1 <!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.01//EN"
2 "http://www.w3.org/TR/html4/strict.dtd">
5 <meta http-equiv="Content-type" content="text/html;charset=UTF-8">
6 <title>LLVM Programmer's Manual</title>
7 <link rel="stylesheet" href="llvm.css" type="text/css">
11 <div class="doc_title">
12 LLVM Programmer's Manual
16 <li><a href="#introduction">Introduction</a></li>
17 <li><a href="#general">General Information</a>
19 <li><a href="#stl">The C++ Standard Template Library</a></li>
21 <li>The <tt>-time-passes</tt> option</li>
22 <li>How to use the LLVM Makefile system</li>
23 <li>How to write a regression test</li>
28 <li><a href="#apis">Important and useful LLVM APIs</a>
30 <li><a href="#isa">The <tt>isa<></tt>, <tt>cast<></tt>
31 and <tt>dyn_cast<></tt> templates</a> </li>
32 <li><a href="#string_apis">Passing strings (the <tt>StringRef</tt>
33 and <tt>Twine</tt> classes)</a>
35 <li><a href="#StringRef">The <tt>StringRef</tt> class</a> </li>
36 <li><a href="#Twine">The <tt>Twine</tt> class</a> </li>
39 <li><a href="#DEBUG">The <tt>DEBUG()</tt> macro and <tt>-debug</tt>
42 <li><a href="#DEBUG_TYPE">Fine grained debug info with <tt>DEBUG_TYPE</tt>
43 and the <tt>-debug-only</tt> option</a> </li>
46 <li><a href="#Statistic">The <tt>Statistic</tt> class & <tt>-stats</tt>
49 <li>The <tt>InstVisitor</tt> template
50 <li>The general graph API
52 <li><a href="#ViewGraph">Viewing graphs while debugging code</a></li>
55 <li><a href="#datastructure">Picking the Right Data Structure for a Task</a>
57 <li><a href="#ds_sequential">Sequential Containers (std::vector, std::list, etc)</a>
59 <li><a href="#dss_fixedarrays">Fixed Size Arrays</a></li>
60 <li><a href="#dss_heaparrays">Heap Allocated Arrays</a></li>
61 <li><a href="#dss_smallvector">"llvm/ADT/SmallVector.h"</a></li>
62 <li><a href="#dss_vector"><vector></a></li>
63 <li><a href="#dss_deque"><deque></a></li>
64 <li><a href="#dss_list"><list></a></li>
65 <li><a href="#dss_ilist">llvm/ADT/ilist.h</a></li>
66 <li><a href="#dss_other">Other Sequential Container Options</a></li>
68 <li><a href="#ds_set">Set-Like Containers (std::set, SmallSet, SetVector, etc)</a>
70 <li><a href="#dss_sortedvectorset">A sorted 'vector'</a></li>
71 <li><a href="#dss_smallset">"llvm/ADT/SmallSet.h"</a></li>
72 <li><a href="#dss_smallptrset">"llvm/ADT/SmallPtrSet.h"</a></li>
73 <li><a href="#dss_denseset">"llvm/ADT/DenseSet.h"</a></li>
74 <li><a href="#dss_FoldingSet">"llvm/ADT/FoldingSet.h"</a></li>
75 <li><a href="#dss_set"><set></a></li>
76 <li><a href="#dss_setvector">"llvm/ADT/SetVector.h"</a></li>
77 <li><a href="#dss_uniquevector">"llvm/ADT/UniqueVector.h"</a></li>
78 <li><a href="#dss_otherset">Other Set-Like ContainerOptions</a></li>
80 <li><a href="#ds_map">Map-Like Containers (std::map, DenseMap, etc)</a>
82 <li><a href="#dss_sortedvectormap">A sorted 'vector'</a></li>
83 <li><a href="#dss_stringmap">"llvm/ADT/StringMap.h"</a></li>
84 <li><a href="#dss_indexedmap">"llvm/ADT/IndexedMap.h"</a></li>
85 <li><a href="#dss_densemap">"llvm/ADT/DenseMap.h"</a></li>
86 <li><a href="#dss_valuemap">"llvm/ADT/ValueMap.h"</a></li>
87 <li><a href="#dss_map"><map></a></li>
88 <li><a href="#dss_othermap">Other Map-Like Container Options</a></li>
90 <li><a href="#ds_string">String-like containers</a>
94 <li><a href="#ds_bit">BitVector-like containers</a>
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>
102 <li><a href="#common">Helpful Hints for Common Operations</a>
104 <li><a href="#inspection">Basic Inspection and Traversal Routines</a>
106 <li><a href="#iterate_function">Iterating over the <tt>BasicBlock</tt>s
107 in a <tt>Function</tt></a> </li>
108 <li><a href="#iterate_basicblock">Iterating over the <tt>Instruction</tt>s
109 in a <tt>BasicBlock</tt></a> </li>
110 <li><a href="#iterate_institer">Iterating over the <tt>Instruction</tt>s
111 in a <tt>Function</tt></a> </li>
112 <li><a href="#iterate_convert">Turning an iterator into a
113 class pointer</a> </li>
114 <li><a href="#iterate_complex">Finding call sites: a more
115 complex example</a> </li>
116 <li><a href="#calls_and_invokes">Treating calls and invokes
117 the same way</a> </li>
118 <li><a href="#iterate_chains">Iterating over def-use &
119 use-def chains</a> </li>
120 <li><a href="#iterate_preds">Iterating over predecessors &
121 successors of blocks</a></li>
124 <li><a href="#simplechanges">Making simple changes</a>
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>
130 with another <tt>Value</tt></a> </li>
131 <li><a href="#schanges_deletingGV">Deleting <tt>GlobalVariable</tt>s</a> </li>
134 <li><a href="#create_types">How to Create Types</a></li>
136 <li>Working with the Control Flow Graph
138 <li>Accessing predecessors and successors of a <tt>BasicBlock</tt>
146 <li><a href="#threading">Threads and LLVM</a>
148 <li><a href="#startmultithreaded">Entering and Exiting Multithreaded Mode
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>
157 <li><a href="#advanced">Advanced Topics</a>
159 <li><a href="#TypeResolve">LLVM Type Resolution</a>
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>
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>
171 <li><a href="#coreclasses">The Core LLVM Class Hierarchy Reference</a>
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>
177 <li><a href="#User">The <tt>User</tt> class</a>
179 <li><a href="#Instruction">The <tt>Instruction</tt> class</a></li>
180 <li><a href="#Constant">The <tt>Constant</tt> class</a>
182 <li><a href="#GlobalValue">The <tt>GlobalValue</tt> class</a>
184 <li><a href="#Function">The <tt>Function</tt> class</a></li>
185 <li><a href="#GlobalVariable">The <tt>GlobalVariable</tt> class</a></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>
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>
209 <!-- *********************************************************************** -->
210 <div class="doc_section">
211 <a name="introduction">Introduction </a>
213 <!-- *********************************************************************** -->
215 <div class="doc_text">
217 <p>This document is meant to highlight some of the important classes and
218 interfaces available in the LLVM source-base. This manual is not
219 intended to explain what LLVM is, how it works, and what LLVM code looks
220 like. It assumes that you know the basics of LLVM and are interested
221 in writing transformations or otherwise analyzing or manipulating the
224 <p>This document should get you oriented so that you can find your
225 way in the continuously growing source code that makes up the LLVM
226 infrastructure. Note that this manual is not intended to serve as a
227 replacement for reading the source code, so if you think there should be
228 a method in one of these classes to do something, but it's not listed,
229 check the source. Links to the <a href="/doxygen/">doxygen</a> sources
230 are provided to make this as easy as possible.</p>
232 <p>The first section of this document describes general information that is
233 useful to know when working in the LLVM infrastructure, and the second describes
234 the Core LLVM classes. In the future this manual will be extended with
235 information describing how to use extension libraries, such as dominator
236 information, CFG traversal routines, and useful utilities like the <tt><a
237 href="/doxygen/InstVisitor_8h-source.html">InstVisitor</a></tt> template.</p>
241 <!-- *********************************************************************** -->
242 <div class="doc_section">
243 <a name="general">General Information</a>
245 <!-- *********************************************************************** -->
247 <div class="doc_text">
249 <p>This section contains general information that is useful if you are working
250 in the LLVM source-base, but that isn't specific to any particular API.</p>
254 <!-- ======================================================================= -->
255 <div class="doc_subsection">
256 <a name="stl">The C++ Standard Template Library</a>
259 <div class="doc_text">
261 <p>LLVM makes heavy use of the C++ Standard Template Library (STL),
262 perhaps much more than you are used to, or have seen before. Because of
263 this, you might want to do a little background reading in the
264 techniques used and capabilities of the library. There are many good
265 pages that discuss the STL, and several books on the subject that you
266 can get, so it will not be discussed in this document.</p>
268 <p>Here are some useful links:</p>
272 <li><a href="http://www.dinkumware.com/refxcpp.html">Dinkumware C++ Library
273 reference</a> - an excellent reference for the STL and other parts of the
274 standard C++ library.</li>
276 <li><a href="http://www.tempest-sw.com/cpp/">C++ In a Nutshell</a> - This is an
277 O'Reilly book in the making. It has a decent Standard Library
278 Reference that rivals Dinkumware's, and is unfortunately no longer free since the
279 book has been published.</li>
281 <li><a href="http://www.parashift.com/c++-faq-lite/">C++ Frequently Asked
284 <li><a href="http://www.sgi.com/tech/stl/">SGI's STL Programmer's Guide</a> -
286 href="http://www.sgi.com/tech/stl/stl_introduction.html">Introduction to the
289 <li><a href="http://www.research.att.com/%7Ebs/C++.html">Bjarne Stroustrup's C++
292 <li><a href="http://64.78.49.204/">
293 Bruce Eckel's Thinking in C++, 2nd ed. Volume 2 Revision 4.0 (even better, get
298 <p>You are also encouraged to take a look at the <a
299 href="CodingStandards.html">LLVM Coding Standards</a> guide which focuses on how
300 to write maintainable code more than where to put your curly braces.</p>
304 <!-- ======================================================================= -->
305 <div class="doc_subsection">
306 <a name="stl">Other useful references</a>
309 <div class="doc_text">
312 <li><a href="http://www.psc.edu/%7Esemke/cvs_branches.html">CVS
313 Branch and Tag Primer</a></li>
314 <li><a href="http://www.fortran-2000.com/ArnaudRecipes/sharedlib.html">Using
315 static and shared libraries across platforms</a></li>
320 <!-- *********************************************************************** -->
321 <div class="doc_section">
322 <a name="apis">Important and useful LLVM APIs</a>
324 <!-- *********************************************************************** -->
326 <div class="doc_text">
328 <p>Here we highlight some LLVM APIs that are generally useful and good to
329 know about when writing transformations.</p>
333 <!-- ======================================================================= -->
334 <div class="doc_subsection">
335 <a name="isa">The <tt>isa<></tt>, <tt>cast<></tt> and
336 <tt>dyn_cast<></tt> templates</a>
339 <div class="doc_text">
341 <p>The LLVM source-base makes extensive use of a custom form of RTTI.
342 These templates have many similarities to the C++ <tt>dynamic_cast<></tt>
343 operator, but they don't have some drawbacks (primarily stemming from
344 the fact that <tt>dynamic_cast<></tt> only works on classes that
345 have a v-table). Because they are used so often, you must know what they
346 do 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>
348 file (note that you very rarely have to include this file directly).</p>
351 <dt><tt>isa<></tt>: </dt>
353 <dd><p>The <tt>isa<></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>
359 <dt><tt>cast<></tt>: </dt>
361 <dd><p>The <tt>cast<></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<></tt>
366 and <tt>cast<></tt> template is:</p>
368 <div class="doc_code">
370 static bool isLoopInvariant(const <a href="#Value">Value</a> *V, const Loop *L) {
371 if (isa<<a href="#Constant">Constant</a>>(V) || isa<<a href="#Argument">Argument</a>>(V) || isa<<a href="#GlobalValue">GlobalValue</a>>(V))
374 // <i>Otherwise, it must be an instruction...</i>
375 return !L->contains(cast<<a href="#Instruction">Instruction</a>>(V)->getParent());
380 <p>Note that you should <b>not</b> use an <tt>isa<></tt> test followed
381 by a <tt>cast<></tt>, for that use the <tt>dyn_cast<></tt>
386 <dt><tt>dyn_cast<></tt>:</dt>
388 <dd><p>The <tt>dyn_cast<></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<></tt> operator in C++, and should be
393 used in the same circumstances. Typically, the <tt>dyn_cast<></tt>
394 operator is used in an <tt>if</tt> statement or some other flow control
395 statement like this:</p>
397 <div class="doc_code">
399 if (<a href="#AllocationInst">AllocationInst</a> *AI = dyn_cast<<a href="#AllocationInst">AllocationInst</a>>(Val)) {
405 <p>This form of the <tt>if</tt> statement effectively combines together a call
406 to <tt>isa<></tt> and a call to <tt>cast<></tt> into one
407 statement, which is very convenient.</p>
409 <p>Note that the <tt>dyn_cast<></tt> operator, like C++'s
410 <tt>dynamic_cast<></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>
418 <dt><tt>cast_or_null<></tt>: </dt>
420 <dd><p>The <tt>cast_or_null<></tt> operator works just like the
421 <tt>cast<></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>
425 <dt><tt>dyn_cast_or_null<></tt>: </dt>
427 <dd><p>The <tt>dyn_cast_or_null<></tt> operator works just like the
428 <tt>dyn_cast<></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>
434 <p>These five templates can be used with any classes, whether they have a
435 v-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
437 to. Describing this is currently outside the scope of this document, but there
438 are lots of examples in the LLVM source base.</p>
443 <!-- ======================================================================= -->
444 <div class="doc_subsection">
445 <a name="string_apis">Passing strings (the <tt>StringRef</tt>
446 and <tt>Twine</tt> classes)</a>
449 <div class="doc_text">
451 <p>Although LLVM generally does not do much string manipulation, we do have
452 several important APIs which take strings. Two important examples are the
453 Value class -- which has names for instructions, functions, etc. -- and the
454 StringMap class which is used extensively in LLVM and Clang.</p>
456 <p>These are generic classes, and they need to be able to accept strings which
457 may have embedded null characters. Therefore, they cannot simply take
458 a <tt>const char *</tt>, and taking a <tt>const std::string&</tt> requires
459 clients to perform a heap allocation which is usually unnecessary. Instead,
460 many LLVM APIs use a <tt>StringRef</tt> or a <tt>const Twine&</tt> for
461 passing strings efficiently.</p>
465 <!-- _______________________________________________________________________ -->
466 <div class="doc_subsubsection">
467 <a name="StringRef">The <tt>StringRef</tt> class</a>
470 <div class="doc_text">
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
474 on <tt>std:string</tt>, but does not require heap allocation.</p>
476 <p>It can be implicitly constructed using a C style null-terminated string,
477 an <tt>std::string</tt>, or explicitly with a character pointer and length.
