturnType()</tt><br>
3720 <tt><a href="#FunctionType">FunctionType</a> *getFunctionType()</tt>
3721
3722 <p>This traverses the <a href="#Type"><tt>Type</tt></a> of the
3723 <tt>Function</tt> and returns the return type of the function, or the <a
3724 href="#FunctionType"><tt>FunctionType</tt></a> of the actual
3725 function.</p></li>
3726
3727 <li><tt><a href="#SymbolTable">SymbolTable</a> *getSymbolTable()</tt>
3728
3729 <p> Return a pointer to the <a href="#SymbolTable"><tt>SymbolTable</tt></a>
3730 for this <tt>Function</tt>.</p></li>
3731</ul>
3732
3733</div>
3734
3735<!-- ======================================================================= -->
3736<div class="doc_subsection">
3737 <a name="GlobalVariable">The <tt>GlobalVariable</tt> class</a>
3738</div>
3739
3740<div class="doc_text">
3741
3742<p><tt>#include "<a
3743href="/doxygen/GlobalVariable_8h-source.html">llvm/GlobalVariable.h</a>"</tt>
3744<br>
3745doxygen info: <a href="/doxygen/classllvm_1_1GlobalVariable.html">GlobalVariable
3746 Class</a><br>
3747Superclasses: <a href="#GlobalValue"><tt>GlobalValue</tt></a>,
3748<a href="#Constant"><tt>Constant</tt></a>,
3749<a href="#User"><tt>User</tt></a>,
3750<a href="#Value"><tt>Value</tt></a></p>
3751
3752<p>Global variables are represented with the (surprise surprise)
3753<tt>GlobalVariable</tt> class. Like functions, <tt>GlobalVariable</tt>s are also
3754subclasses of <a href="#GlobalValue"><tt>GlobalValue</tt></a>, and as such are
3755always referenced by their address (global values must live in memory, so their
3756"name" refers to their constant address). See
3757<a href="#GlobalValue"><tt>GlobalValue</tt></a> for more on this. Global
3758variables may have an initial value (which must be a
3759<a href="#Constant"><tt>Constant</tt></a>), and if they have an initializer,
3760they may be marked as "constant" themselves (indicating that their contents
3761never change at runtime).</p>
3762</div>
3763
3764<!-- _______________________________________________________________________ -->
3765<div class="doc_subsubsection">
3766 <a name="m_GlobalVariable">Important Public Members of the
3767 <tt>GlobalVariable</tt> class</a>
3768</div>
3769
3770<div class="doc_text">
3771
3772<ul>
3773 <li><tt>GlobalVariable(const </tt><tt><a href="#Type">Type</a> *Ty, bool
3774 isConstant, LinkageTypes&amp; Linkage, <a href="#Constant">Constant</a>
3775 *Initializer = 0, const std::string &amp;Name = "", Module* Parent = 0)</tt>
3776
3777 <p>Create a new global variable of the specified type. If
3778 <tt>isConstant</tt> is true then the global variable will be marked as
3779 unchanging for the program. The Linkage parameter specifies the type of
3780 linkage (internal, external, weak, linkonce, appending) for the variable.
3781 If the linkage is InternalLinkage, WeakAnyLinkage, WeakODRLinkage,
3782 LinkOnceAnyLinkage or LinkOnceODRLinkage,&nbsp; then the resultant
3783 global variable will have internal linkage. AppendingLinkage concatenates
3784 together all instances (in different translation units) of the variable
3785 into a single variable but is only applicable to arrays. &nbsp;See
3786 the <a href="LangRef.html#modulestructure">LLVM Language Reference</a> for
3787 further details on linkage types. Optionally an initializer, a name, and the
3788 module to put the variable into may be specified for the global variable as
3789 well.</p></li>
3790
3791 <li><tt>bool isConstant() const</tt>
3792
3793 <p>Returns true if this is a global variable that is known not to
3794 be modified at runtime.</p></li>
3795
3796 <li><tt>bool hasInitializer()</tt>
3797
3798 <p>Returns true if this <tt>GlobalVariable</tt> has an intializer.</p></li>
3799
3800 <li><tt><a href="#Constant">Constant</a> *getInitializer()</tt>
3801
3802 <p>Returns the initial value for a <tt>GlobalVariable</tt>. It is not legal
3803 to call this method if there is no initializer.</p></li>
3804</ul>
3805
3806</div>
3807
3808
3809<!-- ======================================================================= -->
3810<div class="doc_subsection">
3811 <a name="BasicBlock">The <tt>BasicBlock</tt> class</a>
3812</div>
3813
3814<div class="doc_text">
3815
3816<p><tt>#include "<a
3817href="/doxygen/BasicBlock_8h-source.html">llvm/BasicBlock.h</a>"</tt><br>
3818doxygen info: <a href="/doxygen/classllvm_1_1BasicBlock.html">BasicBlock
3819Class</a><br>
3820Superclass: <a href="#Value"><tt>Value</tt></a></p>
3821
3822<p>This class represents a single entry multiple exit section of the code,
3823commonly known as a basic block by the compiler community. The
3824<tt>BasicBlock</tt> class maintains a list of <a
3825href="#Instruction"><tt>Instruction</tt></a>s, which form the body of the block.
