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">
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_arrayref">llvm/ADT/ArrayRef.h</a></li>
60 <li><a href="#dss_fixedarrays">Fixed Size Arrays</a></li>
61 <li><a href="#dss_heaparrays">Heap Allocated Arrays</a></li>
62 <li><a href="#dss_tinyptrvector">"llvm/ADT/TinyPtrVector.h"</a></li>
63 <li><a href="#dss_smallvector">"llvm/ADT/SmallVector.h"</a></li>
64 <li><a href="#dss_vector"><vector></a></li>
65 <li><a href="#dss_deque"><deque></a></li>
66 <li><a href="#dss_list"><list></a></li>
67 <li><a href="#dss_ilist">llvm/ADT/ilist.h</a></li>
68 <li><a href="#dss_packedvector">llvm/ADT/PackedVector.h</a></li>
69 <li><a href="#dss_other">Other Sequential Container Options</a></li>
71 <li><a href="#ds_string">String-like containers</a>
75 <li><a href="#ds_set">Set-Like Containers (std::set, SmallSet, SetVector, etc)</a>
77 <li><a href="#dss_sortedvectorset">A sorted 'vector'</a></li>
78 <li><a href="#dss_smallset">"llvm/ADT/SmallSet.h"</a></li>
79 <li><a href="#dss_smallptrset">"llvm/ADT/SmallPtrSet.h"</a></li>
80 <li><a href="#dss_denseset">"llvm/ADT/DenseSet.h"</a></li>
81 <li><a href="#dss_FoldingSet">"llvm/ADT/FoldingSet.h"</a></li>
82 <li><a href="#dss_set"><set></a></li>
83 <li><a href="#dss_setvector">"llvm/ADT/SetVector.h"</a></li>
84 <li><a href="#dss_uniquevector">"llvm/ADT/UniqueVector.h"</a></li>
85 <li><a href="#dss_otherset">Other Set-Like ContainerOptions</a></li>
87 <li><a href="#ds_map">Map-Like Containers (std::map, DenseMap, etc)</a>
89 <li><a href="#dss_sortedvectormap">A sorted 'vector'</a></li>
90 <li><a href="#dss_stringmap">"llvm/ADT/StringMap.h"</a></li>
91 <li><a href="#dss_indexedmap">"llvm/ADT/IndexedMap.h"</a></li>
92 <li><a href="#dss_densemap">"llvm/ADT/DenseMap.h"</a></li>
93 <li><a href="#dss_valuemap">"llvm/ADT/ValueMap.h"</a></li>
94 <li><a href="#dss_intervalmap">"llvm/ADT/IntervalMap.h"</a></li>
95 <li><a href="#dss_map"><map></a></li>
96 <li><a href="#dss_inteqclasses">"llvm/ADT/IntEqClasses.h"</a></li>
97 <li><a href="#dss_othermap">Other Map-Like Container Options</a></li>
99 <li><a href="#ds_bit">BitVector-like containers</a>
101 <li><a href="#dss_bitvector">A dense bitvector</a></li>
102 <li><a href="#dss_smallbitvector">A "small" dense bitvector</a></li>
103 <li><a href="#dss_sparsebitvector">A sparse bitvector</a></li>
107 <li><a href="#common">Helpful Hints for Common Operations</a>
109 <li><a href="#inspection">Basic Inspection and Traversal Routines</a>
111 <li><a href="#iterate_function">Iterating over the <tt>BasicBlock</tt>s
112 in a <tt>Function</tt></a> </li>
113 <li><a href="#iterate_basicblock">Iterating over the <tt>Instruction</tt>s
114 in a <tt>BasicBlock</tt></a> </li>
115 <li><a href="#iterate_institer">Iterating over the <tt>Instruction</tt>s
116 in a <tt>Function</tt></a> </li>
117 <li><a href="#iterate_convert">Turning an iterator into a
118 class pointer</a> </li>
119 <li><a href="#iterate_complex">Finding call sites: a more
120 complex example</a> </li>
121 <li><a href="#calls_and_invokes">Treating calls and invokes
122 the same way</a> </li>
123 <li><a href="#iterate_chains">Iterating over def-use &
124 use-def chains</a> </li>
125 <li><a href="#iterate_preds">Iterating over predecessors &
126 successors of blocks</a></li>
129 <li><a href="#simplechanges">Making simple changes</a>
131 <li><a href="#schanges_creating">Creating and inserting new
132 <tt>Instruction</tt>s</a> </li>
133 <li><a href="#schanges_deleting">Deleting <tt>Instruction</tt>s</a> </li>
134 <li><a href="#schanges_replacing">Replacing an <tt>Instruction</tt>
135 with another <tt>Value</tt></a> </li>
136 <li><a href="#schanges_deletingGV">Deleting <tt>GlobalVariable</tt>s</a> </li>
139 <li><a href="#create_types">How to Create Types</a></li>
141 <li>Working with the Control Flow Graph
143 <li>Accessing predecessors and successors of a <tt>BasicBlock</tt>
151 <li><a href="#threading">Threads and LLVM</a>
153 <li><a href="#startmultithreaded">Entering and Exiting Multithreaded Mode
155 <li><a href="#shutdown">Ending execution with <tt>llvm_shutdown()</tt></a></li>