478 For example, the <tt>StringRef</tt> find function is declared as:</p>
480 <pre class="doc_code">
481 iterator find(StringRef Key);
484 <p>and clients can call it using any one of:</p>
486 <pre class="doc_code">
487 Map.find("foo"); <i>// Lookup "foo"</i>
488 Map.find(std::string("bar")); <i>// Lookup "bar"</i>
489 Map.find(StringRef("\0baz", 4)); <i>// Lookup "\0baz"</i>
492 <p>Similarly, APIs which need to return a string may return a <tt>StringRef</tt>
493 instance, which can be used directly or converted to an <tt>std::string</tt>
494 using the <tt>str</tt> member function. See
495 "<tt><a href="/doxygen/classllvm_1_1StringRef_8h-source.html">llvm/ADT/StringRef.h</a></tt>"
496 for more information.</p>
498 <p>You should rarely use the <tt>StringRef</tt> class directly, because it contains
499 pointers to external memory it is not generally safe to store an instance of the
500 class (unless you know that the external storage will not be freed). StringRef is
501 small and pervasive enough in LLVM that it should always be passed by value.</p>
505 <!-- _______________________________________________________________________ -->
506 <div class="doc_subsubsection">
507 <a name="Twine">The <tt>Twine</tt> class</a>
510 <div class="doc_text">
512 <p>The <tt>Twine</tt> class is an efficient way for APIs to accept concatenated
513 strings. For example, a common LLVM paradigm is to name one instruction based on
514 the name of another instruction with a suffix, for example:</p>
516 <div class="doc_code">
518 New = CmpInst::Create(<i>...</i>, SO->getName() + ".cmp");
522 <p>The <tt>Twine</tt> class is effectively a
523 lightweight <a href="http://en.wikipedia.org/wiki/Rope_(computer_science)">rope</a>
524 which points to temporary (stack allocated) objects. Twines can be implicitly
525 constructed as the result of the plus operator applied to strings (i.e., a C
526 strings, an <tt>std::string</tt>, or a <tt>StringRef</tt>). The twine delays the
527 actual concatenation of strings until it is actually required, at which point
528 it can be efficiently rendered directly into a character array. This avoids
529 unnecessary heap allocation involved in constructing the temporary results of
530 string concatenation. See
531 "<tt><a href="/doxygen/classllvm_1_1Twine_8h-source.html">llvm/ADT/Twine.h</a></tt>"
532 for more information.</p>
534 <p>As with a <tt>StringRef</tt>, <tt>Twine</tt> objects point to external memory
535 and should almost never be stored or mentioned directly. They are intended
536 solely for use when defining a function which should be able to efficiently
537 accept concatenated strings.</p>
542 <!-- ======================================================================= -->
543 <div class="doc_subsection">
544 <a name="DEBUG">The <tt>DEBUG()</tt> macro and <tt>-debug</tt> option</a>
547 <div class="doc_text">
549 <p>Often when working on your pass you will put a bunch of debugging printouts
550 and other code into your pass. After you get it working, you want to remove
551 it, but you may need it again in the future (to work out new bugs that you run
554 <p> Naturally, because of this, you don't want to delete the debug printouts,
555 but you don't want them to always be noisy. A standard compromise is to comment
556 them out, allowing you to enable them if you need them in the future.</p>
558 <p>The "<tt><a href="/doxygen/Debug_8h-source.html">llvm/Support/Debug.h</a></tt>"
559 file provides a macro named <tt>DEBUG()</tt> that is a much nicer solution to
560 this problem. Basically, you can put arbitrary code into the argument of the
561 <tt>DEBUG</tt> macro, and it is only executed if '<tt>opt</tt>' (or any other
562 tool) is run with the '<tt>-debug</tt>' command line argument:</p>
564 <div class="doc_code">
566 DEBUG(errs() << "I am here!\n");
570 <p>Then you can run your pass like this:</p>
572 <div class="doc_code">
574 $ opt < a.bc > /dev/null -mypass
575 <i><no output></i>
576 $ opt < a.bc > /dev/null -mypass -debug
581 <p>Using the <tt>DEBUG()</tt> macro instead of a home-brewed solution allows you
582 to not have to create "yet another" command line option for the debug output for
583 your pass. Note that <tt>DEBUG()</tt> macros are disabled for optimized builds,
584 so they do not cause a performance impact at all (for the same reason, they
585 should also not contain side-effects!).</p>
587 <p>One additional nice thing about the <tt>DEBUG()</tt> macro is that you can
588 enable or disable it directly in gdb. Just use "<tt>set DebugFlag=0</tt>" or
589 "<tt>set DebugFlag=1</tt>" from the gdb if the program is running. If the
590 program hasn't been started yet, you can always just run it with
595 <!-- _______________________________________________________________________ -->
596 <div class="doc_subsubsection">
597 <a name="DEBUG_TYPE">Fine grained debug info with <tt>DEBUG_TYPE</tt> and
598 the <tt>-debug-only</tt> option</a>
601 <div class="doc_text">
603 <p>Sometimes you may find yourself in a situation where enabling <tt>-debug</tt>
604 just turns on <b>too much</b> information (such as when working on the code
605 generator). If you want to enable debug information with more fine-grained
606 control, you define the <tt>DEBUG_TYPE</tt> macro and the <tt>-debug</tt> only
607 option as follows:</p>
609 <div class="doc_code">
612 DEBUG(errs() << "No debug type\n");
613 #define DEBUG_TYPE "foo"
614 DEBUG(errs() << "'foo' debug type\n");
616 #define DEBUG_TYPE "bar"
617 DEBUG(errs() << "'bar' debug type\n"));
619 #define DEBUG_TYPE ""
620 DEBUG(errs() << "No debug type (2)\n");
624 <p>Then you can run your pass like this:</p>
626 <div class="doc_code">
628 $ opt < a.bc > /dev/null -mypass
629 <i><no output></i>
630 $ opt < a.bc > /dev/null -mypass -debug
635 $ opt < a.bc > /dev/null -mypass -debug-only=foo
637 $ opt < a.bc > /dev/null -mypass -debug-only=bar
642 <p>Of course, in practice, you should only set <tt>DEBUG_TYPE</tt> at the top of
643 a file, to specify the debug type for the entire module (if you do this before
644 you <tt>#include "llvm/Support/Debug.h"</tt>, you don't have to insert the ugly
645 <tt>#undef</tt>'s). Also, you should use names more meaningful than "foo" and
646 "bar", because there is no system in place to ensure that names do not
647 conflict. If two different modules use the same string, they will all be turned
648 on when the name is specified. This allows, for example, all debug information
649 for instruction scheduling to be enabled with <tt>-debug-type=InstrSched</tt>,
650 even if the source lives in multiple files.</p>
652 <p>The <tt>DEBUG_WITH_TYPE</tt> macro is also available for situations where you
653 would like to set <tt>DEBUG_TYPE</tt>, but only for one specific <tt>DEBUG</tt>
654 statement. It takes an additional first parameter, which is the type to use. For
655 example, the preceding example could be written as:</p>
658 <div class="doc_code">
660 DEBUG_WITH_TYPE("", errs() << "No debug type\n");
661 DEBUG_WITH_TYPE("foo", errs() << "'foo' debug type\n");
662 DEBUG_WITH_TYPE("bar", errs() << "'bar' debug type\n"));
663 DEBUG_WITH_TYPE("", errs() << "No debug type (2)\n");
669 <!-- ======================================================================= -->
670 <div class="doc_subsection">
671 <a name="Statistic">The <tt>Statistic</tt> class & <tt>-stats</tt>
675 <div class="doc_text">
678 href="/doxygen/Statistic_8h-source.html">llvm/ADT/Statistic.h</a></tt>" file
679 provides a class named <tt>Statistic</tt> that is used as a unified way to
680 keep track of what the LLVM compiler is doing and how effective various
681 optimizations are. It is useful to see what optimizations are contributing to
682 making a particular program run faster.</p>
684 <p>Often you may run your pass on some big program, and you're interested to see
685 how many times it makes a certain transformation. Although you can do this with
686 hand inspection, or some ad-hoc method, this is a real pain and not very useful
687 for big programs. Using the <tt>Statistic</tt> class makes it very easy to
688 keep track of this information, and the calculated information is presented in a
689 uniform manner with the rest of the passes being executed.</p>
691 <p>There are many examples of <tt>Statistic</tt> uses, but the basics of using
692 it are as follows:</p>
695 <li><p>Define your statistic like this:</p>
697 <div class="doc_code">
699 #define <a href="#DEBUG_TYPE">DEBUG_TYPE</a> "mypassname" <i>// This goes before any #includes.</i>
700 STATISTIC(NumXForms, "The # of times I did stuff");
704 <p>The <tt>STATISTIC</tt> macro defines a static variable, whose name is
705 specified by the first argument. The pass name is taken from the DEBUG_TYPE
706 macro, and the description is taken from the second argument. The variable
707 defined ("NumXForms" in this case) acts like an unsigned integer.</p></li>
709 <li><p>Whenever you make a transformation, bump the counter:</p>
711 <div class="doc_code">
713 ++NumXForms; // <i>I did stuff!</i>
720 <p>That's all you have to do. To get '<tt>opt</tt>' to print out the
721 statistics gathered, use the '<tt>-stats</tt>' option:</p>
723 <div class="doc_code">
725 $ opt -stats -mypassname < program.bc > /dev/null
726 <i>... statistics output ...</i>
730 <p> When running <tt>opt</tt> on a C file from the SPEC benchmark
731 suite, it gives a report that looks like this:</p>
733 <div class="doc_code">
735 7646 bitcodewriter - Number of normal instructions
736 725 bitcodewriter - Number of oversized instructions
737 129996 bitcodewriter - Number of bitcode bytes written
738 2817 raise - Number of insts DCEd or constprop'd
739 3213 raise - Number of cast-of-self removed
740 5046 raise - Number of expression trees converted
741 75 raise - Number of other getelementptr's formed
742 138 raise - Number of load/store peepholes
743 42 deadtypeelim - Number of unused typenames removed from symtab
744 392 funcresolve - Number of varargs functions resolved
745 27 globaldce - Number of global variables removed
746 2 adce - Number of basic blocks removed
747 134 cee - Number of branches revectored
748 49 cee - Number of setcc instruction eliminated
749 532 gcse - Number of loads removed
750 2919 gcse - Number of instructions removed
751 86 indvars - Number of canonical indvars added
752 87 indvars - Number of aux indvars removed
753 25 instcombine - Number of dead inst eliminate
754 434 instcombine - Number of insts combined
755 248 licm - Number of load insts hoisted
756 1298 licm - Number of insts hoisted to a loop pre-header
757 3 licm - Number of insts hoisted to multiple loop preds (bad, no loop pre-header)
758 75 mem2reg - Number of alloca's promoted
759 1444 cfgsimplify - Number of blocks simplified
763 <p>Obviously, with so many optimizations, having a unified framework for this
764 stuff is very nice. Making your pass fit well into the framework makes it more
765 maintainable and useful.</p>
769 <!-- ======================================================================= -->
770 <div class="doc_subsection">
771 <a name="ViewGraph">Viewing graphs while debugging code</a>
774 <div class="doc_text">
776 <p>Several of the important data structures in LLVM are graphs: for example
777 CFGs made out of LLVM <a href="#BasicBlock">BasicBlock</a>s, CFGs made out of
778 LLVM <a href="CodeGenerator.html#machinebasicblock">MachineBasicBlock</a>s, and
779 <a href="CodeGenerator.html#selectiondag_intro">Instruction Selection
780 DAGs</a>. In many cases, while debugging various parts of the compiler, it is
781 nice to instantly visualize these graphs.</p>
783 <p>LLVM provides several callbacks that are available in a debug build to do
784 exactly that. If you call the <tt>Function::viewCFG()</tt> method, for example,
785 the current LLVM tool will pop up a window containing the CFG for the function
786 where each basic block is a node in the graph, and each node contains the
787 instructions in the block. Similarly, there also exists
788 <tt>Function::viewCFGOnly()</tt> (does not include the instructions), the
789 <tt>MachineFunction::viewCFG()</tt> and <tt>MachineFunction::viewCFGOnly()</tt>,
790 and the <tt>SelectionDAG::viewGraph()</tt> methods. Within GDB, for example,
791 you can usually use something like <tt>call DAG.viewGraph()</tt> to pop
792 up a window. Alternatively, you can sprinkle calls to these functions in your
793 code in places you want to debug.</p>
795 <p>Getting this to work requires a small amount of configuration. On Unix
796 systems with X11, install the <a href="http://www.graphviz.org">graphviz</a>
797 toolkit, and make sure 'dot' and 'gv' are in your path. If you are running on
798 Mac OS/X, download and install the Mac OS/X <a
799 href="http://www.pixelglow.com/graphviz/">Graphviz program</a>, and add
800 <tt>/Applications/Graphviz.app/Contents/MacOS/</tt> (or wherever you install
801 it) to your path. Once in your system and path are set up, rerun the LLVM
802 configure script and rebuild LLVM to enable this functionality.</p>
804 <p><tt>SelectionDAG</tt> has been extended to make it easier to locate
805 <i>interesting</i> nodes in large complex graphs. From gdb, if you
806 <tt>call DAG.setGraphColor(<i>node</i>, "<i>color</i>")</tt>, then the
807 next <tt>call DAG.viewGraph()</tt> would highlight the node in the
808 specified color (choices of colors can be found at <a
809 href="http://www.graphviz.org/doc/info/colors.html">colors</a>.) More
810 complex node attributes can be provided with <tt>call
811 DAG.setGraphAttrs(<i>node</i>, "<i>attributes</i>")</tt> (choices can be
812 found at <a href="http://www.graphviz.org/doc/info/attrs.html">Graph
813 Attributes</a>.) If you want to restart and clear all the current graph
814 attributes, then you can <tt>call DAG.clearGraphAttrs()</tt>. </p>
818 <!-- *********************************************************************** -->
819 <div class="doc_section">
820 <a name="datastructure">Picking the Right Data Structure for a Task</a>
822 <!-- *********************************************************************** -->
824 <div class="doc_text">
826 <p>LLVM has a plethora of data structures in the <tt>llvm/ADT/</tt> directory,
827 and we commonly use STL data structures. This section describes the trade-offs
828 you should consider when you pick one.</p>
831 The first step is a choose your own adventure: do you want a sequential
832 container, a set-like container, or a map-like container? The most important
833 thing when choosing a container is the algorithmic properties of how you plan to
834 access the container. Based on that, you should use:</p>
837 <li>a <a href="#ds_map">map-like</a> container if you need efficient look-up
838 of an value based on another value. Map-like containers also support
839 efficient queries for containment (whether a key is in the map). Map-like
840 containers generally do not support efficient reverse mapping (values to
841 keys). If you need that, use two maps. Some map-like containers also
842 support efficient iteration through the keys in sorted order. Map-like
843 containers are the most expensive sort, only use them if you need one of
844 these capabilities.</li>
846 <li>a <a href="#ds_set">set-like</a> container if you need to put a bunch of
847 stuff into a container that automatically eliminates duplicates. Some
848 set-like containers support efficient iteration through the elements in
849 sorted order. Set-like containers are more expensive than sequential
853 <li>a <a href="#ds_sequential">sequential</a> container provides
854 the most efficient way to add elements and keeps track of the order they are
855 added to the collection. They permit duplicates and support efficient
856 iteration, but do not support efficient look-up based on a key.
859 <li>a <a href="#ds_string">string</a> container is a specialized sequential
860 container or reference structure that is used for character or byte
863 <li>a <a href="#ds_bit">bit</a> container provides an efficient way to store and
864 perform set operations on sets of numeric id's, while automatically
865 eliminating duplicates. Bit containers require a maximum of 1 bit for each
866 identifier you want to store.
871 Once the proper category of container is determined, you can fine tune the
872 memory use, constant factors, and cache behaviors of access by intelligently
873 picking a member of the category. Note that constant factors and cache behavior
874 can be a big deal. If you have a vector that usually only contains a few
875 elements (but could contain many), for example, it's much better to use
876 <a href="#dss_smallvector">SmallVector</a> than <a href="#dss_vector">vector</a>
877 . Doing so avoids (relatively) expensive malloc/free calls, which dwarf the
878 cost of adding the elements to the container. </p>
882 <!-- ======================================================================= -->
883 <div class="doc_subsection">
884 <a name="ds_sequential">Sequential Containers (std::vector, std::list, etc)</a>
887 <div class="doc_text">
888 There are a variety of sequential containers available for you, based on your
889 needs. Pick the first in this section that will do what you want.
892 <!-- _______________________________________________________________________ -->
893 <div class="doc_subsubsection">
894 <a name="dss_fixedarrays">Fixed Size Arrays</a>
897 <div class="doc_text">
898 <p>Fixed size arrays are very simple and very fast. They are good if you know
899 exactly how many elements you have, or you have a (low) upper bound on how many
903 <!-- _______________________________________________________________________ -->
904 <div class="doc_subsubsection">
905 <a name="dss_heaparrays">Heap Allocated Arrays</a>
908 <div class="doc_text">
909 <p>Heap allocated arrays (new[] + delete[]) are also simple. They are good if
910 the number of elements is variable, if you know how many elements you will need
911 before the array is allocated, and if the array is usually large (if not,
912 consider a <a href="#dss_smallvector">SmallVector</a>). The cost of a heap
913 allocated array is the cost of the new/delete (aka malloc/free). Also note that
914 if you are allocating an array of a type with a constructor, the constructor and
915 destructors will be run for every element in the array (re-sizable vectors only
916 construct those elements actually used).</p>
919 <!-- _______________________________________________________________________ -->
920 <div class="doc_subsubsection">
921 <a name="dss_smallvector">"llvm/ADT/SmallVector.h"</a>
924 <div class="doc_text">
925 <p><tt>SmallVector<Type, N></tt> is a simple class that looks and smells
926 just like <tt>vector<Type></tt>:
927 it supports efficient iteration, lays out elements in memory order (so you can
928 do pointer arithmetic between elements), supports efficient push_back/pop_back
929 operations, supports efficient random access to its elements, etc.</p>
931 <p>The advantage of SmallVector is that it allocates space for
932 some number of elements (N) <b>in the object itself</b>. Because of this, if
933 the SmallVector is dynamically smaller than N, no malloc is performed. This can
934 be a big win in cases where the malloc/free call is far more expensive than the
935 code that fiddles around with the elements.</p>
937 <p>This is good for vectors that are "usually small" (e.g. the number of
938 predecessors/successors of a block is usually less than 8). On the other hand,
939 this makes the size of the SmallVector itself large, so you don't want to
940 allocate lots of them (doing so will waste a lot of space). As such,
941 SmallVectors are most useful when on the stack.</p>
943 <p>SmallVector also provides a nice portable and efficient replacement for
948 <!-- _______________________________________________________________________ -->
949 <div class="doc_subsubsection">
950 <a name="dss_vector"><vector></a>
953 <div class="doc_text">
955 std::vector is well loved and respected. It is useful when SmallVector isn't:
956 when the size of the vector is often large (thus the small optimization will
957 rarely be a benefit) or if you will be allocating many instances of the vector
958 itself (which would waste space for elements that aren't in the container).
959 vector is also useful when interfacing with code that expects vectors :).
962 <p>One worthwhile note about std::vector: avoid code like this:</p>
964 <div class="doc_code">
967 std::vector<foo> V;
973 <p>Instead, write this as:</p>
975 <div class="doc_code">
977 std::vector<foo> V;
985 <p>Doing so will save (at least) one heap allocation and free per iteration of
990 <!-- _______________________________________________________________________ -->
991 <div class="doc_subsubsection">
992 <a name="dss_deque"><deque></a>
995 <div class="doc_text">
996 <p>std::deque is, in some senses, a generalized version of std::vector. Like
997 std::vector, it provides constant time random access and other similar
998 properties, but it also provides efficient access to the front of the list. It
999 does not guarantee continuity of elements within memory.</p>
1001 <p>In exchange for this extra flexibility, std::deque has significantly higher
1002 constant factor costs than std::vector. If possible, use std::vector or
1003 something cheaper.</p>
1006 <!-- _______________________________________________________________________ -->
1007 <div class="doc_subsubsection">
1008 <a name="dss_list"><list></a>
1011 <div class="doc_text">
1012 <p>std::list is an extremely inefficient class that is rarely useful.