3826Matching the language definition, the last element of this list of instructions
3827is always a terminator instruction (a subclass of the <a
3828href="#TerminatorInst"><tt>TerminatorInst</tt></a> class).</p>
3829
3830<p>In addition to tracking the list of instructions that make up the block, the
3831<tt>BasicBlock</tt> class also keeps track of the <a
3832href="#Function"><tt>Function</tt></a> that it is embedded into.</p>
3833
3834<p>Note that <tt>BasicBlock</tt>s themselves are <a
3835href="#Value"><tt>Value</tt></a>s, because they are referenced by instructions
3836like branches and can go in the switch tables. <tt>BasicBlock</tt>s have type
3837<tt>label</tt>.</p>
3838
3839</div>
3840
3841<!-- _______________________________________________________________________ -->
3842<div class="doc_subsubsection">
3843 <a name="m_BasicBlock">Important Public Members of the <tt>BasicBlock</tt>
3844 class</a>
3845</div>
3846
3847<div class="doc_text">
3848<ul>
3849
3850<li><tt>BasicBlock(const std::string &amp;Name = "", </tt><tt><a
3851 href="#Function">Function</a> *Parent = 0)</tt>
3852
3853<p>The <tt>BasicBlock</tt> constructor is used to create new basic blocks for
3854insertion into a function. The constructor optionally takes a name for the new
3855block, and a <a href="#Function"><tt>Function</tt></a> to insert it into. If
3856the <tt>Parent</tt> parameter is specified, the new <tt>BasicBlock</tt> is
3857automatically inserted at the end of the specified <a
3858href="#Function"><tt>Function</tt></a>, if not specified, the BasicBlock must be
3859manually inserted into the <a href="#Function"><tt>Function</tt></a>.</p></li>
3860
3861<li><tt>BasicBlock::iterator</tt> - Typedef for instruction list iterator<br>
3862<tt>BasicBlock::const_iterator</tt> - Typedef for const_iterator.<br>
3863<tt>begin()</tt>, <tt>end()</tt>, <tt>front()</tt>, <tt>back()</tt>,
3864<tt>size()</tt>, <tt>empty()</tt>
3865STL-style functions for accessing the instruction list.
3866
3867<p>These methods and typedefs are forwarding functions that have the same
3868semantics as the standard library methods of the same names. These methods
3869expose the underlying instruction list of a basic block in a way that is easy to
3870manipulate. To get the full complement of container operations (including
3871operations to update the list), you must use the <tt>getInstList()</tt>
3872method.</p></li>
3873
3874<li><tt>BasicBlock::InstListType &amp;getInstList()</tt>
3875
3876<p>This method is used to get access to the underlying container that actually
3877holds the Instructions. This method must be used when there isn't a forwarding
3878function in the <tt>BasicBlock</tt> class for the operation that you would like
3879to perform. Because there are no forwarding functions for "updating"
3880operations, you need to use this if you want to update the contents of a
3881<tt>BasicBlock</tt>.</p></li>
3882
3883<li><tt><a href="#Function">Function</a> *getParent()</tt>
3884
3885<p> Returns a pointer to <a href="#Function"><tt>Function</tt></a> the block is
3886embedded into, or a null pointer if it is homeless.</p></li>
3887
3888<li><tt><a href="#TerminatorInst">TerminatorInst</a> *getTerminator()</tt>
3889
3890<p> Returns a pointer to the terminator instruction that appears at the end of
3891the <tt>BasicBlock</tt>. If there is no terminator instruction, or if the last
3892instruction in the block is not a terminator, then a null pointer is
3893returned.</p></li>
3894
3895</ul>
3896
3897</div>
3898
3899
3900<!-- ======================================================================= -->
3901<div class="doc_subsection">
3902 <a name="Argument">The <tt>Argument</tt> class</a>
3903</div>
3904
3905<div class="doc_text">
3906
3907<p>This subclass of Value defines the interface for incoming formal
3908arguments to a function. A Function maintains a list of its formal
3909arguments. An argument has a pointer to the parent Function.</p>
3910
3911</div>
3912
3913<!-- *********************************************************************** -->
3914<hr>
3915<address>
3916 <a href="http://jigsaw.w3.org/css-validator/check/referer"><img
3917 src="http://jigsaw.w3.org/css-validator/images/vcss-blue" alt="Valid CSS"></a>
3918 <a href="http://validator.w3.org/check/referer"><img
3919 src="http://www.w3.org/Icons/valid-html401" alt="Valid HTML 4.01 Strict"></a>
3920
3921 <a href="mailto:dhurjati@cs.uiuc.edu">Dinakar Dhurjati</a> and
3922 <a href="mailto:sabre@nondot.org">Chris Lattner</a><br>
3923 <a href="http://llvm.org">The LLVM Compiler Infrastructure</a><br>
3924 Last modified: $Date: 2010-01-29 11:10:38 -0800 (Fri, 29 Jan 2010) $
3925</address>
3926
3927</body>
3928</html>