156 <li><a href="#managedstatic">Lazy initialization with <tt>ManagedStatic</tt></a></li>
157 <li><a href="#llvmcontext">Achieving Isolation with <tt>LLVMContext</tt></a></li>
158 <li><a href="#jitthreading">Threads and the JIT</a></li>
162 <li><a href="#advanced">Advanced Topics</a>
165 <li><a href="#SymbolTable">The <tt>ValueSymbolTable</tt> class</a></li>
166 <li><a href="#UserLayout">The <tt>User</tt> and owned <tt>Use</tt> classes' memory layout</a></li>
169 <li><a href="#coreclasses">The Core LLVM Class Hierarchy Reference</a>
171 <li><a href="#Type">The <tt>Type</tt> class</a> </li>
172 <li><a href="#Module">The <tt>Module</tt> class</a></li>
173 <li><a href="#Value">The <tt>Value</tt> class</a>
175 <li><a href="#User">The <tt>User</tt> class</a>
177 <li><a href="#Instruction">The <tt>Instruction</tt> class</a></li>
178 <li><a href="#Constant">The <tt>Constant</tt> class</a>
180 <li><a href="#GlobalValue">The <tt>GlobalValue</tt> class</a>
182 <li><a href="#Function">The <tt>Function</tt> class</a></li>
183 <li><a href="#GlobalVariable">The <tt>GlobalVariable</tt> class</a></li>
190 <li><a href="#BasicBlock">The <tt>BasicBlock</tt> class</a></li>
191 <li><a href="#Argument">The <tt>Argument</tt> class</a></li>
198 <div class="doc_author">
199 <p>Written by <a href="mailto:sabre@nondot.org">Chris Lattner</a>,
200 <a href="mailto:dhurjati@cs.uiuc.edu">Dinakar Dhurjati</a>,
201 <a href="mailto:ggreif@gmail.com">Gabor Greif</a>,
202 <a href="mailto:jstanley@cs.uiuc.edu">Joel Stanley</a>,
203 <a href="mailto:rspencer@x10sys.com">Reid Spencer</a> and
204 <a href="mailto:owen@apple.com">Owen Anderson</a></p>
207 <!-- *********************************************************************** -->
209 <a name="introduction">Introduction </a>
211 <!-- *********************************************************************** -->
215 <p>This document is meant to highlight some of the important classes and
216 interfaces available in the LLVM source-base. This manual is not
217 intended to explain what LLVM is, how it works, and what LLVM code looks
218 like. It assumes that you know the basics of LLVM and are interested
219 in writing transformations or otherwise analyzing or manipulating the
222 <p>This document should get you oriented so that you can find your
223 way in the continuously growing source code that makes up the LLVM
224 infrastructure. Note that this manual is not intended to serve as a
225 replacement for reading the source code, so if you think there should be
226 a method in one of these classes to do something, but it's not listed,
227 check the source. Links to the <a href="/doxygen/">doxygen</a> sources
228 are provided to make this as easy as possible.</p>
230 <p>The first section of this document describes general information that is
231 useful to know when working in the LLVM infrastructure, and the second describes
232 the Core LLVM classes. In the future this manual will be extended with
233 information describing how to use extension libraries, such as dominator
234 information, CFG traversal routines, and useful utilities like the <tt><a
235 href="/doxygen/InstVisitor_8h-source.html">InstVisitor</a></tt> template.</p>
239 <!-- *********************************************************************** -->
241 <a name="general">General Information</a>
243 <!-- *********************************************************************** -->
247 <p>This section contains general information that is useful if you are working
248 in the LLVM source-base, but that isn't specific to any particular API.</p>
250 <!-- ======================================================================= -->
252 <a name="stl">The C++ Standard Template Library</a>
257 <p>LLVM makes heavy use of the C++ Standard Template Library (STL),
258 perhaps much more than you are used to, or have seen before. Because of
259 this, you might want to do a little background reading in the
260 techniques used and capabilities of the library. There are many good
261 pages that discuss the STL, and several books on the subject that you