1013 It performs a heap allocation for every element inserted into it, thus having an
1014 extremely high constant factor, particularly for small data types. std::list
1015 also only supports bidirectional iteration, not random access iteration.</p>
1017 <p>In exchange for this high cost, std::list supports efficient access to both
1018 ends of the list (like std::deque, but unlike std::vector or SmallVector). In
1019 addition, the iterator invalidation characteristics of std::list are stronger
1020 than that of a vector class: inserting or removing an element into the list does
1021 not invalidate iterator or pointers to other elements in the list.</p>
1024 <!-- _______________________________________________________________________ -->
1025 <div class="doc_subsubsection">
1026 <a name="dss_ilist">llvm/ADT/ilist.h</a>
1029 <div class="doc_text">
1030 <p><tt>ilist<T></tt> implements an 'intrusive' doubly-linked list. It is
1031 intrusive, because it requires the element to store and provide access to the
1032 prev/next pointers for the list.</p>
1034 <p><tt>ilist</tt> has the same drawbacks as <tt>std::list</tt>, and additionally
1035 requires an <tt>ilist_traits</tt> implementation for the element type, but it
1036 provides some novel characteristics. In particular, it can efficiently store
1037 polymorphic objects, the traits class is informed when an element is inserted or
1038 removed from the list, and <tt>ilist</tt>s are guaranteed to support a
1039 constant-time splice operation.</p>
1041 <p>These properties are exactly what we want for things like
1042 <tt>Instruction</tt>s and basic blocks, which is why these are implemented with
1043 <tt>ilist</tt>s.</p>
1045 Related classes of interest are explained in the following subsections:
1047 <li><a href="#dss_ilist_traits">ilist_traits</a></li>
1048 <li><a href="#dss_iplist">iplist</a></li>
1049 <li><a href="#dss_ilist_node">llvm/ADT/ilist_node.h</a></li>
1050 <li><a href="#dss_ilist_sentinel">Sentinels</a></li>
1054 <!-- _______________________________________________________________________ -->
1055 <div class="doc_subsubsection">
1056 <a name="dss_ilist_traits">ilist_traits</a>
1059 <div class="doc_text">
1060 <p><tt>ilist_traits<T></tt> is <tt>ilist<T></tt>'s customization
1061 mechanism. <tt>iplist<T></tt> (and consequently <tt>ilist<T></tt>)
1062 publicly derive from this traits class.</p>
1065 <!-- _______________________________________________________________________ -->
1066 <div class="doc_subsubsection">
1067 <a name="dss_iplist">iplist</a>
1070 <div class="doc_text">
1071 <p><tt>iplist<T></tt> is <tt>ilist<T></tt>'s base and as such
1072 supports a slightly narrower interface. Notably, inserters from
1073 <tt>T&</tt> are absent.</p>
1075 <p><tt>ilist_traits<T></tt> is a public base of this class and can be
1076 used for a wide variety of customizations.</p>
1079 <!-- _______________________________________________________________________ -->
1080 <div class="doc_subsubsection">
1081 <a name="dss_ilist_node">llvm/ADT/ilist_node.h</a>
1084 <div class="doc_text">
1085 <p><tt>ilist_node<T></tt> implements a the forward and backward links
1086 that are expected by the <tt>ilist<T></tt> (and analogous containers)
1087 in the default manner.</p>
1089 <p><tt>ilist_node<T></tt>s are meant to be embedded in the node type
1090 <tt>T</tt>, usually <tt>T</tt> publicly derives from
1091 <tt>ilist_node<T></tt>.</p>
1094 <!-- _______________________________________________________________________ -->
1095 <div class="doc_subsubsection">
1096 <a name="dss_ilist_sentinel">Sentinels</a>
1099 <div class="doc_text">
1100 <p><tt>ilist</tt>s have another specialty that must be considered. To be a good
1101 citizen in the C++ ecosystem, it needs to support the standard container
1102 operations, such as <tt>begin</tt> and <tt>end</tt> iterators, etc. Also, the
1103 <tt>operator--</tt> must work correctly on the <tt>end</tt> iterator in the
1104 case of non-empty <tt>ilist</tt>s.</p>
1106 <p>The only sensible solution to this problem is to allocate a so-called
1107 <i>sentinel</i> along with the intrusive list, which serves as the <tt>end</tt>
1108 iterator, providing the back-link to the last element. However conforming to the
1109 C++ convention it is illegal to <tt>operator++</tt> beyond the sentinel and it
1110 also must not be dereferenced.</p>
1112 <p>These constraints allow for some implementation freedom to the <tt>ilist</tt>
1113 how to allocate and store the sentinel. The corresponding policy is dictated
1114 by <tt>ilist_traits<T></tt>. By default a <tt>T</tt> gets heap-allocated
1115 whenever the need for a sentinel arises.</p>
1117 <p>While the default policy is sufficient in most cases, it may break down when
1118 <tt>T</tt> does not provide a default constructor. Also, in the case of many
1119 instances of <tt>ilist</tt>s, the memory overhead of the associated sentinels
1120 is wasted. To alleviate the situation with numerous and voluminous
1121 <tt>T</tt>-sentinels, sometimes a trick is employed, leading to <i>ghostly
1124 <p>Ghostly sentinels are obtained by specially-crafted <tt>ilist_traits<T></tt>
1125 which superpose the sentinel with the <tt>ilist</tt> instance in memory. Pointer
1126 arithmetic is used to obtain the sentinel, which is relative to the
1127 <tt>ilist</tt>'s <tt>this</tt> pointer. The <tt>ilist</tt> is augmented by an
1128 extra pointer, which serves as the back-link of the sentinel. This is the only
1129 field in the ghostly sentinel which can be legally accessed.</p>
1132 <!-- _______________________________________________________________________ -->
1133 <div class="doc_subsubsection">
1134 <a name="dss_other">Other Sequential Container options</a>
1137 <div class="doc_text">
1138 <p>Other STL containers are available, such as std::string.</p>
1140 <p>There are also various STL adapter classes such as std::queue,
1141 std::priority_queue, std::stack, etc. These provide simplified access to an
1142 underlying container but don't affect the cost of the container itself.</p>
1147 <!-- ======================================================================= -->
1148 <div class="doc_subsection">
1149 <a name="ds_set">Set-Like Containers (std::set, SmallSet, SetVector, etc)</a>
1152 <div class="doc_text">
1154 <p>Set-like containers are useful when you need to canonicalize multiple values
1155 into a single representation. There are several different choices for how to do
1156 this, providing various trade-offs.</p>
1161 <!-- _______________________________________________________________________ -->
1162 <div class="doc_subsubsection">
1163 <a name="dss_sortedvectorset">A sorted 'vector'</a>
1166 <div class="doc_text">
1168 <p>If you intend to insert a lot of elements, then do a lot of queries, a
1169 great approach is to use a vector (or other sequential container) with
1170 std::sort+std::unique to remove duplicates. This approach works really well if
1171 your usage pattern has these two distinct phases (insert then query), and can be
1172 coupled with a good choice of <a href="#ds_sequential">sequential container</a>.
1176 This combination provides the several nice properties: the result data is
1177 contiguous in memory (good for cache locality), has few allocations, is easy to
1178 address (iterators in the final vector are just indices or pointers), and can be
1179 efficiently queried with a standard binary or radix search.</p>
1183 <!-- _______________________________________________________________________ -->
1184 <div class="doc_subsubsection">
1185 <a name="dss_smallset">"llvm/ADT/SmallSet.h"</a>
1188 <div class="doc_text">
1190 <p>If you have a set-like data structure that is usually small and whose elements
1191 are reasonably small, a <tt>SmallSet<Type, N></tt> is a good choice. This set
1192 has space for N elements in place (thus, if the set is dynamically smaller than
1193 N, no malloc traffic is required) and accesses them with a simple linear search.
1194 When the set grows beyond 'N' elements, it allocates a more expensive representation that
1195 guarantees efficient access (for most types, it falls back to std::set, but for
1196 pointers it uses something far better, <a
1197 href="#dss_smallptrset">SmallPtrSet</a>).</p>
1199 <p>The magic of this class is that it handles small sets extremely efficiently,
1200 but gracefully handles extremely large sets without loss of efficiency. The
1201 drawback is that the interface is quite small: it supports insertion, queries
1202 and erasing, but does not support iteration.</p>
1206 <!-- _______________________________________________________________________ -->
1207 <div class="doc_subsubsection">
1208 <a name="dss_smallptrset">"llvm/ADT/SmallPtrSet.h"</a>
1211 <div class="doc_text">
1213 <p>SmallPtrSet has all the advantages of <tt>SmallSet</tt> (and a <tt>SmallSet</tt> of pointers is
1214 transparently implemented with a <tt>SmallPtrSet</tt>), but also supports iterators. If
1215 more than 'N' insertions are performed, a single quadratically
1216 probed hash table is allocated and grows as needed, providing extremely
1217 efficient access (constant time insertion/deleting/queries with low constant
1218 factors) and is very stingy with malloc traffic.</p>
1220 <p>Note that, unlike <tt>std::set</tt>, the iterators of <tt>SmallPtrSet</tt> are invalidated
1221 whenever an insertion occurs. Also, the values visited by the iterators are not
1222 visited in sorted order.</p>
1226 <!-- _______________________________________________________________________ -->
1227 <div class="doc_subsubsection">
1228 <a name="dss_denseset">"llvm/ADT/DenseSet.h"</a>
1231 <div class="doc_text">
1234 DenseSet is a simple quadratically probed hash table. It excels at supporting
1235 small values: it uses a single allocation to hold all of the pairs that
1236 are currently inserted in the set. DenseSet is a great way to unique small
1237 values that are not simple pointers (use <a
1238 href="#dss_smallptrset">SmallPtrSet</a> for pointers). Note that DenseSet has
1239 the same requirements for the value type that <a
1240 href="#dss_densemap">DenseMap</a> has.
1245 <!-- _______________________________________________________________________ -->
1246 <div class="doc_subsubsection">
1247 <a name="dss_FoldingSet">"llvm/ADT/FoldingSet.h"</a>
1250 <div class="doc_text">
1253 FoldingSet is an aggregate class that is really good at uniquing
1254 expensive-to-create or polymorphic objects. It is a combination of a chained
1255 hash table with intrusive links (uniqued objects are required to inherit from
1256 FoldingSetNode) that uses <a href="#dss_smallvector">SmallVector</a> as part of
1259 <p>Consider a case where you want to implement a "getOrCreateFoo" method for
1260 a complex object (for example, a node in the code generator). The client has a
1261 description of *what* it wants to generate (it knows the opcode and all the
1262 operands), but we don't want to 'new' a node, then try inserting it into a set
1263 only to find out it already exists, at which point we would have to delete it
1264 and return the node that already exists.
1267 <p>To support this style of client, FoldingSet perform a query with a
1268 FoldingSetNodeID (which wraps SmallVector) that can be used to describe the
1269 element that we want to query for. The query either returns the element
1270 matching the ID or it returns an opaque ID that indicates where insertion should
1271 take place. Construction of the ID usually does not require heap traffic.</p>
1273 <p>Because FoldingSet uses intrusive links, it can support polymorphic objects
1274 in the set (for example, you can have SDNode instances mixed with LoadSDNodes).
1275 Because the elements are individually allocated, pointers to the elements are
1276 stable: inserting or removing elements does not invalidate any pointers to other
1282 <!-- _______________________________________________________________________ -->
1283 <div class="doc_subsubsection">
1284 <a name="dss_set"><set></a>
1287 <div class="doc_text">
1289 <p><tt>std::set</tt> is a reasonable all-around set class, which is decent at
1290 many things but great at nothing. std::set allocates memory for each element
1291 inserted (thus it is very malloc intensive) and typically stores three pointers
1292 per element in the set (thus adding a large amount of per-element space
1293 overhead). It offers guaranteed log(n) performance, which is not particularly
1294 fast from a complexity standpoint (particularly if the elements of the set are
1295 expensive to compare, like strings), and has extremely high constant factors for
1296 lookup, insertion and removal.</p>
1298 <p>The advantages of std::set are that its iterators are stable (deleting or
1299 inserting an element from the set does not affect iterators or pointers to other
1300 elements) and that iteration over the set is guaranteed to be in sorted order.
1301 If the elements in the set are large, then the relative overhead of the pointers
1302 and malloc traffic is not a big deal, but if the elements of the set are small,
1303 std::set is almost never a good choice.</p>
1307 <!-- _______________________________________________________________________ -->
1308 <div class="doc_subsubsection">
1309 <a name="dss_setvector">"llvm/ADT/SetVector.h"</a>
1312 <div class="doc_text">
1313 <p>LLVM's SetVector<Type> is an adapter class that combines your choice of
1314 a set-like container along with a <a href="#ds_sequential">Sequential
1315 Container</a>. The important property
1316 that this provides is efficient insertion with uniquing (duplicate elements are
1317 ignored) with iteration support. It implements this by inserting elements into
1318 both a set-like container and the sequential container, using the set-like
1319 container for uniquing and the sequential container for iteration.
1322 <p>The difference between SetVector and other sets is that the order of
1323 iteration is guaranteed to match the order of insertion into the SetVector.
1324 This property is really important for things like sets of pointers. Because
1325 pointer values are non-deterministic (e.g. vary across runs of the program on
1326 different machines), iterating over the pointers in the set will
1327 not be in a well-defined order.</p>
1330 The drawback of SetVector is that it requires twice as much space as a normal
1331 set and has the sum of constant factors from the set-like container and the
1332 sequential container that it uses. Use it *only* if you need to iterate over
1333 the elements in a deterministic order. SetVector is also expensive to delete
1334 elements out of (linear time), unless you use it's "pop_back" method, which is
1338 <p>SetVector is an adapter class that defaults to using std::vector and std::set
1339 for the underlying containers, so it is quite expensive. However,
1340 <tt>"llvm/ADT/SetVector.h"</tt> also provides a SmallSetVector class, which
1341 defaults to using a SmallVector and SmallSet of a specified size. If you use
1342 this, and if your sets are dynamically smaller than N, you will save a lot of
1347 <!-- _______________________________________________________________________ -->
1348 <div class="doc_subsubsection">
1349 <a name="dss_uniquevector">"llvm/ADT/UniqueVector.h"</a>
1352 <div class="doc_text">
1355 UniqueVector is similar to <a href="#dss_setvector">SetVector</a>, but it
1356 retains a unique ID for each element inserted into the set. It internally
1357 contains a map and a vector, and it assigns a unique ID for each value inserted
1360 <p>UniqueVector is very expensive: its cost is the sum of the cost of
1361 maintaining both the map and vector, it has high complexity, high constant
1362 factors, and produces a lot of malloc traffic. It should be avoided.</p>
1367 <!-- _______________________________________________________________________ -->
1368 <div class="doc_subsubsection">
1369 <a name="dss_otherset">Other Set-Like Container Options</a>
1372 <div class="doc_text">
1375 The STL provides several other options, such as std::multiset and the various
1376 "hash_set" like containers (whether from C++ TR1 or from the SGI library). We
1377 never use hash_set and unordered_set because they are generally very expensive
1378 (each insertion requires a malloc) and very non-portable.
1381 <p>std::multiset is useful if you're not interested in elimination of
1382 duplicates, but has all the drawbacks of std::set. A sorted vector (where you
1383 don't delete duplicate entries) or some other approach is almost always
1388 <!-- ======================================================================= -->
1389 <div class="doc_subsection">
1390 <a name="ds_map">Map-Like Containers (std::map, DenseMap, etc)</a>
1393 <div class="doc_text">
1394 Map-like containers are useful when you want to associate data to a key. As
1395 usual, there are a lot of different ways to do this. :)
1398 <!-- _______________________________________________________________________ -->
1399 <div class="doc_subsubsection">
1400 <a name="dss_sortedvectormap">A sorted 'vector'</a>
1403 <div class="doc_text">
1406 If your usage pattern follows a strict insert-then-query approach, you can
1407 trivially use the same approach as <a href="#dss_sortedvectorset">sorted vectors
1408 for set-like containers</a>. The only difference is that your query function
1409 (which uses std::lower_bound to get efficient log(n) lookup) should only compare
1410 the key, not both the key and value. This yields the same advantages as sorted
1415 <!-- _______________________________________________________________________ -->
1416 <div class="doc_subsubsection">
1417 <a name="dss_stringmap">"llvm/ADT/StringMap.h"</a>
1420 <div class="doc_text">
1423 Strings are commonly used as keys in maps, and they are difficult to support
1424 efficiently: they are variable length, inefficient to hash and compare when
1425 long, expensive to copy, etc. StringMap is a specialized container designed to
1426 cope with these issues. It supports mapping an arbitrary range of bytes to an
1427 arbitrary other object.</p>
1429 <p>The StringMap implementation uses a quadratically-probed hash table, where
1430 the buckets store a pointer to the heap allocated entries (and some other
1431 stuff). The entries in the map must be heap allocated because the strings are
1432 variable length. The string data (key) and the element object (value) are
1433 stored in the same allocation with the string data immediately after the element
1434 object. This container guarantees the "<tt>(char*)(&Value+1)</tt>" points
1435 to the key string for a value.</p>
1437 <p>The StringMap is very fast for several reasons: quadratic probing is very
1438 cache efficient for lookups, the hash value of strings in buckets is not
1439 recomputed when looking up an element, StringMap rarely has to touch the
1440 memory for unrelated objects when looking up a value (even when hash collisions
1441 happen), hash table growth does not recompute the hash values for strings
1442 already in the table, and each pair in the map is store in a single allocation
1443 (the string data is stored in the same allocation as the Value of a pair).</p>
1445 <p>StringMap also provides query methods that take byte ranges, so it only ever
1446 copies a string if a value is inserted into the table.</p>
1449 <!-- _______________________________________________________________________ -->
1450 <div class="doc_subsubsection">
1451 <a name="dss_indexedmap">"llvm/ADT/IndexedMap.h"</a>
1454 <div class="doc_text">
1456 IndexedMap is a specialized container for mapping small dense integers (or
1457 values that can be mapped to small dense integers) to some other type. It is
1458 internally implemented as a vector with a mapping function that maps the keys to
1459 the dense integer range.
1463 This is useful for cases like virtual registers in the LLVM code generator: they
1464 have a dense mapping that is offset by a compile-time constant (the first
1465 virtual register ID).</p>
1469 <!-- _______________________________________________________________________ -->
1470 <div class="doc_subsubsection">
1471 <a name="dss_densemap">"llvm/ADT/DenseMap.h"</a>
1474 <div class="doc_text">
1477 DenseMap is a simple quadratically probed hash table. It excels at supporting
1478 small keys and values: it uses a single allocation to hold all of the pairs that
1479 are currently inserted in the map. DenseMap is a great way to map pointers to
1480 pointers, or map other small types to each other.
1484 There are several aspects of DenseMap that you should be aware of, however. The
1485 iterators in a densemap are invalidated whenever an insertion occurs, unlike
1486 map. Also, because DenseMap allocates space for a large number of key/value
1487 pairs (it starts with 64 by default), it will waste a lot of space if your keys
1488 or values are large. Finally, you must implement a partial specialization of
1489 DenseMapInfo for the key that you want, if it isn't already supported. This
1490 is required to tell DenseMap about two special marker values (which can never be
1491 inserted into the map) that it needs internally.</p>
1495 <!-- _______________________________________________________________________ -->
1496 <div class="doc_subsubsection">
1497 <a name="dss_valuemap">"llvm/ADT/ValueMap.h"</a>
1500 <div class="doc_text">
1503 ValueMap is a wrapper around a <a href="#dss_densemap">DenseMap</a> mapping
1504 Value*s (or subclasses) to another type. When a Value is deleted or RAUW'ed,
1505 ValueMap will update itself so the new version of the key is mapped to the same
1506 value, just as if the key were a WeakVH. You can configure exactly how this
1507 happens, and what else happens on these two events, by passing
1508 a <code>Config</code> parameter to the ValueMap template.</p>
1512 <!-- _______________________________________________________________________ -->
1513 <div class="doc_subsubsection">
1514 <a name="dss_map"><map></a>
1517 <div class="doc_text">
1520 std::map has similar characteristics to <a href="#dss_set">std::set</a>: it uses
1521 a single allocation per pair inserted into the map, it offers log(n) lookup with
1522 an extremely large constant factor, imposes a space penalty of 3 pointers per
1523 pair in the map, etc.</p>
1525 <p>std::map is most useful when your keys or values are very large, if you need
1526 to iterate over the collection in sorted order, or if you need stable iterators
1527 into the map (i.e. they don't get invalidated if an insertion or deletion of
1528 another element takes place).</p>
1532 <!-- _______________________________________________________________________ -->
1533 <div class="doc_subsubsection">
1534 <a name="dss_othermap">Other Map-Like Container Options</a>
1537 <div class="doc_text">
1540 The STL provides several other options, such as std::multimap and the various
1541 "hash_map" like containers (whether from C++ TR1 or from the SGI library). We
1542 never use hash_set and unordered_set because they are generally very expensive
1543 (each insertion requires a malloc) and very non-portable.</p>
1545 <p>std::multimap is useful if you want to map a key to multiple values, but has
1546 all the drawbacks of std::map. A sorted vector or some other approach is almost
1551 <!-- ======================================================================= -->
1552 <div class="doc_subsection">
1553 <a name="ds_string">String-like containers</a>
1556 <div class="doc_text">
1559 TODO: const char* vs stringref vs smallstring vs std::string. Describe twine,
1560 xref to #string_apis.
1565 <!-- ======================================================================= -->
1566 <div class="doc_subsection">
1567 <a name="ds_bit">Bit storage containers (BitVector, SparseBitVector)</a>
1570 <div class="doc_text">
1571 <p>Unlike the other containers, there are only two bit storage containers, and
1572 choosing when to use each is relatively straightforward.</p>
1574 <p>One additional option is
1575 <tt>std::vector<bool></tt>: we discourage its use for two reasons 1) the
1576 implementation in many common compilers (e.g. commonly available versions of
1577 GCC) is extremely inefficient and 2) the C++ standards committee is likely to
1578 deprecate this container and/or change it significantly somehow. In any case,
1579 please don't use it.</p>
1582 <!-- _______________________________________________________________________ -->
1583 <div class="doc_subsubsection">
1584 <a name="dss_bitvector">BitVector</a>
1587 <div class="doc_text">
1588 <p> The BitVector container provides a dynamic size set of bits for manipulation.
1589 It supports individual bit setting/testing, as well as set operations. The set
1590 operations take time O(size of bitvector), but operations are performed one word
1591 at a time, instead of one bit at a time. This makes the BitVector very fast for
1592 set operations compared to other containers. Use the BitVector when you expect
1593 the number of set bits to be high (IE a dense set).
1597 <!-- _______________________________________________________________________ -->
1598 <div class="doc_subsubsection">
1599 <a name="dss_smallbitvector">SmallBitVector</a>
1602 <div class="doc_text">
1603 <p> The SmallBitVector container provides the same interface as BitVector, but
1604 it is optimized for the case where only a small number of bits, less than
1605 25 or so, are needed. It also transparently supports larger bit counts, but
1606 slightly less efficiently than a plain BitVector, so SmallBitVector should
1607 only be used when larger counts are rare.
1611 At this time, SmallBitVector does not support set operations (and, or, xor),
1612 and its operator[] does not provide an assignable lvalue.
1616 <!-- _______________________________________________________________________ -->
1617 <div class="doc_subsubsection">
1618 <a name="dss_sparsebitvector">SparseBitVector</a>
1621 <div class="doc_text">
1622 <p> The SparseBitVector container is much like BitVector, with one major
1623 difference: Only the bits that are set, are stored. This makes the
1624 SparseBitVector much more space efficient than BitVector when the set is sparse,
1625 as well as making set operations O(number of set bits) instead of O(size of
1626 universe). The downside to the SparseBitVector is that setting and testing of random bits is O(N), and on large SparseBitVectors, this can be slower than BitVector. In our implementation, setting or testing bits in sorted order
1627 (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).
1631 <!-- *********************************************************************** -->
1632 <div class="doc_section">
1633 <a name="common">Helpful Hints for Common Operations</a>
1635 <!-- *********************************************************************** -->
1637 <div class="doc_text">
1639 <p>This section describes how to perform some very simple transformations of
1640 LLVM code. This is meant to give examples of common idioms used, showing the
1641 practical side of LLVM transformations. <p> Because this is a "how-to" section,
1642 you should also read about the main classes that you will be working with. The
1643 <a href="#coreclasses">Core LLVM Class Hierarchy Reference</a> contains details
1644 and descriptions of the main classes that you should know about.</p>
1648 <!-- NOTE: this section should be heavy on example code -->
1649 <!-- ======================================================================= -->
1650 <div class="doc_subsection">
1651 <a name="inspection">Basic Inspection and Traversal Routines</a>
1654 <div class="doc_text">
1656 <p>The LLVM compiler infrastructure have many different data structures that may
1657 be traversed. Following the example of the C++ standard template library, the
1658 techniques used to traverse these various data structures are all basically the
1659 same. For a enumerable sequence of values, the <tt>XXXbegin()</tt> function (or
1660 method) returns an iterator to the start of the sequence, the <tt>XXXend()</tt>
1661 function returns an iterator pointing to one past the last valid element of the
1662 sequence, and there is some <tt>XXXiterator</tt> data type that is common
1663 between the two operations.</p>
1665 <p>Because the pattern for iteration is common across many different aspects of
1666 the program representation, the standard template library algorithms may be used
1667 on them, and it is easier to remember how to iterate. First we show a few common
1668 examples of the data structures that need to be traversed. Other data
1669 structures are traversed in very similar ways.</p>
1673 <!-- _______________________________________________________________________ -->
1674 <div class="doc_subsubsection">
1675 <a name="iterate_function">Iterating over the </a><a
1676 href="#BasicBlock"><tt>BasicBlock</tt></a>s in a <a
1677 href="#Function"><tt>Function</tt></a>
1680 <div class="doc_text">
1682 <p>It's quite common to have a <tt>Function</tt> instance that you'd like to
1683 transform in some way; in particular, you'd like to manipulate its
1684 <tt>BasicBlock</tt>s. To facilitate this, you'll need to iterate over all of
1685 the <tt>BasicBlock</tt>s that constitute the <tt>Function</tt>. The following is
1686 an example that prints the name of a <tt>BasicBlock</tt> and the number of
1687 <tt>Instruction</tt>s it contains:</p>
1689 <div class="doc_code">
1691 // <i>func is a pointer to a Function instance</i>
1692 for (Function::iterator i = func->begin(), e = func->end(); i != e; ++i)
1693 // <i>Print out the name of the basic block if it has one, and then the</i>
1694 // <i>number of instructions that it contains</i>
1695 errs() << "Basic block (name=" << i->getName() << ") has "
1696 << i->size() << " instructions.\n";
1700 <p>Note that i can be used as if it were a pointer for the purposes of
1701 invoking member functions of the <tt>Instruction</tt> class. This is
1702 because the indirection operator is overloaded for the iterator
1703 classes. In the above code, the expression <tt>i->size()</tt> is
1704 exactly equivalent to <tt>(*i).size()</tt> just like you'd expect.</p>
1708 <!-- _______________________________________________________________________ -->
1709 <div class="doc_subsubsection">
1710 <a name="iterate_basicblock">Iterating over the </a><a
1711 href="#Instruction"><tt>Instruction</tt></a>s in a <a
1712 href="#BasicBlock"><tt>BasicBlock</tt></a>
1715 <div class="doc_text">
1717 <p>Just like when dealing with <tt>BasicBlock</tt>s in <tt>Function</tt>s, it's
1718 easy to iterate over the individual instructions that make up
1719 <tt>BasicBlock</tt>s. Here's a code snippet that prints out each instruction in
1720 a <tt>BasicBlock</tt>:</p>
1722 <div class="doc_code">
1724 // <i>blk is a pointer to a BasicBlock instance</i>
1725 for (BasicBlock::iterator i = blk->begin(), e = blk->end(); i != e; ++i)
1726 // <i>The next statement works since operator<<(ostream&,...)</i>
1727 // <i>is overloaded for Instruction&</i>
1728 errs() << *i << "\n";
1732 <p>However, this isn't really the best way to print out the contents of a
1733 <tt>BasicBlock</tt>! Since the ostream operators are overloaded for virtually
1734 anything you'll care about, you could have just invoked the print routine on the
1735 basic block itself: <tt>errs() << *blk << "\n";</tt>.</p>
1739 <!-- _______________________________________________________________________ -->
1740 <div class="doc_subsubsection">
1741 <a name="iterate_institer">Iterating over the </a><a
1742 href="#Instruction"><tt>Instruction</tt></a>s in a <a
1743 href="#Function"><tt>Function</tt></a>
1746 <div class="doc_text">
1748 <p>If you're finding that you commonly iterate over a <tt>Function</tt>'s
1749 <tt>BasicBlock</tt>s and then that <tt>BasicBlock</tt>'s <tt>Instruction</tt>s,
1750 <tt>InstIterator</tt> should be used instead. You'll need to include <a
1751 href="/doxygen/InstIterator_8h-source.html"><tt>llvm/Support/InstIterator.h</tt></a>,
1752 and then instantiate <tt>InstIterator</tt>s explicitly in your code. Here's a
1753 small example that shows how to dump all instructions in a function to the standard error stream:<p>
1755 <div class="doc_code">
1757 #include "<a href="/doxygen/InstIterator_8h-source.html">llvm/Support/InstIterator.h</a>"
1759 // <i>F is a pointer to a Function instance</i>
1760 for (inst_iterator I = inst_begin(F), E = inst_end(F); I != E; ++I)
1761 errs() << *I << "\n";
1765 <p>Easy, isn't it? You can also use <tt>InstIterator</tt>s to fill a
1766 work list with its initial contents. For example, if you wanted to
1767 initialize a work list to contain all instructions in a <tt>Function</tt>
1768 F, all you would need to do is something like:</p>
1770 <div class="doc_code">
1772 std::set<Instruction*> worklist;
1773 // or better yet, SmallPtrSet<Instruction*, 64> worklist;
1775 for (inst_iterator I = inst_begin(F), E = inst_end(F); I != E; ++I)
1776 worklist.insert(&*I);
1780 <p>The STL set <tt>worklist</tt> would now contain all instructions in the
1781 <tt>Function</tt> pointed to by F.</p>
1785 <!-- _______________________________________________________________________ -->
1786 <div class="doc_subsubsection">
1787 <a name="iterate_convert">Turning an iterator into a class pointer (and
1791 <div class="doc_text">
1793 <p>Sometimes, it'll be useful to grab a reference (or pointer) to a class
1794 instance when all you've got at hand is an iterator. Well, extracting
1795 a reference or a pointer from an iterator is very straight-forward.
1796 Assuming that <tt>i</tt> is a <tt>BasicBlock::iterator</tt> and <tt>j</tt>
1797 is a <tt>BasicBlock::const_iterator</tt>:</p>
1799 <div class="doc_code">
1801 Instruction& inst = *i; // <i>Grab reference to instruction reference</i>
1802 Instruction* pinst = &*i; // <i>Grab pointer to instruction reference</i>
1803 const Instruction& inst = *j;
1807 <p>However, the iterators you'll be working with in the LLVM framework are
1808 special: they will automatically convert to a ptr-to-instance type whenever they
1809 need to. Instead of dereferencing the iterator and then taking the address of
1810 the result, you can simply assign the iterator to the proper pointer type and
1811 you get the dereference and address-of operation as a result of the assignment
1812 (behind the scenes, this is a result of overloading casting mechanisms). Thus
1813 the last line of the last example,</p>
1815 <div class="doc_code">
1817 Instruction *pinst = &*i;
1821 <p>is semantically equivalent to</p>
1823 <div class="doc_code">
1825 Instruction *pinst = i;
1829 <p>It's also possible to turn a class pointer into the corresponding iterator,
1830 and this is a constant time operation (very efficient). The following code
1831 snippet illustrates use of the conversion constructors provided by LLVM
1832 iterators. By using these, you can explicitly grab the iterator of something
1833 without actually obtaining it via iteration over some structure:</p>
1835 <div class="doc_code">
1837 void printNextInstruction(Instruction* inst) {
1838 BasicBlock::iterator it(inst);
1839 ++it; // <i>After this line, it refers to the instruction after *inst</i>
1840 if (it != inst->getParent()->end()) errs() << *it << "\n";
1845 <p>Unfortunately, these implicit conversions come at a cost; they prevent
1846 these iterators from conforming to standard iterator conventions, and thus
1847 from being usable with standard algorithms and containers. For example, they
1848 prevent the following code, where <tt>B</tt> is a <tt>BasicBlock</tt>,
1851 <div class="doc_code">
1853 llvm::SmallVector<llvm::Instruction *, 16>(B->begin(), B->end());
1857 <p>Because of this, these implicit conversions may be removed some day,
1858 and <tt>operator*</tt> changed to return a pointer instead of a reference.</p>
1862 <!--_______________________________________________________________________-->
1863 <div class="doc_subsubsection">
1864 <a name="iterate_complex">Finding call sites: a slightly more complex
1868 <div class="doc_text">
1870 <p>Say that you're writing a FunctionPass and would like to count all the
1871 locations in the entire module (that is, across every <tt>Function</tt>) where a
1872 certain function (i.e., some <tt>Function</tt>*) is already in scope. As you'll
1873 learn later, you may want to use an <tt>InstVisitor</tt> to accomplish this in a
1874 much more straight-forward manner, but this example will allow us to explore how
1875 you'd do it if you didn't have <tt>InstVisitor</tt> around. In pseudo-code, this
1876 is what we want to do:</p>
1878 <div class="doc_code">
1880 initialize callCounter to zero
1881 for each Function f in the Module
1882 for each BasicBlock b in f
1883 for each Instruction i in b
1884 if (i is a CallInst and calls the given function)
1885 increment callCounter
1889 <p>And the actual code is (remember, because we're writing a
1890 <tt>FunctionPass</tt>, our <tt>FunctionPass</tt>-derived class simply has to
1891 override the <tt>runOnFunction</tt> method):</p>
1893 <div class="doc_code">
1895 Function* targetFunc = ...;
1897 class OurFunctionPass : public FunctionPass {
1899 OurFunctionPass(): callCounter(0) { }
1901 virtual runOnFunction(Function& F) {
1902 for (Function::iterator b = F.begin(), be = F.end(); b != be; ++b) {
1903 for (BasicBlock::iterator i = b->begin(), ie = b->end(); i != ie; ++i) {
1904 if (<a href="#CallInst">CallInst</a>* callInst = <a href="#isa">dyn_cast</a><<a
1905 href="#CallInst">CallInst</a>>(&*i)) {
1906 // <i>We know we've encountered a call instruction, so we</i>
1907 // <i>need to determine if it's a call to the</i>
1908 // <i>function pointed to by m_func or not.</i>
1909 if (callInst->getCalledFunction() == targetFunc)
1917 unsigned callCounter;
1924 <!--_______________________________________________________________________-->
1925 <div class="doc_subsubsection">
1926 <a name="calls_and_invokes">Treating calls and invokes the same way</a>
1929 <div class="doc_text">
1931 <p>You may have noticed that the previous example was a bit oversimplified in
1932 that it did not deal with call sites generated by 'invoke' instructions. In
1933 this, and in other situations, you may find that you want to treat
1934 <tt>CallInst</tt>s and <tt>InvokeInst</tt>s the same way, even though their
1935 most-specific common base class is <tt>Instruction</tt>, which includes lots of
1936 less closely-related things. For these cases, LLVM provides a handy wrapper
1938 href="http://llvm.org/doxygen/classllvm_1_1CallSite.html"><tt>CallSite</tt></a>.
1939 It is essentially a wrapper around an <tt>Instruction</tt> pointer, with some
1940 methods that provide functionality common to <tt>CallInst</tt>s and
1941 <tt>InvokeInst</tt>s.</p>
1943 <p>This class has "value semantics": it should be passed by value, not by
1944 reference and it should not be dynamically allocated or deallocated using
1945 <tt>operator new</tt> or <tt>operator delete</tt>. It is efficiently copyable,
1946 assignable and constructable, with costs equivalents to that of a bare pointer.
1947 If you look at its definition, it has only a single pointer member.</p>
1951 <!--_______________________________________________________________________-->
1952 <div class="doc_subsubsection">
1953 <a name="iterate_chains">Iterating over def-use & use-def chains</a>
1956 <div class="doc_text">
1958 <p>Frequently, we might have an instance of the <a
1959 href="/doxygen/classllvm_1_1Value.html">Value Class</a> and we want to
1960 determine which <tt>User</tt>s use the <tt>Value</tt>. The list of all
1961 <tt>User</tt>s of a particular <tt>Value</tt> is called a <i>def-use</i> chain.