262 can get, so it will not be discussed in this document.</p>
264 <p>Here are some useful links:</p>
268 <li><a href="http://www.dinkumware.com/manuals/#Standard C++ Library">Dinkumware
269 C++ Library reference</a> - an excellent reference for the STL and other parts
270 of the standard C++ library.</li>
272 <li><a href="http://www.tempest-sw.com/cpp/">C++ In a Nutshell</a> - This is an
273 O'Reilly book in the making. It has a decent Standard Library
274 Reference that rivals Dinkumware's, and is unfortunately no longer free since the
275 book has been published.</li>
277 <li><a href="http://www.parashift.com/c++-faq-lite/">C++ Frequently Asked
280 <li><a href="http://www.sgi.com/tech/stl/">SGI's STL Programmer's Guide</a> -
282 href="http://www.sgi.com/tech/stl/stl_introduction.html">Introduction to the
285 <li><a href="http://www.research.att.com/%7Ebs/C++.html">Bjarne Stroustrup's C++
288 <li><a href="http://64.78.49.204/">
289 Bruce Eckel's Thinking in C++, 2nd ed. Volume 2 Revision 4.0 (even better, get
294 <p>You are also encouraged to take a look at the <a
295 href="CodingStandards.html">LLVM Coding Standards</a> guide which focuses on how
296 to write maintainable code more than where to put your curly braces.</p>
300 <!-- ======================================================================= -->
302 <a name="stl">Other useful references</a>
308 <li><a href="http://www.fortran-2000.com/ArnaudRecipes/sharedlib.html">Using
309 static and shared libraries across platforms</a></li>
316 <!-- *********************************************************************** -->
318 <a name="apis">Important and useful LLVM APIs</a>
320 <!-- *********************************************************************** -->
324 <p>Here we highlight some LLVM APIs that are generally useful and good to
325 know about when writing transformations.</p>
327 <!-- ======================================================================= -->
329 <a name="isa">The <tt>isa<></tt>, <tt>cast<></tt> and
330 <tt>dyn_cast<></tt> templates</a>
335 <p>The LLVM source-base makes extensive use of a custom form of RTTI.
336 These templates have many similarities to the C++ <tt>dynamic_cast<></tt>
337 operator, but they don't have some drawbacks (primarily stemming from
338 the fact that <tt>dynamic_cast<></tt> only works on classes that
339 have a v-table). Because they are used so often, you must know what they
340 do and how they work. All of these templates are defined in the <a
341 href="/doxygen/Casting_8h-source.html"><tt>llvm/Support/Casting.h</tt></a>
342 file (note that you very rarely have to include this file directly).</p>
345 <dt><tt>isa<></tt>: </dt>
347 <dd><p>The <tt>isa<></tt> operator works exactly like the Java
348 "<tt>instanceof</tt>" operator. It returns true or false depending on whether
349 a reference or pointer points to an instance of the specified class. This can
350 be very useful for constraint checking of various sorts (example below).</p>
353 <dt><tt>cast<></tt>: </dt>
355 <dd><p>The <tt>cast<></tt> operator is a "checked cast" operation. It
356 converts a pointer or reference from a base class to a derived class, causing
357 an assertion failure if it is not really an instance of the right type. This
358 should be used in cases where you have some information that makes you believe
359 that something is of the right type. An example of the <tt>isa<></tt>
360 and <tt>cast<></tt> template is:</p>
362 <div class="doc_code">
364 static bool isLoopInvariant(const <a href="#Value">Value</a> *V, const Loop *L) {
365 if (isa<<a href="#Constant">Constant</a>>(V) || isa<<a href="#Argument">Argument</a>>(V) || isa<<a href="#GlobalValue">GlobalValue</a>>(V))
368 // <i>Otherwise, it must be an instruction...</i>
369 return !L->contains(cast<<a href="#Instruction">Instruction</a>>(V)->getParent());
374 <p>Note that you should <b>not</b> use an <tt>isa<></tt> test followed
375 by a <tt>cast<></tt>, for that use the <tt>dyn_cast<></tt>
380 <dt><tt>dyn_cast<></tt>:</dt>
382 <dd><p>The <tt>dyn_cast<></tt> operator is a "checking cast" operation.