1962 For example, let's say we have a <tt>Function*</tt> named <tt>F</tt> to a
1963 particular function <tt>foo</tt>. Finding all of the instructions that
1964 <i>use</i> <tt>foo</tt> is as simple as iterating over the <i>def-use</i> chain
1967 <div class="doc_code">
1971 for (Value::use_iterator i = F->use_begin(), e = F->use_end(); i != e; ++i)
1972 if (Instruction *Inst = dyn_cast<Instruction>(*i)) {
1973 errs() << "F is used in instruction:\n";
1974 errs() << *Inst << "\n";
1979 <p>Note that dereferencing a <tt>Value::use_iterator</tt> is not a very cheap
1980 operation. Instead of performing <tt>*i</tt> above several times, consider
1981 doing it only once in the loop body and reusing its result.</p>
1983 <p>Alternatively, it's common to have an instance of the <a
1984 href="/doxygen/classllvm_1_1User.html">User Class</a> and need to know what
1985 <tt>Value</tt>s are used by it. The list of all <tt>Value</tt>s used by a
1986 <tt>User</tt> is known as a <i>use-def</i> chain. Instances of class
1987 <tt>Instruction</tt> are common <tt>User</tt>s, so we might want to iterate over
1988 all of the values that a particular instruction uses (that is, the operands of
1989 the particular <tt>Instruction</tt>):</p>
1991 <div class="doc_code">
1993 Instruction *pi = ...;
1995 for (User::op_iterator i = pi->op_begin(), e = pi->op_end(); i != e; ++i) {
2002 <p>Declaring objects as <tt>const</tt> is an important tool of enforcing
2003 mutation free algorithms (such as analyses, etc.). For this purpose above
2004 iterators come in constant flavors as <tt>Value::const_use_iterator</tt>
2005 and <tt>Value::const_op_iterator</tt>. They automatically arise when
2006 calling <tt>use/op_begin()</tt> on <tt>const Value*</tt>s or
2007 <tt>const User*</tt>s respectively. Upon dereferencing, they return
2008 <tt>const Use*</tt>s. Otherwise the above patterns remain unchanged.</p>
2012 <!--_______________________________________________________________________-->
2013 <div class="doc_subsubsection">
2014 <a name="iterate_preds">Iterating over predecessors &
2015 successors of blocks</a>
2018 <div class="doc_text">
2020 <p>Iterating over the predecessors and successors of a block is quite easy
2021 with the routines defined in <tt>"llvm/Support/CFG.h"</tt>. Just use code like
2022 this to iterate over all predecessors of BB:</p>
2024 <div class="doc_code">
2026 #include "llvm/Support/CFG.h"
2027 BasicBlock *BB = ...;
2029 for (pred_iterator PI = pred_begin(BB), E = pred_end(BB); PI != E; ++PI) {
2030 BasicBlock *Pred = *PI;
2036 <p>Similarly, to iterate over successors use
2037 succ_iterator/succ_begin/succ_end.</p>
2042 <!-- ======================================================================= -->
2043 <div class="doc_subsection">
2044 <a name="simplechanges">Making simple changes</a>
2047 <div class="doc_text">
2049 <p>There are some primitive transformation operations present in the LLVM
2050 infrastructure that are worth knowing about. When performing
2051 transformations, it's fairly common to manipulate the contents of basic
2052 blocks. This section describes some of the common methods for doing so
2053 and gives example code.</p>
2057 <!--_______________________________________________________________________-->
2058 <div class="doc_subsubsection">
2059 <a name="schanges_creating">Creating and inserting new
2060 <tt>Instruction</tt>s</a>
2063 <div class="doc_text">
2065 <p><i>Instantiating Instructions</i></p>
2067 <p>Creation of <tt>Instruction</tt>s is straight-forward: simply call the
2068 constructor for the kind of instruction to instantiate and provide the necessary
2069 parameters. For example, an <tt>AllocaInst</tt> only <i>requires</i> a
2070 (const-ptr-to) <tt>Type</tt>. Thus:</p>
2072 <div class="doc_code">
2074 AllocaInst* ai = new AllocaInst(Type::Int32Ty);
2078 <p>will create an <tt>AllocaInst</tt> instance that represents the allocation of
2079 one integer in the current stack frame, at run time. Each <tt>Instruction</tt>
2080 subclass is likely to have varying default parameters which change the semantics
2081 of the instruction, so refer to the <a
2082 href="/doxygen/classllvm_1_1Instruction.html">doxygen documentation for the subclass of
2083 Instruction</a> that you're interested in instantiating.</p>
2085 <p><i>Naming values</i></p>
2087 <p>It is very useful to name the values of instructions when you're able to, as
2088 this facilitates the debugging of your transformations. If you end up looking
2089 at generated LLVM machine code, you definitely want to have logical names
2090 associated with the results of instructions! By supplying a value for the
2091 <tt>Name</tt> (default) parameter of the <tt>Instruction</tt> constructor, you
2092 associate a logical name with the result of the instruction's execution at
2093 run time. For example, say that I'm writing a transformation that dynamically
2094 allocates space for an integer on the stack, and that integer is going to be
2095 used as some kind of index by some other code. To accomplish this, I place an
2096 <tt>AllocaInst</tt> at the first point in the first <tt>BasicBlock</tt> of some
2097 <tt>Function</tt>, and I'm intending to use it within the same
2098 <tt>Function</tt>. I might do:</p>
2100 <div class="doc_code">
2102 AllocaInst* pa = new AllocaInst(Type::Int32Ty, 0, "indexLoc");
2106 <p>where <tt>indexLoc</tt> is now the logical name of the instruction's
2107 execution value, which is a pointer to an integer on the run time stack.</p>
2109 <p><i>Inserting instructions</i></p>
2111 <p>There are essentially two ways to insert an <tt>Instruction</tt>
2112 into an existing sequence of instructions that form a <tt>BasicBlock</tt>:</p>
2115 <li>Insertion into an explicit instruction list
2117 <p>Given a <tt>BasicBlock* pb</tt>, an <tt>Instruction* pi</tt> within that
2118 <tt>BasicBlock</tt>, and a newly-created instruction we wish to insert
2119 before <tt>*pi</tt>, we do the following: </p>
2121 <div class="doc_code">
2123 BasicBlock *pb = ...;
2124 Instruction *pi = ...;
2125 Instruction *newInst = new Instruction(...);
2127 pb->getInstList().insert(pi, newInst); // <i>Inserts newInst before pi in pb</i>
2131 <p>Appending to the end of a <tt>BasicBlock</tt> is so common that
2132 the <tt>Instruction</tt> class and <tt>Instruction</tt>-derived
2133 classes provide constructors which take a pointer to a
2134 <tt>BasicBlock</tt> to be appended to. For example code that
2137 <div class="doc_code">
2139 BasicBlock *pb = ...;
2140 Instruction *newInst = new Instruction(...);
2142 pb->getInstList().push_back(newInst); // <i>Appends newInst to pb</i>
2148 <div class="doc_code">
2150 BasicBlock *pb = ...;
2151 Instruction *newInst = new Instruction(..., pb);
2155 <p>which is much cleaner, especially if you are creating
2156 long instruction streams.</p></li>
2158 <li>Insertion into an implicit instruction list
2160 <p><tt>Instruction</tt> instances that are already in <tt>BasicBlock</tt>s
2161 are implicitly associated with an existing instruction list: the instruction
2162 list of the enclosing basic block. Thus, we could have accomplished the same
2163 thing as the above code without being given a <tt>BasicBlock</tt> by doing:
2166 <div class="doc_code">
2168 Instruction *pi = ...;
2169 Instruction *newInst = new Instruction(...);
2171 pi->getParent()->getInstList().insert(pi, newInst);
2175 <p>In fact, this sequence of steps occurs so frequently that the
2176 <tt>Instruction</tt> class and <tt>Instruction</tt>-derived classes provide
2177 constructors which take (as a default parameter) a pointer to an
2178 <tt>Instruction</tt> which the newly-created <tt>Instruction</tt> should
2179 precede. That is, <tt>Instruction</tt> constructors are capable of
2180 inserting the newly-created instance into the <tt>BasicBlock</tt> of a
2181 provided instruction, immediately before that instruction. Using an
2182 <tt>Instruction</tt> constructor with a <tt>insertBefore</tt> (default)
2183 parameter, the above code becomes:</p>
2185 <div class="doc_code">
2187 Instruction* pi = ...;
2188 Instruction* newInst = new Instruction(..., pi);
2192 <p>which is much cleaner, especially if you're creating a lot of
2193 instructions and adding them to <tt>BasicBlock</tt>s.</p></li>
2198 <!--_______________________________________________________________________-->
2199 <div class="doc_subsubsection">
2200 <a name="schanges_deleting">Deleting <tt>Instruction</tt>s</a>
2203 <div class="doc_text">
2205 <p>Deleting an instruction from an existing sequence of instructions that form a
2206 <a href="#BasicBlock"><tt>BasicBlock</tt></a> is very straight-forward. First,
2207 you must have a pointer to the instruction that you wish to delete. Second, you
2208 need to obtain the pointer to that instruction's basic block. You use the
2209 pointer to the basic block to get its list of instructions and then use the
2210 erase function to remove your instruction. For example:</p>
2212 <div class="doc_code">
2214 <a href="#Instruction">Instruction</a> *I = .. ;
2215 I->eraseFromParent();
2221 <!--_______________________________________________________________________-->
2222 <div class="doc_subsubsection">
2223 <a name="schanges_replacing">Replacing an <tt>Instruction</tt> with another
2227 <div class="doc_text">
2229 <p><i>Replacing individual instructions</i></p>
2231 <p>Including "<a href="/doxygen/BasicBlockUtils_8h-source.html">llvm/Transforms/Utils/BasicBlockUtils.h</a>"
2232 permits use of two very useful replace functions: <tt>ReplaceInstWithValue</tt>
2233 and <tt>ReplaceInstWithInst</tt>.</p>
2235 <h4><a name="schanges_deleting">Deleting <tt>Instruction</tt>s</a></h4>
2238 <li><tt>ReplaceInstWithValue</tt>
2240 <p>This function replaces all uses of a given instruction with a value,
2241 and then removes the original instruction. The following example
2242 illustrates the replacement of the result of a particular
2243 <tt>AllocaInst</tt> that allocates memory for a single integer with a null
2244 pointer to an integer.</p>
2246 <div class="doc_code">
2248 AllocaInst* instToReplace = ...;
2249 BasicBlock::iterator ii(instToReplace);
2251 ReplaceInstWithValue(instToReplace->getParent()->getInstList(), ii,
2252 Constant::getNullValue(PointerType::getUnqual(Type::Int32Ty)));
2255 <li><tt>ReplaceInstWithInst</tt>
2257 <p>This function replaces a particular instruction with another
2258 instruction, inserting the new instruction into the basic block at the
2259 location where the old instruction was, and replacing any uses of the old
2260 instruction with the new instruction. The following example illustrates
2261 the replacement of one <tt>AllocaInst</tt> with another.</p>
2263 <div class="doc_code">
2265 AllocaInst* instToReplace = ...;
2266 BasicBlock::iterator ii(instToReplace);
2268 ReplaceInstWithInst(instToReplace->getParent()->getInstList(), ii,
2269 new AllocaInst(Type::Int32Ty, 0, "ptrToReplacedInt"));
2273 <p><i>Replacing multiple uses of <tt>User</tt>s and <tt>Value</tt>s</i></p>
2275 <p>You can use <tt>Value::replaceAllUsesWith</tt> and
2276 <tt>User::replaceUsesOfWith</tt> to change more than one use at a time. See the
2277 doxygen documentation for the <a href="/doxygen/classllvm_1_1Value.html">Value Class</a>
2278 and <a href="/doxygen/classllvm_1_1User.html">User Class</a>, respectively, for more
2281 <!-- Value::replaceAllUsesWith User::replaceUsesOfWith Point out:
2282 include/llvm/Transforms/Utils/ especially BasicBlockUtils.h with:
2283 ReplaceInstWithValue, ReplaceInstWithInst -->
2287 <!--_______________________________________________________________________-->
2288 <div class="doc_subsubsection">
2289 <a name="schanges_deletingGV">Deleting <tt>GlobalVariable</tt>s</a>
2292 <div class="doc_text">
2294 <p>Deleting a global variable from a module is just as easy as deleting an
2295 Instruction. First, you must have a pointer to the global variable that you wish
2296 to delete. You use this pointer to erase it from its parent, the module.
2299 <div class="doc_code">
2301 <a href="#GlobalVariable">GlobalVariable</a> *GV = .. ;
2303 GV->eraseFromParent();
2309 <!-- ======================================================================= -->
2310 <div class="doc_subsection">
2311 <a name="create_types">How to Create Types</a>
2314 <div class="doc_text">
2316 <p>In generating IR, you may need some complex types. If you know these types
2317 statically, you can use <tt>TypeBuilder<...>::get()</tt>, defined
2318 in <tt>llvm/Support/TypeBuilder.h</tt>, to retrieve them. <tt>TypeBuilder</tt>
2319 has two forms depending on whether you're building types for cross-compilation
2320 or native library use. <tt>TypeBuilder<T, true></tt> requires
2321 that <tt>T</tt> be independent of the host environment, meaning that it's built
2323 the <a href="/doxygen/namespacellvm_1_1types.html"><tt>llvm::types</tt></a>
2324 namespace and pointers, functions, arrays, etc. built of
2325 those. <tt>TypeBuilder<T, false></tt> additionally allows native C types
2326 whose size may depend on the host compiler. For example,</p>
2328 <div class="doc_code">
2330 FunctionType *ft = TypeBuilder<types::i<8>(types::i<32>*), true>::get();
2334 <p>is easier to read and write than the equivalent</p>
2336 <div class="doc_code">
2338 std::vector<const Type*> params;
2339 params.push_back(PointerType::getUnqual(Type::Int32Ty));
2340 FunctionType *ft = FunctionType::get(Type::Int8Ty, params, false);
2344 <p>See the <a href="/doxygen/TypeBuilder_8h-source.html#l00001">class
2345 comment</a> for more details.</p>
2349 <!-- *********************************************************************** -->
2350 <div class="doc_section">
2351 <a name="threading">Threads and LLVM</a>
2353 <!-- *********************************************************************** -->
2355 <div class="doc_text">
2357 This section describes the interaction of the LLVM APIs with multithreading,
2358 both on the part of client applications, and in the JIT, in the hosted
2363 Note that LLVM's support for multithreading is still relatively young. Up
2364 through version 2.5, the execution of threaded hosted applications was
2365 supported, but not threaded client access to the APIs. While this use case is
2366 now supported, clients <em>must</em> adhere to the guidelines specified below to
2367 ensure proper operation in multithreaded mode.
2371 Note that, on Unix-like platforms, LLVM requires the presence of GCC's atomic
2372 intrinsics in order to support threaded operation. If you need a
2373 multhreading-capable LLVM on a platform without a suitably modern system
2374 compiler, consider compiling LLVM and LLVM-GCC in single-threaded mode, and
2375 using the resultant compiler to build a copy of LLVM with multithreading
2380 <!-- ======================================================================= -->
2381 <div class="doc_subsection">
2382 <a name="startmultithreaded">Entering and Exiting Multithreaded Mode</a>
2385 <div class="doc_text">
2388 In order to properly protect its internal data structures while avoiding
2389 excessive locking overhead in the single-threaded case, the LLVM must intialize
2390 certain data structures necessary to provide guards around its internals. To do
2391 so, the client program must invoke <tt>llvm_start_multithreaded()</tt> before
2392 making any concurrent LLVM API calls. To subsequently tear down these
2393 structures, use the <tt>llvm_stop_multithreaded()</tt> call. You can also use
2394 the <tt>llvm_is_multithreaded()</tt> call to check the status of multithreaded
2399 Note that both of these calls must be made <em>in isolation</em>. That is to
2400 say that no other LLVM API calls may be executing at any time during the
2401 execution of <tt>llvm_start_multithreaded()</tt> or <tt>llvm_stop_multithreaded
2402 </tt>. It's is the client's responsibility to enforce this isolation.
2406 The return value of <tt>llvm_start_multithreaded()</tt> indicates the success or
2407 failure of the initialization. Failure typically indicates that your copy of
2408 LLVM was built without multithreading support, typically because GCC atomic
2409 intrinsics were not found in your system compiler. In this case, the LLVM API
2410 will not be safe for concurrent calls. However, it <em>will</em> be safe for
2411 hosting threaded applications in the JIT, though <a href="#jitthreading">care
2412 must be taken</a> to ensure that side exits and the like do not accidentally
2413 result in concurrent LLVM API calls.
2417 <!-- ======================================================================= -->
2418 <div class="doc_subsection">
2419 <a name="shutdown">Ending Execution with <tt>llvm_shutdown()</tt></a>
2422 <div class="doc_text">
2424 When you are done using the LLVM APIs, you should call <tt>llvm_shutdown()</tt>
2425 to deallocate memory used for internal structures. This will also invoke
2426 <tt>llvm_stop_multithreaded()</tt> if LLVM is operating in multithreaded mode.
2427 As such, <tt>llvm_shutdown()</tt> requires the same isolation guarantees as
2428 <tt>llvm_stop_multithreaded()</tt>.
2432 Note that, if you use scope-based shutdown, you can use the
2433 <tt>llvm_shutdown_obj</tt> class, which calls <tt>llvm_shutdown()</tt> in its
2437 <!-- ======================================================================= -->
2438 <div class="doc_subsection">
2439 <a name="managedstatic">Lazy Initialization with <tt>ManagedStatic</tt></a>
2442 <div class="doc_text">
2444 <tt>ManagedStatic</tt> is a utility class in LLVM used to implement static
2445 initialization of static resources, such as the global type tables. Before the
2446 invocation of <tt>llvm_shutdown()</tt>, it implements a simple lazy
2447 initialization scheme. Once <tt>llvm_start_multithreaded()</tt> returns,
2448 however, it uses double-checked locking to implement thread-safe lazy
2453 Note that, because no other threads are allowed to issue LLVM API calls before
2454 <tt>llvm_start_multithreaded()</tt> returns, it is possible to have
2455 <tt>ManagedStatic</tt>s of <tt>llvm::sys::Mutex</tt>s.
2459 The <tt>llvm_acquire_global_lock()</tt> and <tt>llvm_release_global_lock</tt>
2460 APIs provide access to the global lock used to implement the double-checked
2461 locking for lazy initialization. These should only be used internally to LLVM,
2462 and only if you know what you're doing!
2466 <!-- ======================================================================= -->
2467 <div class="doc_subsection">
2468 <a name="llvmcontext">Achieving Isolation with <tt>LLVMContext</tt></a>
2471 <div class="doc_text">
2473 <tt>LLVMContext</tt> is an opaque class in the LLVM API which clients can use
2474 to operate multiple, isolated instances of LLVM concurrently within the same
2475 address space. For instance, in a hypothetical compile-server, the compilation
2476 of an individual translation unit is conceptually independent from all the
2477 others, and it would be desirable to be able to compile incoming translation
2478 units concurrently on independent server threads. Fortunately,
2479 <tt>LLVMContext</tt> exists to enable just this kind of scenario!
2483 Conceptually, <tt>LLVMContext</tt> provides isolation. Every LLVM entity
2484 (<tt>Module</tt>s, <tt>Value</tt>s, <tt>Type</tt>s, <tt>Constant</tt>s, etc.)
2485 in LLVM's in-memory IR belongs to an <tt>LLVMContext</tt>. Entities in
2486 different contexts <em>cannot</em> interact with each other: <tt>Module</tt>s in
2487 different contexts cannot be linked together, <tt>Function</tt>s cannot be added
2488 to <tt>Module</tt>s in different contexts, etc. What this means is that is is
2489 safe to compile on multiple threads simultaneously, as long as no two threads
2490 operate on entities within the same context.
2494 In practice, very few places in the API require the explicit specification of a
2495 <tt>LLVMContext</tt>, other than the <tt>Type</tt> creation/lookup APIs.
2496 Because every <tt>Type</tt> carries a reference to its owning context, most
2497 other entities can determine what context they belong to by looking at their
2498 own <tt>Type</tt>. If you are adding new entities to LLVM IR, please try to
2499 maintain this interface design.
2503 For clients that do <em>not</em> require the benefits of isolation, LLVM
2504 provides a convenience API <tt>getGlobalContext()</tt>. This returns a global,
2505 lazily initialized <tt>LLVMContext</tt> that may be used in situations where
2506 isolation is not a concern.
2510 <!-- ======================================================================= -->
2511 <div class="doc_subsection">
2512 <a name="jitthreading">Threads and the JIT</a>
2515 <div class="doc_text">
2517 LLVM's "eager" JIT compiler is safe to use in threaded programs. Multiple
2518 threads can call <tt>ExecutionEngine::getPointerToFunction()</tt> or
2519 <tt>ExecutionEngine::runFunction()</tt> concurrently, and multiple threads can
2520 run code output by the JIT concurrently. The user must still ensure that only
2521 one thread accesses IR in a given <tt>LLVMContext</tt> while another thread
2522 might be modifying it. One way to do that is to always hold the JIT lock while
2523 accessing IR outside the JIT (the JIT <em>modifies</em> the IR by adding
2524 <tt>CallbackVH</tt>s). Another way is to only
2525 call <tt>getPointerToFunction()</tt> from the <tt>LLVMContext</tt>'s thread.
2528 <p>When the JIT is configured to compile lazily (using
2529 <tt>ExecutionEngine::DisableLazyCompilation(false)</tt>), there is currently a
2530 <a href="http://llvm.org/bugs/show_bug.cgi?id=5184">race condition</a> in
2531 updating call sites after a function is lazily-jitted. It's still possible to
2532 use the lazy JIT in a threaded program if you ensure that only one thread at a
2533 time can call any particular lazy stub and that the JIT lock guards any IR
2534 access, but we suggest using only the eager JIT in threaded programs.
2538 <!-- *********************************************************************** -->
2539 <div class="doc_section">
2540 <a name="advanced">Advanced Topics</a>
2542 <!-- *********************************************************************** -->
2544 <div class="doc_text">
2546 This section describes some of the advanced or obscure API's that most clients
2547 do not need to be aware of. These API's tend manage the inner workings of the
2548 LLVM system, and only need to be accessed in unusual circumstances.
2552 <!-- ======================================================================= -->
2553 <div class="doc_subsection">
2554 <a name="TypeResolve">LLVM Type Resolution</a>
2557 <div class="doc_text">
2560 The LLVM type system has a very simple goal: allow clients to compare types for
2561 structural equality with a simple pointer comparison (aka a shallow compare).
2562 This goal makes clients much simpler and faster, and is used throughout the LLVM
2567 Unfortunately achieving this goal is not a simple matter. In particular,
2568 recursive types and late resolution of opaque types makes the situation very
2569 difficult to handle. Fortunately, for the most part, our implementation makes
2570 most clients able to be completely unaware of the nasty internal details. The
2571 primary case where clients are exposed to the inner workings of it are when
2572 building a recursive type. In addition to this case, the LLVM bitcode reader,
2573 assembly parser, and linker also have to be aware of the inner workings of this
2578 For our purposes below, we need three concepts. First, an "Opaque Type" is
2579 exactly as defined in the <a href="LangRef.html#t_opaque">language
2580 reference</a>. Second an "Abstract Type" is any type which includes an
2581 opaque type as part of its type graph (for example "<tt>{ opaque, i32 }</tt>").
2582 Third, a concrete type is a type that is not an abstract type (e.g. "<tt>{ i32,
2588 <!-- ______________________________________________________________________ -->
2589 <div class="doc_subsubsection">
2590 <a name="BuildRecType">Basic Recursive Type Construction</a>
2593 <div class="doc_text">
2596 Because the most common question is "how do I build a recursive type with LLVM",
2597 we answer it now and explain it as we go. Here we include enough to cause this
2598 to be emitted to an output .ll file:
2601 <div class="doc_code">
2603 %mylist = type { %mylist*, i32 }
2608 To build this, use the following LLVM APIs:
2611 <div class="doc_code">
2613 // <i>Create the initial outer struct</i>
2614 <a href="#PATypeHolder">PATypeHolder</a> StructTy = OpaqueType::get();
2615 std::vector<const Type*> Elts;
2616 Elts.push_back(PointerType::getUnqual(StructTy));
2617 Elts.push_back(Type::Int32Ty);
2618 StructType *NewSTy = StructType::get(Elts);
2620 // <i>At this point, NewSTy = "{ opaque*, i32 }". Tell VMCore that</i>
2621 // <i>the struct and the opaque type are actually the same.</i>
2622 cast<OpaqueType>(StructTy.get())-><a href="#refineAbstractTypeTo">refineAbstractTypeTo</a>(NewSTy);
2624 // <i>NewSTy is potentially invalidated, but StructTy (a <a href="#PATypeHolder">PATypeHolder</a>) is</i>
2625 // <i>kept up-to-date</i>
2626 NewSTy = cast<StructType>(StructTy.get());
2628 // <i>Add a name for the type to the module symbol table (optional)</i>
2629 MyModule->addTypeName("mylist", NewSTy);
2634 This code shows the basic approach used to build recursive types: build a
2635 non-recursive type using 'opaque', then use type unification to close the cycle.
2636 The type unification step is performed by the <tt><a
2637 href="#refineAbstractTypeTo">refineAbstractTypeTo</a></tt> method, which is
2638 described next. After that, we describe the <a
2639 href="#PATypeHolder">PATypeHolder class</a>.
2644 <!-- ______________________________________________________________________ -->
2645 <div class="doc_subsubsection">
2646 <a name="refineAbstractTypeTo">The <tt>refineAbstractTypeTo</tt> method</a>
2649 <div class="doc_text">
2651 The <tt>refineAbstractTypeTo</tt> method starts the type unification process.
2652 While this method is actually a member of the DerivedType class, it is most
2653 often used on OpaqueType instances. Type unification is actually a recursive
2654 process. After unification, types can become structurally isomorphic to
2655 existing types, and all duplicates are deleted (to preserve pointer equality).
2659 In the example above, the OpaqueType object is definitely deleted.
2660 Additionally, if there is an "{ \2*, i32}" type already created in the system,
2661 the pointer and struct type created are <b>also</b> deleted. Obviously whenever
2662 a type is deleted, any "Type*" pointers in the program are invalidated. As
2663 such, it is safest to avoid having <i>any</i> "Type*" pointers to abstract types
2664 live across a call to <tt>refineAbstractTypeTo</tt> (note that non-abstract
2665 types can never move or be deleted). To deal with this, the <a
2666 href="#PATypeHolder">PATypeHolder</a> class is used to maintain a stable
2667 reference to a possibly refined type, and the <a
2668 href="#AbstractTypeUser">AbstractTypeUser</a> class is used to update more
2669 complex datastructures.
2674 <!-- ______________________________________________________________________ -->
2675 <div class="doc_subsubsection">
2676 <a name="PATypeHolder">The PATypeHolder Class</a>
2679 <div class="doc_text">
2681 PATypeHolder is a form of a "smart pointer" for Type objects. When VMCore
2682 happily goes about nuking types that become isomorphic to existing types, it
2683 automatically updates all PATypeHolder objects to point to the new type. In the
2684 example above, this allows the code to maintain a pointer to the resultant
2685 resolved recursive type, even though the Type*'s are potentially invalidated.
2689 PATypeHolder is an extremely light-weight object that uses a lazy union-find
2690 implementation to update pointers. For example the pointer from a Value to its
2691 Type is maintained by PATypeHolder objects.
2696 <!-- ______________________________________________________________________ -->
2697 <div class="doc_subsubsection">
2698 <a name="AbstractTypeUser">The AbstractTypeUser Class</a>
2701 <div class="doc_text">
2704 Some data structures need more to perform more complex updates when types get
2705 resolved. To support this, a class can derive from the AbstractTypeUser class.
2707 allows it to get callbacks when certain types are resolved. To register to get
2708 callbacks for a particular type, the DerivedType::{add/remove}AbstractTypeUser
2709 methods can be called on a type. Note that these methods only work for <i>
2710 abstract</i> types. Concrete types (those that do not include any opaque
2711 objects) can never be refined.
2716 <!-- ======================================================================= -->
2717 <div class="doc_subsection">
2718 <a name="SymbolTable">The <tt>ValueSymbolTable</tt> and
2719 <tt>TypeSymbolTable</tt> classes</a>
2722 <div class="doc_text">
2723 <p>The <tt><a href="http://llvm.org/doxygen/classllvm_1_1ValueSymbolTable.html">
2724 ValueSymbolTable</a></tt> class provides a symbol table that the <a
2725 href="#Function"><tt>Function</tt></a> and <a href="#Module">
2726 <tt>Module</tt></a> classes use for naming value definitions. The symbol table
2727 can provide a name for any <a href="#Value"><tt>Value</tt></a>.
2728 The <tt><a href="http://llvm.org/doxygen/classllvm_1_1TypeSymbolTable.html">
2729 TypeSymbolTable</a></tt> class is used by the <tt>Module</tt> class to store
2730 names for types.</p>
2732 <p>Note that the <tt>SymbolTable</tt> class should not be directly accessed
2733 by most clients. It should only be used when iteration over the symbol table
2734 names themselves are required, which is very special purpose. Note that not
2736 <tt><a href="#Value">Value</a></tt>s have names, and those without names (i.e. they have
2737 an empty name) do not exist in the symbol table.
2740 <p>These symbol tables support iteration over the values/types in the symbol
2741 table with <tt>begin/end/iterator</tt> and supports querying to see if a
2742 specific name is in the symbol table (with <tt>lookup</tt>). The
2743 <tt>ValueSymbolTable</tt> class exposes no public mutator methods, instead,
2744 simply call <tt>setName</tt> on a value, which will autoinsert it into the
2745 appropriate symbol table. For types, use the Module::addTypeName method to
2746 insert entries into the symbol table.</p>
2752 <!-- ======================================================================= -->
2753 <div class="doc_subsection">
2754 <a name="UserLayout">The <tt>User</tt> and owned <tt>Use</tt> classes' memory layout</a>
2757 <div class="doc_text">
2758 <p>The <tt><a href="http://llvm.org/doxygen/classllvm_1_1User.html">
2759 User</a></tt> class provides a basis for expressing the ownership of <tt>User</tt>
2760 towards other <tt><a href="http://llvm.org/doxygen/classllvm_1_1Value.html">
2761 Value</a></tt>s. The <tt><a href="http://llvm.org/doxygen/classllvm_1_1Use.html">
2762 Use</a></tt> helper class is employed to do the bookkeeping and to facilitate <i>O(1)</i>
2763 addition and removal.</p>
2765 <!-- ______________________________________________________________________ -->
2766 <div class="doc_subsubsection">
2767 <a name="Use2User">Interaction and relationship between <tt>User</tt> and <tt>Use</tt> objects</a>
2770 <div class="doc_text">
2772 A subclass of <tt>User</tt> can choose between incorporating its <tt>Use</tt> objects
2773 or refer to them out-of-line by means of a pointer. A mixed variant
2774 (some <tt>Use</tt>s inline others hung off) is impractical and breaks the invariant
2775 that the <tt>Use</tt> objects belonging to the same <tt>User</tt> form a contiguous array.
2780 We have 2 different layouts in the <tt>User</tt> (sub)classes:
2783 The <tt>Use</tt> object(s) are inside (resp. at fixed offset) of the <tt>User</tt>
2784 object and there are a fixed number of them.</p>
2787 The <tt>Use</tt> object(s) are referenced by a pointer to an
2788 array from the <tt>User</tt> object and there may be a variable
2792 As of v2.4 each layout still possesses a direct pointer to the
2793 start of the array of <tt>Use</tt>s. Though not mandatory for layout a),
2794 we stick to this redundancy for the sake of simplicity.
2795 The <tt>User</tt> object also stores the number of <tt>Use</tt> objects it
2796 has. (Theoretically this information can also be calculated
2797 given the scheme presented below.)</p>
2799 Special forms of allocation operators (<tt>operator new</tt>)
2800 enforce the following memory layouts:</p>
2803 <li><p>Layout a) is modelled by prepending the <tt>User</tt> object by the <tt>Use[]</tt> array.</p>
2806 ...---.---.---.---.-------...
2807 | P | P | P | P | User
2808 '''---'---'---'---'-------'''
2811 <li><p>Layout b) is modelled by pointing at the <tt>Use[]</tt> array.</p>
2823 <i>(In the above figures '<tt>P</tt>' stands for the <tt>Use**</tt> that
2824 is stored in each <tt>Use</tt> object in the member <tt>Use::Prev</tt>)</i>
2826 <!-- ______________________________________________________________________ -->
2827 <div class="doc_subsubsection">
2828 <a name="Waymarking">The waymarking algorithm</a>
2831 <div class="doc_text">
2833 Since the <tt>Use</tt> objects are deprived of the direct (back)pointer to
2834 their <tt>User</tt> objects, there must be a fast and exact method to
2835 recover it. This is accomplished by the following scheme:</p>
2838 A bit-encoding in the 2 LSBits (least significant bits) of the <tt>Use::Prev</tt> allows to find the
2839 start of the <tt>User</tt> object:
2841 <li><tt>00</tt> —> binary digit 0</li>
2842 <li><tt>01</tt> —> binary digit 1</li>
2843 <li><tt>10</tt> —> stop and calculate (<tt>s</tt>)</li>
2844 <li><tt>11</tt> —> full stop (<tt>S</tt>)</li>
2847 Given a <tt>Use*</tt>, all we have to do is to walk till we get
2848 a stop and we either have a <tt>User</tt> immediately behind or
2849 we have to walk to the next stop picking up digits
2850 and calculating the offset:</p>
2852 .---.---.---.---.---.---.---.---.---.---.---.---.---.---.---.---.----------------
2853 | 1 | s | 1 | 0 | 1 | 0 | s | 1 | 1 | 0 | s | 1 | 1 | s | 1 | S | User (or User*)
2854 '---'---'---'---'---'---'---'---'---'---'---'---'---'---'---'---'----------------
2855 |+15 |+10 |+6 |+3 |+1
2858 | | |______________________>
2859 | |______________________________________>
2860 |__________________________________________________________>
2863 Only the significant number of bits need to be stored between the
2864 stops, so that the <i>worst case is 20 memory accesses</i> when there are
2865 1000 <tt>Use</tt> objects associated with a <tt>User</tt>.</p>
2867 <!-- ______________________________________________________________________ -->
2868 <div class="doc_subsubsection">
2869 <a name="ReferenceImpl">Reference implementation</a>
2872 <div class="doc_text">
2874 The following literate Haskell fragment demonstrates the concept:</p>
2877 <div class="doc_code">
2879 > import Test.QuickCheck
2881 > digits :: Int -> [Char] -> [Char]
2882 > digits 0 acc = '0' : acc
2883 > digits 1 acc = '1' : acc
2884 > digits n acc = digits (n `div` 2) $ digits (n `mod` 2) acc
2886 > dist :: Int -> [Char] -> [Char]
2889 > dist 1 acc = let r = dist 0 acc in 's' : digits (length r) r
2890 > dist n acc = dist (n - 1) $ dist 1 acc
2892 > takeLast n ss = reverse $ take n $ reverse ss
2894 > test = takeLast 40 $ dist 20 []
2899 Printing <test> gives: <tt>"1s100000s11010s10100s1111s1010s110s11s1S"</tt></p>
2901 The reverse algorithm computes the length of the string just by examining
2902 a certain prefix:</p>
2904 <div class="doc_code">
2906 > pref :: [Char] -> Int
2908 > pref ('s':'1':rest) = decode 2 1 rest
2909 > pref (_:rest) = 1 + pref rest
2911 > decode walk acc ('0':rest) = decode (walk + 1) (acc * 2) rest
2912 > decode walk acc ('1':rest) = decode (walk + 1) (acc * 2 + 1) rest
2913 > decode walk acc _ = walk + acc
2918 Now, as expected, printing <pref test> gives <tt>40</tt>.</p>
2920 We can <i>quickCheck</i> this with following property:</p>
2922 <div class="doc_code">
2924 > testcase = dist 2000 []
2925 > testcaseLength = length testcase
2927 > identityProp n = n > 0 && n <= testcaseLength ==> length arr == pref arr
2928 > where arr = takeLast n testcase
2933 As expected <quickCheck identityProp> gives:</p>
2936 *Main> quickCheck identityProp
2937 OK, passed 100 tests.
2940 Let's be a bit more exhaustive:</p>
2942 <div class="doc_code">
2945 > deepCheck p = check (defaultConfig { configMaxTest = 500 }) p
2950 And here is the result of <deepCheck identityProp>:</p>
2953 *Main> deepCheck identityProp
2954 OK, passed 500 tests.
2957 <!-- ______________________________________________________________________ -->
2958 <div class="doc_subsubsection">
2959 <a name="Tagging">Tagging considerations</a>
2963 To maintain the invariant that the 2 LSBits of each <tt>Use**</tt> in <tt>Use</tt>
2964 never change after being set up, setters of <tt>Use::Prev</tt> must re-tag the
2965 new <tt>Use**</tt> on every modification. Accordingly getters must strip the
2968 For layout b) instead of the <tt>User</tt> we find a pointer (<tt>User*</tt> with LSBit set).
2969 Following this pointer brings us to the <tt>User</tt>. A portable trick ensures
2970 that the first bytes of <tt>User</tt> (if interpreted as a pointer) never has
2971 the LSBit set. (Portability is relying on the fact that all known compilers place the
2972 <tt>vptr</tt> in the first word of the instances.)</p>
2976 <!-- *********************************************************************** -->
2977 <div class="doc_section">
2978 <a name="coreclasses">The Core LLVM Class Hierarchy Reference </a>
2980 <!-- *********************************************************************** -->
2982 <div class="doc_text">
2983 <p><tt>#include "<a href="/doxygen/Type_8h-source.html">llvm/Type.h</a>"</tt>
2984 <br>doxygen info: <a href="/doxygen/classllvm_1_1Type.html">Type Class</a></p>
2986 <p>The Core LLVM classes are the primary means of representing the program
2987 being inspected or transformed. The core LLVM classes are defined in
2988 header files in the <tt>include/llvm/</tt> directory, and implemented in
2989 the <tt>lib/VMCore</tt> directory.</p>
2993 <!-- ======================================================================= -->
2994 <div class="doc_subsection">
2995 <a name="Type">The <tt>Type</tt> class and Derived Types</a>
2998 <div class="doc_text">
3000 <p><tt>Type</tt> is a superclass of all type classes. Every <tt>Value</tt> has
3001 a <tt>Type</tt>. <tt>Type</tt> cannot be instantiated directly but only
3002 through its subclasses. Certain primitive types (<tt>VoidType</tt>,
3003 <tt>LabelType</tt>, <tt>FloatType</tt> and <tt>DoubleType</tt>) have hidden
3004 subclasses. They are hidden because they offer no useful functionality beyond
3005 what the <tt>Type</tt> class offers except to distinguish themselves from
3006 other subclasses of <tt>Type</tt>.</p>
3007 <p>All other types are subclasses of <tt>DerivedType</tt>. Types can be
3008 named, but this is not a requirement. There exists exactly
3009 one instance of a given shape at any one time. This allows type equality to
3010 be performed with address equality of the Type Instance. That is, given two
3011 <tt>Type*</tt> values, the types are identical if the pointers are identical.
3015 <!-- _______________________________________________________________________ -->
3016 <div class="doc_subsubsection">
3017 <a name="m_Type">Important Public Methods</a>
3020 <div class="doc_text">
3023 <li><tt>bool isIntegerTy() const</tt>: Returns true for any integer type.</li>
3025 <li><tt>bool isFloatingPointTy()</tt>: Return true if this is one of the five
3026 floating point types.</li>
3028 <li><tt>bool isAbstract()</tt>: Return true if the type is abstract (contains
3029 an OpaqueType anywhere in its definition).</li>
3031 <li><tt>bool isSized()</tt>: Return true if the type has known size. Things
3032 that don't have a size are abstract types, labels and void.</li>
3037 <!-- _______________________________________________________________________ -->
3038 <div class="doc_subsubsection">
3039 <a name="derivedtypes">Important Derived Types</a>
3041 <div class="doc_text">
3043 <dt><tt>IntegerType</tt></dt>
3044 <dd>Subclass of DerivedType that represents integer types of any bit width.
3045 Any bit width between <tt>IntegerType::MIN_INT_BITS</tt> (1) and
3046 <tt>IntegerType::MAX_INT_BITS</tt> (~8 million) can be represented.
3048 <li><tt>static const IntegerType* get(unsigned NumBits)</tt>: get an integer
3049 type of a specific bit width.</li>
3050 <li><tt>unsigned getBitWidth() const</tt>: Get the bit width of an integer
3054 <dt><tt>SequentialType</tt></dt>
3055 <dd>This is subclassed by ArrayType and PointerType
3057 <li><tt>const Type * getElementType() const</tt>: Returns the type of each
3058 of the elements in the sequential type. </li>
3061 <dt><tt>ArrayType</tt></dt>
3062 <dd>This is a subclass of SequentialType and defines the interface for array
3065 <li><tt>unsigned getNumElements() const</tt>: Returns the number of
3066 elements in the array. </li>
3069 <dt><tt>PointerType</tt></dt>
3070 <dd>Subclass of SequentialType for pointer types.</dd>
3071 <dt><tt>VectorType</tt></dt>
3072 <dd>Subclass of SequentialType for vector types. A
3073 vector type is similar to an ArrayType but is distinguished because it is
3074 a first class type whereas ArrayType is not. Vector types are used for
3075 vector operations and are usually small vectors of of an integer or floating
3077 <dt><tt>StructType</tt></dt>
3078 <dd>Subclass of DerivedTypes for struct types.</dd>
3079 <dt><tt><a name="FunctionType">FunctionType</a></tt></dt>
3080 <dd>Subclass of DerivedTypes for function types.