383 It checks to see if the operand is of the specified type, and if so, returns a
384 pointer to it (this operator does not work with references). If the operand is
385 not of the correct type, a null pointer is returned. Thus, this works very
386 much like the <tt>dynamic_cast<></tt> operator in C++, and should be
387 used in the same circumstances. Typically, the <tt>dyn_cast<></tt>
388 operator is used in an <tt>if</tt> statement or some other flow control
389 statement like this:</p>
391 <div class="doc_code">
393 if (<a href="#AllocationInst">AllocationInst</a> *AI = dyn_cast<<a href="#AllocationInst">AllocationInst</a>>(Val)) {
399 <p>This form of the <tt>if</tt> statement effectively combines together a call
400 to <tt>isa<></tt> and a call to <tt>cast<></tt> into one
401 statement, which is very convenient.</p>
403 <p>Note that the <tt>dyn_cast<></tt> operator, like C++'s
404 <tt>dynamic_cast<></tt> or Java's <tt>instanceof</tt> operator, can be
405 abused. In particular, you should not use big chained <tt>if/then/else</tt>
406 blocks to check for lots of different variants of classes. If you find
407 yourself wanting to do this, it is much cleaner and more efficient to use the
408 <tt>InstVisitor</tt> class to dispatch over the instruction type directly.</p>
412 <dt><tt>cast_or_null<></tt>: </dt>
414 <dd><p>The <tt>cast_or_null<></tt> operator works just like the
415 <tt>cast<></tt> operator, except that it allows for a null pointer as an
416 argument (which it then propagates). This can sometimes be useful, allowing
417 you to combine several null checks into one.</p></dd>
419 <dt><tt>dyn_cast_or_null<></tt>: </dt>
421 <dd><p>The <tt>dyn_cast_or_null<></tt> operator works just like the
422 <tt>dyn_cast<></tt> operator, except that it allows for a null pointer
423 as an argument (which it then propagates). This can sometimes be useful,
424 allowing you to combine several null checks into one.</p></dd>
428 <p>These five templates can be used with any classes, whether they have a
429 v-table or not. To add support for these templates, you simply need to add
430 <tt>classof</tt> static methods to the class you are interested casting
431 to. Describing this is currently outside the scope of this document, but there
432 are lots of examples in the LLVM source base.</p>
437 <!-- ======================================================================= -->
439 <a name="string_apis">Passing strings (the <tt>StringRef</tt>
440 and <tt>Twine</tt> classes)</a>
445 <p>Although LLVM generally does not do much string manipulation, we do have
446 several important APIs which take strings. Two important examples are the
447 Value class -- which has names for instructions, functions, etc. -- and the
448 StringMap class which is used extensively in LLVM and Clang.</p>
450 <p>These are generic classes, and they need to be able to accept strings which
451 may have embedded null characters. Therefore, they cannot simply take
452 a <tt>const char *</tt>, and taking a <tt>const std::string&</tt> requires
453 clients to perform a heap allocation which is usually unnecessary. Instead,
454 many LLVM APIs use a <tt>StringRef</tt> or a <tt>const Twine&</tt> for
455 passing strings efficiently.</p>
457 <!-- _______________________________________________________________________ -->
459 <a name="StringRef">The <tt>StringRef</tt> class</a>
464 <p>The <tt>StringRef</tt> data type represents a reference to a constant string
465 (a character array and a length) and supports the common operations available
466 on <tt>std:string</tt>, but does not require heap allocation.</p>
468 <p>It can be implicitly constructed using a C style null-terminated string,
469 an <tt>std::string</tt>, or explicitly with a character pointer and length.