3082 <li><tt>bool isVarArg() const</tt>: Returns true if it's a vararg
3084 <li><tt> const Type * getReturnType() const</tt>: Returns the
3085 return type of the function.</li>
3086 <li><tt>const Type * getParamType (unsigned i)</tt>: Returns
3087 the type of the ith parameter.</li>
3088 <li><tt> const unsigned getNumParams() const</tt>: Returns the
3089 number of formal parameters.</li>
3092 <dt><tt>OpaqueType</tt></dt>
3093 <dd>Sublcass of DerivedType for abstract types. This class
3094 defines no content and is used as a placeholder for some other type. Note
3095 that OpaqueType is used (temporarily) during type resolution for forward
3096 references of types. Once the referenced type is resolved, the OpaqueType
3097 is replaced with the actual type. OpaqueType can also be used for data
3098 abstraction. At link time opaque types can be resolved to actual types
3099 of the same name.</dd>
3105 <!-- ======================================================================= -->
3106 <div class="doc_subsection">
3107 <a name="Module">The <tt>Module</tt> class</a>
3110 <div class="doc_text">
3113 href="/doxygen/Module_8h-source.html">llvm/Module.h</a>"</tt><br> doxygen info:
3114 <a href="/doxygen/classllvm_1_1Module.html">Module Class</a></p>
3116 <p>The <tt>Module</tt> class represents the top level structure present in LLVM
3117 programs. An LLVM module is effectively either a translation unit of the
3118 original program or a combination of several translation units merged by the
3119 linker. The <tt>Module</tt> class keeps track of a list of <a
3120 href="#Function"><tt>Function</tt></a>s, a list of <a
3121 href="#GlobalVariable"><tt>GlobalVariable</tt></a>s, and a <a
3122 href="#SymbolTable"><tt>SymbolTable</tt></a>. Additionally, it contains a few
3123 helpful member functions that try to make common operations easy.</p>
3127 <!-- _______________________________________________________________________ -->
3128 <div class="doc_subsubsection">
3129 <a name="m_Module">Important Public Members of the <tt>Module</tt> class</a>
3132 <div class="doc_text">
3135 <li><tt>Module::Module(std::string name = "")</tt></li>
3138 <p>Constructing a <a href="#Module">Module</a> is easy. You can optionally
3139 provide a name for it (probably based on the name of the translation unit).</p>
3142 <li><tt>Module::iterator</tt> - Typedef for function list iterator<br>
3143 <tt>Module::const_iterator</tt> - Typedef for const_iterator.<br>
3145 <tt>begin()</tt>, <tt>end()</tt>
3146 <tt>size()</tt>, <tt>empty()</tt>
3148 <p>These are forwarding methods that make it easy to access the contents of
3149 a <tt>Module</tt> object's <a href="#Function"><tt>Function</tt></a>
3152 <li><tt>Module::FunctionListType &getFunctionList()</tt>
3154 <p> Returns the list of <a href="#Function"><tt>Function</tt></a>s. This is
3155 necessary to use when you need to update the list or perform a complex
3156 action that doesn't have a forwarding method.</p>
3158 <p><!-- Global Variable --></p></li>
3164 <li><tt>Module::global_iterator</tt> - Typedef for global variable list iterator<br>
3166 <tt>Module::const_global_iterator</tt> - Typedef for const_iterator.<br>
3168 <tt>global_begin()</tt>, <tt>global_end()</tt>
3169 <tt>global_size()</tt>, <tt>global_empty()</tt>
3171 <p> These are forwarding methods that make it easy to access the contents of
3172 a <tt>Module</tt> object's <a
3173 href="#GlobalVariable"><tt>GlobalVariable</tt></a> list.</p></li>
3175 <li><tt>Module::GlobalListType &getGlobalList()</tt>
3177 <p>Returns the list of <a
3178 href="#GlobalVariable"><tt>GlobalVariable</tt></a>s. This is necessary to
3179 use when you need to update the list or perform a complex action that
3180 doesn't have a forwarding method.</p>
3182 <p><!-- Symbol table stuff --> </p></li>
3188 <li><tt><a href="#SymbolTable">SymbolTable</a> *getSymbolTable()</tt>
3190 <p>Return a reference to the <a href="#SymbolTable"><tt>SymbolTable</tt></a>
3191 for this <tt>Module</tt>.</p>
3193 <p><!-- Convenience methods --></p></li>
3199 <li><tt><a href="#Function">Function</a> *getFunction(const std::string
3200 &Name, const <a href="#FunctionType">FunctionType</a> *Ty)</tt>
3202 <p>Look up the specified function in the <tt>Module</tt> <a
3203 href="#SymbolTable"><tt>SymbolTable</tt></a>. If it does not exist, return
3204 <tt>null</tt>.</p></li>
3206 <li><tt><a href="#Function">Function</a> *getOrInsertFunction(const
3207 std::string &Name, const <a href="#FunctionType">FunctionType</a> *T)</tt>
3209 <p>Look up the specified function in the <tt>Module</tt> <a
3210 href="#SymbolTable"><tt>SymbolTable</tt></a>. If it does not exist, add an
3211 external declaration for the function and return it.</p></li>
3213 <li><tt>std::string getTypeName(const <a href="#Type">Type</a> *Ty)</tt>
3215 <p>If there is at least one entry in the <a
3216 href="#SymbolTable"><tt>SymbolTable</tt></a> for the specified <a
3217 href="#Type"><tt>Type</tt></a>, return it. Otherwise return the empty
3220 <li><tt>bool addTypeName(const std::string &Name, const <a
3221 href="#Type">Type</a> *Ty)</tt>
3223 <p>Insert an entry in the <a href="#SymbolTable"><tt>SymbolTable</tt></a>
3224 mapping <tt>Name</tt> to <tt>Ty</tt>. If there is already an entry for this
3225 name, true is returned and the <a
3226 href="#SymbolTable"><tt>SymbolTable</tt></a> is not modified.</p></li>
3232 <!-- ======================================================================= -->
3233 <div class="doc_subsection">
3234 <a name="Value">The <tt>Value</tt> class</a>
3237 <div class="doc_text">
3239 <p><tt>#include "<a href="/doxygen/Value_8h-source.html">llvm/Value.h</a>"</tt>
3241 doxygen info: <a href="/doxygen/classllvm_1_1Value.html">Value Class</a></p>
3243 <p>The <tt>Value</tt> class is the most important class in the LLVM Source
3244 base. It represents a typed value that may be used (among other things) as an
3245 operand to an instruction. There are many different types of <tt>Value</tt>s,
3246 such as <a href="#Constant"><tt>Constant</tt></a>s,<a
3247 href="#Argument"><tt>Argument</tt></a>s. Even <a
3248 href="#Instruction"><tt>Instruction</tt></a>s and <a
3249 href="#Function"><tt>Function</tt></a>s are <tt>Value</tt>s.</p>
3251 <p>A particular <tt>Value</tt> may be used many times in the LLVM representation
3252 for a program. For example, an incoming argument to a function (represented
3253 with an instance of the <a href="#Argument">Argument</a> class) is "used" by
3254 every instruction in the function that references the argument. To keep track
3255 of this relationship, the <tt>Value</tt> class keeps a list of all of the <a
3256 href="#User"><tt>User</tt></a>s that is using it (the <a
3257 href="#User"><tt>User</tt></a> class is a base class for all nodes in the LLVM
3258 graph that can refer to <tt>Value</tt>s). This use list is how LLVM represents
3259 def-use information in the program, and is accessible through the <tt>use_</tt>*
3260 methods, shown below.</p>
3262 <p>Because LLVM is a typed representation, every LLVM <tt>Value</tt> is typed,
3263 and this <a href="#Type">Type</a> is available through the <tt>getType()</tt>
3264 method. In addition, all LLVM values can be named. The "name" of the
3265 <tt>Value</tt> is a symbolic string printed in the LLVM code:</p>
3267 <div class="doc_code">
3269 %<b>foo</b> = add i32 1, 2
3273 <p><a name="nameWarning">The name of this instruction is "foo".</a> <b>NOTE</b>
3274 that the name of any value may be missing (an empty string), so names should
3275 <b>ONLY</b> be used for debugging (making the source code easier to read,
3276 debugging printouts), they should not be used to keep track of values or map
3277 between them. For this purpose, use a <tt>std::map</tt> of pointers to the
3278 <tt>Value</tt> itself instead.</p>
3280 <p>One important aspect of LLVM is that there is no distinction between an SSA
3281 variable and the operation that produces it. Because of this, any reference to
3282 the value produced by an instruction (or the value available as an incoming
3283 argument, for example) is represented as a direct pointer to the instance of
3285 represents this value. Although this may take some getting used to, it
3286 simplifies the representation and makes it easier to manipulate.</p>
3290 <!-- _______________________________________________________________________ -->
3291 <div class="doc_subsubsection">
3292 <a name="m_Value">Important Public Members of the <tt>Value</tt> class</a>
3295 <div class="doc_text">
3298 <li><tt>Value::use_iterator</tt> - Typedef for iterator over the
3300 <tt>Value::const_use_iterator</tt> - Typedef for const_iterator over
3302 <tt>unsigned use_size()</tt> - Returns the number of users of the
3304 <tt>bool use_empty()</tt> - Returns true if there are no users.<br>
3305 <tt>use_iterator use_begin()</tt> - Get an iterator to the start of
3307 <tt>use_iterator use_end()</tt> - Get an iterator to the end of the
3309 <tt><a href="#User">User</a> *use_back()</tt> - Returns the last
3310 element in the list.
3311 <p> These methods are the interface to access the def-use
3312 information in LLVM. As with all other iterators in LLVM, the naming
3313 conventions follow the conventions defined by the <a href="#stl">STL</a>.</p>
3315 <li><tt><a href="#Type">Type</a> *getType() const</tt>
3316 <p>This method returns the Type of the Value.</p>
3318 <li><tt>bool hasName() const</tt><br>
3319 <tt>std::string getName() const</tt><br>
3320 <tt>void setName(const std::string &Name)</tt>
3321 <p> This family of methods is used to access and assign a name to a <tt>Value</tt>,
3322 be aware of the <a href="#nameWarning">precaution above</a>.</p>
3324 <li><tt>void replaceAllUsesWith(Value *V)</tt>
3326 <p>This method traverses the use list of a <tt>Value</tt> changing all <a
3327 href="#User"><tt>User</tt>s</a> of the current value to refer to
3328 "<tt>V</tt>" instead. For example, if you detect that an instruction always
3329 produces a constant value (for example through constant folding), you can
3330 replace all uses of the instruction with the constant like this:</p>
3332 <div class="doc_code">
3334 Inst->replaceAllUsesWith(ConstVal);
3342 <!-- ======================================================================= -->
3343 <div class="doc_subsection">
3344 <a name="User">The <tt>User</tt> class</a>
3347 <div class="doc_text">
3350 <tt>#include "<a href="/doxygen/User_8h-source.html">llvm/User.h</a>"</tt><br>
3351 doxygen info: <a href="/doxygen/classllvm_1_1User.html">User Class</a><br>
3352 Superclass: <a href="#Value"><tt>Value</tt></a></p>
3354 <p>The <tt>User</tt> class is the common base class of all LLVM nodes that may
3355 refer to <a href="#Value"><tt>Value</tt></a>s. It exposes a list of "Operands"
3356 that are all of the <a href="#Value"><tt>Value</tt></a>s that the User is
3357 referring to. The <tt>User</tt> class itself is a subclass of
3360 <p>The operands of a <tt>User</tt> point directly to the LLVM <a
3361 href="#Value"><tt>Value</tt></a> that it refers to. Because LLVM uses Static
3362 Single Assignment (SSA) form, there can only be one definition referred to,
3363 allowing this direct connection. This connection provides the use-def
3364 information in LLVM.</p>
3368 <!-- _______________________________________________________________________ -->
3369 <div class="doc_subsubsection">
3370 <a name="m_User">Important Public Members of the <tt>User</tt> class</a>
3373 <div class="doc_text">
3375 <p>The <tt>User</tt> class exposes the operand list in two ways: through
3376 an index access interface and through an iterator based interface.</p>
3379 <li><tt>Value *getOperand(unsigned i)</tt><br>
3380 <tt>unsigned getNumOperands()</tt>
3381 <p> These two methods expose the operands of the <tt>User</tt> in a
3382 convenient form for direct access.</p></li>
3384 <li><tt>User::op_iterator</tt> - Typedef for iterator over the operand
3386 <tt>op_iterator op_begin()</tt> - Get an iterator to the start of
3387 the operand list.<br>
3388 <tt>op_iterator op_end()</tt> - Get an iterator to the end of the
3390 <p> Together, these methods make up the iterator based interface to
3391 the operands of a <tt>User</tt>.</p></li>
3396 <!-- ======================================================================= -->
3397 <div class="doc_subsection">
3398 <a name="Instruction">The <tt>Instruction</tt> class</a>
3401 <div class="doc_text">
3403 <p><tt>#include "</tt><tt><a
3404 href="/doxygen/Instruction_8h-source.html">llvm/Instruction.h</a>"</tt><br>
3405 doxygen info: <a href="/doxygen/classllvm_1_1Instruction.html">Instruction Class</a><br>
3406 Superclasses: <a href="#User"><tt>User</tt></a>, <a
3407 href="#Value"><tt>Value</tt></a></p>
3409 <p>The <tt>Instruction</tt> class is the common base class for all LLVM
3410 instructions. It provides only a few methods, but is a very commonly used
3411 class. The primary data tracked by the <tt>Instruction</tt> class itself is the
3412 opcode (instruction type) and the parent <a
3413 href="#BasicBlock"><tt>BasicBlock</tt></a> the <tt>Instruction</tt> is embedded
3414 into. To represent a specific type of instruction, one of many subclasses of
3415 <tt>Instruction</tt> are used.</p>
3417 <p> Because the <tt>Instruction</tt> class subclasses the <a
3418 href="#User"><tt>User</tt></a> class, its operands can be accessed in the same
3419 way as for other <a href="#User"><tt>User</tt></a>s (with the
3420 <tt>getOperand()</tt>/<tt>getNumOperands()</tt> and
3421 <tt>op_begin()</tt>/<tt>op_end()</tt> methods).</p> <p> An important file for
3422 the <tt>Instruction</tt> class is the <tt>llvm/Instruction.def</tt> file. This
3423 file contains some meta-data about the various different types of instructions
3424 in LLVM. It describes the enum values that are used as opcodes (for example
3425 <tt>Instruction::Add</tt> and <tt>Instruction::ICmp</tt>), as well as the
3426 concrete sub-classes of <tt>Instruction</tt> that implement the instruction (for
3427 example <tt><a href="#BinaryOperator">BinaryOperator</a></tt> and <tt><a
3428 href="#CmpInst">CmpInst</a></tt>). Unfortunately, the use of macros in
3429 this file confuses doxygen, so these enum values don't show up correctly in the
3430 <a href="/doxygen/classllvm_1_1Instruction.html">doxygen output</a>.</p>
3434 <!-- _______________________________________________________________________ -->
3435 <div class="doc_subsubsection">
3436 <a name="s_Instruction">Important Subclasses of the <tt>Instruction</tt>
3439 <div class="doc_text">
3441 <li><tt><a name="BinaryOperator">BinaryOperator</a></tt>
3442 <p>This subclasses represents all two operand instructions whose operands
3443 must be the same type, except for the comparison instructions.</p></li>
3444 <li><tt><a name="CastInst">CastInst</a></tt>
3445 <p>This subclass is the parent of the 12 casting instructions. It provides
3446 common operations on cast instructions.</p>
3447 <li><tt><a name="CmpInst">CmpInst</a></tt>
3448 <p>This subclass respresents the two comparison instructions,
3449 <a href="LangRef.html#i_icmp">ICmpInst</a> (integer opreands), and
3450 <a href="LangRef.html#i_fcmp">FCmpInst</a> (floating point operands).</p>
3451 <li><tt><a name="TerminatorInst">TerminatorInst</a></tt>
3452 <p>This subclass is the parent of all terminator instructions (those which
3453 can terminate a block).</p>
3457 <!-- _______________________________________________________________________ -->
3458 <div class="doc_subsubsection">
3459 <a name="m_Instruction">Important Public Members of the <tt>Instruction</tt>
3463 <div class="doc_text">
3466 <li><tt><a href="#BasicBlock">BasicBlock</a> *getParent()</tt>
3467 <p>Returns the <a href="#BasicBlock"><tt>BasicBlock</tt></a> that
3468 this <tt>Instruction</tt> is embedded into.</p></li>
3469 <li><tt>bool mayWriteToMemory()</tt>
3470 <p>Returns true if the instruction writes to memory, i.e. it is a
3471 <tt>call</tt>,<tt>free</tt>,<tt>invoke</tt>, or <tt>store</tt>.</p></li>
3472 <li><tt>unsigned getOpcode()</tt>
3473 <p>Returns the opcode for the <tt>Instruction</tt>.</p></li>
3474 <li><tt><a href="#Instruction">Instruction</a> *clone() const</tt>
3475 <p>Returns another instance of the specified instruction, identical
3476 in all ways to the original except that the instruction has no parent
3477 (ie it's not embedded into a <a href="#BasicBlock"><tt>BasicBlock</tt></a>),
3478 and it has no name</p></li>
3483 <!-- ======================================================================= -->
3484 <div class="doc_subsection">
3485 <a name="Constant">The <tt>Constant</tt> class and subclasses</a>
3488 <div class="doc_text">
3490 <p>Constant represents a base class for different types of constants. It
3491 is subclassed by ConstantInt, ConstantArray, etc. for representing
3492 the various types of Constants. <a href="#GlobalValue">GlobalValue</a> is also
3493 a subclass, which represents the address of a global variable or function.
3498 <!-- _______________________________________________________________________ -->
3499 <div class="doc_subsubsection">Important Subclasses of Constant </div>
3500 <div class="doc_text">
3502 <li>ConstantInt : This subclass of Constant represents an integer constant of
3505 <li><tt>const APInt& getValue() const</tt>: Returns the underlying
3506 value of this constant, an APInt value.</li>
3507 <li><tt>int64_t getSExtValue() const</tt>: Converts the underlying APInt
3508 value to an int64_t via sign extension. If the value (not the bit width)
3509 of the APInt is too large to fit in an int64_t, an assertion will result.
3510 For this reason, use of this method is discouraged.</li>
3511 <li><tt>uint64_t getZExtValue() const</tt>: Converts the underlying APInt
3512 value to a uint64_t via zero extension. IF the value (not the bit width)
3513 of the APInt is too large to fit in a uint64_t, an assertion will result.
3514 For this reason, use of this method is discouraged.</li>
3515 <li><tt>static ConstantInt* get(const APInt& Val)</tt>: Returns the
3516 ConstantInt object that represents the value provided by <tt>Val</tt>.
3517 The type is implied as the IntegerType that corresponds to the bit width
3518 of <tt>Val</tt>.</li>
3519 <li><tt>static ConstantInt* get(const Type *Ty, uint64_t Val)</tt>:
3520 Returns the ConstantInt object that represents the value provided by
3521 <tt>Val</tt> for integer type <tt>Ty</tt>.</li>
3524 <li>ConstantFP : This class represents a floating point constant.
3526 <li><tt>double getValue() const</tt>: Returns the underlying value of
3527 this constant. </li>
3530 <li>ConstantArray : This represents a constant array.
3532 <li><tt>const std::vector<Use> &getValues() const</tt>: Returns
3533 a vector of component constants that makeup this array. </li>
3536 <li>ConstantStruct : This represents a constant struct.
3538 <li><tt>const std::vector<Use> &getValues() const</tt>: Returns
3539 a vector of component constants that makeup this array. </li>
3542 <li>GlobalValue : This represents either a global variable or a function. In
3543 either case, the value is a constant fixed address (after linking).
3549 <!-- ======================================================================= -->
3550 <div class="doc_subsection">
3551 <a name="GlobalValue">The <tt>GlobalValue</tt> class</a>
3554 <div class="doc_text">
3557 href="/doxygen/GlobalValue_8h-source.html">llvm/GlobalValue.h</a>"</tt><br>
3558 doxygen info: <a href="/doxygen/classllvm_1_1GlobalValue.html">GlobalValue
3560 Superclasses: <a href="#Constant"><tt>Constant</tt></a>,
3561 <a href="#User"><tt>User</tt></a>, <a href="#Value"><tt>Value</tt></a></p>
3563 <p>Global values (<a href="#GlobalVariable"><tt>GlobalVariable</tt></a>s or <a
3564 href="#Function"><tt>Function</tt></a>s) are the only LLVM values that are
3565 visible in the bodies of all <a href="#Function"><tt>Function</tt></a>s.
3566 Because they are visible at global scope, they are also subject to linking with
3567 other globals defined in different translation units. To control the linking
3568 process, <tt>GlobalValue</tt>s know their linkage rules. Specifically,
3569 <tt>GlobalValue</tt>s know whether they have internal or external linkage, as
3570 defined by the <tt>LinkageTypes</tt> enumeration.</p>
3572 <p>If a <tt>GlobalValue</tt> has internal linkage (equivalent to being
3573 <tt>static</tt> in C), it is not visible to code outside the current translation
3574 unit, and does not participate in linking. If it has external linkage, it is
3575 visible to external code, and does participate in linking. In addition to
3576 linkage information, <tt>GlobalValue</tt>s keep track of which <a
3577 href="#Module"><tt>Module</tt></a> they are currently part of.</p>
3579 <p>Because <tt>GlobalValue</tt>s are memory objects, they are always referred to
3580 by their <b>address</b>. As such, the <a href="#Type"><tt>Type</tt></a> of a
3581 global is always a pointer to its contents. It is important to remember this
3582 when using the <tt>GetElementPtrInst</tt> instruction because this pointer must
3583 be dereferenced first. For example, if you have a <tt>GlobalVariable</tt> (a
3584 subclass of <tt>GlobalValue)</tt> that is an array of 24 ints, type <tt>[24 x
3585 i32]</tt>, then the <tt>GlobalVariable</tt> is a pointer to that array. Although
3586 the address of the first element of this array and the value of the
3587 <tt>GlobalVariable</tt> are the same, they have different types. The
3588 <tt>GlobalVariable</tt>'s type is <tt>[24 x i32]</tt>. The first element's type
3589 is <tt>i32.</tt> Because of this, accessing a global value requires you to
3590 dereference the pointer with <tt>GetElementPtrInst</tt> first, then its elements
3591 can be accessed. This is explained in the <a href="LangRef.html#globalvars">LLVM
3592 Language Reference Manual</a>.</p>
3596 <!-- _______________________________________________________________________ -->
3597 <div class="doc_subsubsection">
3598 <a name="m_GlobalValue">Important Public Members of the <tt>GlobalValue</tt>
3602 <div class="doc_text">
3605 <li><tt>bool hasInternalLinkage() const</tt><br>
3606 <tt>bool hasExternalLinkage() const</tt><br>
3607 <tt>void setInternalLinkage(bool HasInternalLinkage)</tt>
3608 <p> These methods manipulate the linkage characteristics of the <tt>GlobalValue</tt>.</p>
3611 <li><tt><a href="#Module">Module</a> *getParent()</tt>
3612 <p> This returns the <a href="#Module"><tt>Module</tt></a> that the
3613 GlobalValue is currently embedded into.</p></li>
3618 <!-- ======================================================================= -->
3619 <div class="doc_subsection">
3620 <a name="Function">The <tt>Function</tt> class</a>
3623 <div class="doc_text">
3626 href="/doxygen/Function_8h-source.html">llvm/Function.h</a>"</tt><br> doxygen
3627 info: <a href="/doxygen/classllvm_1_1Function.html">Function Class</a><br>
3628 Superclasses: <a href="#GlobalValue"><tt>GlobalValue</tt></a>,
3629 <a href="#Constant"><tt>Constant</tt></a>,
3630 <a href="#User"><tt>User</tt></a>,
3631 <a href="#Value"><tt>Value</tt></a></p>
3633 <p>The <tt>Function</tt> class represents a single procedure in LLVM. It is
3634 actually one of the more complex classes in the LLVM hierarchy because it must
3635 keep track of a large amount of data. The <tt>Function</tt> class keeps track
3636 of a list of <a href="#BasicBlock"><tt>BasicBlock</tt></a>s, a list of formal
3637 <a href="#Argument"><tt>Argument</tt></a>s, and a
3638 <a href="#SymbolTable"><tt>SymbolTable</tt></a>.</p>
3640 <p>The list of <a href="#BasicBlock"><tt>BasicBlock</tt></a>s is the most
3641 commonly used part of <tt>Function</tt> objects. The list imposes an implicit
3642 ordering of the blocks in the function, which indicate how the code will be
3643 laid out by the backend. Additionally, the first <a
3644 href="#BasicBlock"><tt>BasicBlock</tt></a> is the implicit entry node for the
3645 <tt>Function</tt>. It is not legal in LLVM to explicitly branch to this initial
3646 block. There are no implicit exit nodes, and in fact there may be multiple exit
3647 nodes from a single <tt>Function</tt>. If the <a
3648 href="#BasicBlock"><tt>BasicBlock</tt></a> list is empty, this indicates that
3649 the <tt>Function</tt> is actually a function declaration: the actual body of the
3650 function hasn't been linked in yet.</p>
3652 <p>In addition to a list of <a href="#BasicBlock"><tt>BasicBlock</tt></a>s, the
3653 <tt>Function</tt> class also keeps track of the list of formal <a
3654 href="#Argument"><tt>Argument</tt></a>s that the function receives. This
3655 container manages the lifetime of the <a href="#Argument"><tt>Argument</tt></a>
3656 nodes, just like the <a href="#BasicBlock"><tt>BasicBlock</tt></a> list does for
3657 the <a href="#BasicBlock"><tt>BasicBlock</tt></a>s.</p>
3659 <p>The <a href="#SymbolTable"><tt>SymbolTable</tt></a> is a very rarely used
3660 LLVM feature that is only used when you have to look up a value by name. Aside
3661 from that, the <a href="#SymbolTable"><tt>SymbolTable</tt></a> is used
3662 internally to make sure that there are not conflicts between the names of <a
3663 href="#Instruction"><tt>Instruction</tt></a>s, <a
3664 href="#BasicBlock"><tt>BasicBlock</tt></a>s, or <a
3665 href="#Argument"><tt>Argument</tt></a>s in the function body.</p>
3667 <p>Note that <tt>Function</tt> is a <a href="#GlobalValue">GlobalValue</a>
3668 and therefore also a <a href="#Constant">Constant</a>. The value of the function
3669 is its address (after linking) which is guaranteed to be constant.</p>
3672 <!-- _______________________________________________________________________ -->
3673 <div class="doc_subsubsection">
3674 <a name="m_Function">Important Public Members of the <tt>Function</tt>
3678 <div class="doc_text">
3681 <li><tt>Function(const </tt><tt><a href="#FunctionType">FunctionType</a>
3682 *Ty, LinkageTypes Linkage, const std::string &N = "", Module* Parent = 0)</tt>
3684 <p>Constructor used when you need to create new <tt>Function</tt>s to add
3685 the the program. The constructor must specify the type of the function to
3686 create and what type of linkage the function should have. The <a
3687 href="#FunctionType"><tt>FunctionType</tt></a> argument
3688 specifies the formal arguments and return value for the function. The same
3689 <a href="#FunctionType"><tt>FunctionType</tt></a> value can be used to
3690 create multiple functions. The <tt>Parent</tt> argument specifies the Module
3691 in which the function is defined. If this argument is provided, the function
3692 will automatically be inserted into that module's list of
3695 <li><tt>bool isDeclaration()</tt>
3697 <p>Return whether or not the <tt>Function</tt> has a body defined. If the
3698 function is "external", it does not have a body, and thus must be resolved
3699 by linking with a function defined in a different translation unit.</p></li>
3701 <li><tt>Function::iterator</tt> - Typedef for basic block list iterator<br>
3702 <tt>Function::const_iterator</tt> - Typedef for const_iterator.<br>
3704 <tt>begin()</tt>, <tt>end()</tt>
3705 <tt>size()</tt>, <tt>empty()</tt>
3707 <p>These are forwarding methods that make it easy to access the contents of
3708 a <tt>Function</tt> object's <a href="#BasicBlock"><tt>BasicBlock</tt></a>
3711 <li><tt>Function::BasicBlockListType &getBasicBlockList()</tt>
3713 <p>Returns the list of <a href="#BasicBlock"><tt>BasicBlock</tt></a>s. This
3714 is necessary to use when you need to update the list or perform a complex
3715 action that doesn't have a forwarding method.</p></li>
3717 <li><tt>Function::arg_iterator</tt> - Typedef for the argument list
3719 <tt>Function::const_arg_iterator</tt> - Typedef for const_iterator.<br>
3721 <tt>arg_begin()</tt>, <tt>arg_end()</tt>
3722 <tt>arg_size()</tt>, <tt>arg_empty()</tt>
3724 <p>These are forwarding methods that make it easy to access the contents of
3725 a <tt>Function</tt> object's <a href="#Argument"><tt>Argument</tt></a>
3728 <li><tt>Function::ArgumentListType &getArgumentList()</tt>
3730 <p>Returns the list of <a href="#Argument"><tt>Argument</tt></a>s. This is
3731 necessary to use when you need to update the list or perform a complex
3732 action that doesn't have a forwarding method.</p></li>
3734 <li><tt><a href="#BasicBlock">BasicBlock</a> &getEntryBlock()</tt>
3736 <p>Returns the entry <a href="#BasicBlock"><tt>BasicBlock</tt></a> for the
3737 function. Because the entry block for the function is always the first
3738 block, this returns the first block of the <tt>Function</tt>.</p></li>
3740 <li><tt><a href="#Type">Type</a> *getReturnType()</tt><br>
3741 <tt><a href="#FunctionType">FunctionType</a> *getFunctionType()</tt>
3743 <p>This traverses the <a href="#Type"><tt>Type</tt></a> of the
3744 <tt>Function</tt> and returns the return type of the function, or the <a
3745 href="#FunctionType"><tt>FunctionType</tt></a> of the actual
3748 <li><tt><a href="#SymbolTable">SymbolTable</a> *getSymbolTable()</tt>
3750 <p> Return a pointer to the <a href="#SymbolTable"><tt>SymbolTable</tt></a>
3751 for this <tt>Function</tt>.</p></li>
3756 <!-- ======================================================================= -->
3757 <div class="doc_subsection">
3758 <a name="GlobalVariable">The <tt>GlobalVariable</tt> class</a>
3761 <div class="doc_text">
3764 href="/doxygen/GlobalVariable_8h-source.html">llvm/GlobalVariable.h</a>"</tt>
3766 doxygen info: <a href="/doxygen/classllvm_1_1GlobalVariable.html">GlobalVariable
3768 Superclasses: <a href="#GlobalValue"><tt>GlobalValue</tt></a>,
3769 <a href="#Constant"><tt>Constant</tt></a>,
3770 <a href="#User"><tt>User</tt></a>,
3771 <a href="#Value"><tt>Value</tt></a></p>
3773 <p>Global variables are represented with the (surprise surprise)
3774 <tt>GlobalVariable</tt> class. Like functions, <tt>GlobalVariable</tt>s are also
3775 subclasses of <a href="#GlobalValue"><tt>GlobalValue</tt></a>, and as such are
3776 always referenced by their address (global values must live in memory, so their
3777 "name" refers to their constant address). See
3778 <a href="#GlobalValue"><tt>GlobalValue</tt></a> for more on this. Global
3779 variables may have an initial value (which must be a
3780 <a href="#Constant"><tt>Constant</tt></a>), and if they have an initializer,
3781 they may be marked as "constant" themselves (indicating that their contents
3782 never change at runtime).</p>
3785 <!-- _______________________________________________________________________ -->
3786 <div class="doc_subsubsection">
3787 <a name="m_GlobalVariable">Important Public Members of the
3788 <tt>GlobalVariable</tt> class</a>
3791 <div class="doc_text">
3794 <li><tt>GlobalVariable(const </tt><tt><a href="#Type">Type</a> *Ty, bool
3795 isConstant, LinkageTypes& Linkage, <a href="#Constant">Constant</a>
3796 *Initializer = 0, const std::string &Name = "", Module* Parent = 0)</tt>
3798 <p>Create a new global variable of the specified type. If
3799 <tt>isConstant</tt> is true then the global variable will be marked as
3800 unchanging for the program. The Linkage parameter specifies the type of
3801 linkage (internal, external, weak, linkonce, appending) for the variable.
3802 If the linkage is InternalLinkage, WeakAnyLinkage, WeakODRLinkage,
3803 LinkOnceAnyLinkage or LinkOnceODRLinkage, then the resultant
3804 global variable will have internal linkage. AppendingLinkage concatenates
3805 together all instances (in different translation units) of the variable
3806 into a single variable but is only applicable to arrays. See
3807 the <a href="LangRef.html#modulestructure">LLVM Language Reference</a> for
3808 further details on linkage types. Optionally an initializer, a name, and the
3809 module to put the variable into may be specified for the global variable as
3812 <li><tt>bool isConstant() const</tt>
3814 <p>Returns true if this is a global variable that is known not to
3815 be modified at runtime.</p></li>
3817 <li><tt>bool hasInitializer()</tt>
3819 <p>Returns true if this <tt>GlobalVariable</tt> has an intializer.</p></li>
3821 <li><tt><a href="#Constant">Constant</a> *getInitializer()</tt>
3823 <p>Returns the initial value for a <tt>GlobalVariable</tt>. It is not legal
3824 to call this method if there is no initializer.</p></li>
3830 <!-- ======================================================================= -->
3831 <div class="doc_subsection">
3832 <a name="BasicBlock">The <tt>BasicBlock</tt> class</a>
3835 <div class="doc_text">
3838 href="/doxygen/BasicBlock_8h-source.html">llvm/BasicBlock.h</a>"</tt><br>
3839 doxygen info: <a href="/doxygen/classllvm_1_1BasicBlock.html">BasicBlock
3841 Superclass: <a href="#Value"><tt>Value</tt></a></p>
3843 <p>This class represents a single entry multiple exit section of the code,
3844 commonly known as a basic block by the compiler community. The
3845 <tt>BasicBlock</tt> class maintains a list of <a
3846 href="#Instruction"><tt>Instruction</tt></a>s, which form the body of the block.
3847 Matching the language definition, the last element of this list of instructions
3848 is always a terminator instruction (a subclass of the <a
3849 href="#TerminatorInst"><tt>TerminatorInst</tt></a> class).</p>
3851 <p>In addition to tracking the list of instructions that make up the block, the
3852 <tt>BasicBlock</tt> class also keeps track of the <a
3853 href="#Function"><tt>Function</tt></a> that it is embedded into.</p>
3855 <p>Note that <tt>BasicBlock</tt>s themselves are <a
3856 href="#Value"><tt>Value</tt></a>s, because they are referenced by instructions
3857 like branches and can go in the switch tables. <tt>BasicBlock</tt>s have type
3862 <!-- _______________________________________________________________________ -->
3863 <div class="doc_subsubsection">
3864 <a name="m_BasicBlock">Important Public Members of the <tt>BasicBlock</tt>
3868 <div class="doc_text">
3871 <li><tt>BasicBlock(const std::string &Name = "", </tt><tt><a
3872 href="#Function">Function</a> *Parent = 0)</tt>
3874 <p>The <tt>BasicBlock</tt> constructor is used to create new basic blocks for
3875 insertion into a function. The constructor optionally takes a name for the new
3876 block, and a <a href="#Function"><tt>Function</tt></a> to insert it into. If
3877 the <tt>Parent</tt> parameter is specified, the new <tt>BasicBlock</tt> is
3878 automatically inserted at the end of the specified <a
3879 href="#Function"><tt>Function</tt></a>, if not specified, the BasicBlock must be
3880 manually inserted into the <a href="#Function"><tt>Function</tt></a>.</p></li>
3882 <li><tt>BasicBlock::iterator</tt> - Typedef for instruction list iterator<br>
3883 <tt>BasicBlock::const_iterator</tt> - Typedef for const_iterator.<br>
3884 <tt>begin()</tt>, <tt>end()</tt>, <tt>front()</tt>, <tt>back()</tt>,
3885 <tt>size()</tt>, <tt>empty()</tt>
3886 STL-style functions for accessing the instruction list.
3888 <p>These methods and typedefs are forwarding functions that have the same
3889 semantics as the standard library methods of the same names. These methods
3890 expose the underlying instruction list of a basic block in a way that is easy to
3891 manipulate. To get the full complement of container operations (including
3892 operations to update the list), you must use the <tt>getInstList()</tt>
3895 <li><tt>BasicBlock::InstListType &getInstList()</tt>
3897 <p>This method is used to get access to the underlying container that actually
3898 holds the Instructions. This method must be used when there isn't a forwarding
3899 function in the <tt>BasicBlock</tt> class for the operation that you would like
3900 to perform. Because there are no forwarding functions for "updating"
3901 operations, you need to use this if you want to update the contents of a
3902 <tt>BasicBlock</tt>.</p></li>
3904 <li><tt><a href="#Function">Function</a> *getParent()</tt>
3906 <p> Returns a pointer to <a href="#Function"><tt>Function</tt></a> the block is
3907 embedded into, or a null pointer if it is homeless.</p></li>
3909 <li><tt><a href="#TerminatorInst">TerminatorInst</a> *getTerminator()</tt>
3911 <p> Returns a pointer to the terminator instruction that appears at the end of
3912 the <tt>BasicBlock</tt>. If there is no terminator instruction, or if the last
3913 instruction in the block is not a terminator, then a null pointer is
3921 <!-- ======================================================================= -->
3922 <div class="doc_subsection">
3923 <a name="Argument">The <tt>Argument</tt> class</a>
3926 <div class="doc_text">
3928 <p>This subclass of Value defines the interface for incoming formal
3929 arguments to a function. A Function maintains a list of its formal
3930 arguments. An argument has a pointer to the parent Function.</p>
3934 <!-- *********************************************************************** -->
3937 <a href="http://jigsaw.w3.org/css-validator/check/referer"><img
3938 src="http://jigsaw.w3.org/css-validator/images/vcss-blue" alt="Valid CSS"></a>
3939 <a href="http://validator.w3.org/check/referer"><img
3940 src="http://www.w3.org/Icons/valid-html401" alt="Valid HTML 4.01 Strict"></a>
3942 <a href="mailto:dhurjati@cs.uiuc.edu">Dinakar Dhurjati</a> and
3943 <a href="mailto:sabre@nondot.org">Chris Lattner</a><br>
3944 <a href="http://llvm.org">The LLVM Compiler Infrastructure</a><br>
3945 Last modified: $Date$