470 For example, the <tt>StringRef</tt> find function is declared as:</p>
472 <pre class="doc_code">
473 iterator find(StringRef Key);
476 <p>and clients can call it using any one of:</p>
478 <pre class="doc_code">
479 Map.find("foo"); <i>// Lookup "foo"</i>
480 Map.find(std::string("bar")); <i>// Lookup "bar"</i>
481 Map.find(StringRef("\0baz", 4)); <i>// Lookup "\0baz"</i>
484 <p>Similarly, APIs which need to return a string may return a <tt>StringRef</tt>
485 instance, which can be used directly or converted to an <tt>std::string</tt>
486 using the <tt>str</tt> member function. See
487 "<tt><a href="/doxygen/classllvm_1_1StringRef_8h-source.html">llvm/ADT/StringRef.h</a></tt>"
488 for more information.</p>
490 <p>You should rarely use the <tt>StringRef</tt> class directly, because it contains
491 pointers to external memory it is not generally safe to store an instance of the
492 class (unless you know that the external storage will not be freed). StringRef is
493 small and pervasive enough in LLVM that it should always be passed by value.</p>
497 <!-- _______________________________________________________________________ -->
499 <a name="Twine">The <tt>Twine</tt> class</a>
504 <p>The <tt>Twine</tt> class is an efficient way for APIs to accept concatenated
505 strings. For example, a common LLVM paradigm is to name one instruction based on
506 the name of another instruction with a suffix, for example:</p>
508 <div class="doc_code">
510 New = CmpInst::Create(<i>...</i>, SO->getName() + ".cmp");
514 <p>The <tt>Twine</tt> class is effectively a
515 lightweight <a href="http://en.wikipedia.org/wiki/Rope_(computer_science)">rope</a>
516 which points to temporary (stack allocated) objects. Twines can be implicitly
517 constructed as the result of the plus operator applied to strings (i.e., a C
518 strings, an <tt>std::string</tt>, or a <tt>StringRef</tt>). The twine delays the
519 actual concatenation of strings until it is actually required, at which point
520 it can be efficiently rendered directly into a character array. This avoids
521 unnecessary heap allocation involved in constructing the temporary results of
522 string concatenation. See
523 "<tt><a href="/doxygen/classllvm_1_1Twine_8h-source.html">llvm/ADT/Twine.h</a></tt>"
524 for more information.</p>
526 <p>As with a <tt>StringRef</tt>, <tt>Twine</tt> objects point to external memory
527 and should almost never be stored or mentioned directly. They are intended
528 solely for use when defining a function which should be able to efficiently
529 accept concatenated strings.</p>
535 <!-- ======================================================================= -->
537 <a name="DEBUG">The <tt>DEBUG()</tt> macro and <tt>-debug</tt> option</a>
542 <p>Often when working on your pass you will put a bunch of debugging printouts
543 and other code into your pass. After you get it working, you want to remove
544 it, but you may need it again in the future (to work out new bugs that you run
547 <p> Naturally, because of this, you don't want to delete the debug printouts,
548 but you don't want them to always be noisy. A standard compromise is to comment
549 them out, allowing you to enable them if you need them in the future.</p>
551 <p>The "<tt><a href="/doxygen/Debug_8h-source.html">llvm/Support/Debug.h</a></tt>"
552 file provides a macro named <tt>DEBUG()</tt> that is a much nicer solution to
553 this problem. Basically, you can put arbitrary code into the argument of the
554 <tt>DEBUG</tt> macro, and it is only executed if '<tt>opt</tt>' (or any other
555 tool) is run with the '<tt>-debug</tt>' command line argument:</p>
557 <div class="doc_code">
559 DEBUG(errs() << "I am here!\n");
563 <p>Then you can run your pass like this:</p>
565 <div class="doc_code">
567 $ opt < a.bc > /dev/null -mypass
568 <i><no output></i>
569 $ opt < a.bc > /dev/null -mypass -debug
574 <p>Using the <tt>DEBUG()</tt> macro instead of a home-brewed solution allows you
575 to not have to create "yet another" command line option for the debug output for
576 your pass. Note that <tt>DEBUG()</tt> macros are disabled for optimized builds,
577 so they do not cause a performance impact at all (for the same reason, they
578 should also not contain side-effects!).</p>
580 <p>One additional nice thing about the <tt>DEBUG()</tt> macro is that you can
581 enable or disable it directly in gdb. Just use "<tt>set DebugFlag=0</tt>" or
582 "<tt>set DebugFlag=1</tt>" from the gdb if the program is running. If the
583 program hasn't been started yet, you can always just run it with
586 <!-- _______________________________________________________________________ -->
588 <a name="DEBUG_TYPE">Fine grained debug info with <tt>DEBUG_TYPE</tt> and
589 the <tt>-debug-only</tt> option</a>
594 <p>Sometimes you may find yourself in a situation where enabling <tt>-debug</tt>
595 just turns on <b>too much</b> information (such as when working on the code
596 generator). If you want to enable debug information with more fine-grained
597 control, you define the <tt>DEBUG_TYPE</tt> macro and the <tt>-debug</tt> only
598 option as follows:</p>
600 <div class="doc_code">
603 DEBUG(errs() << "No debug type\n");
604 #define DEBUG_TYPE "foo"
605 DEBUG(errs() << "'foo' debug type\n");
607 #define DEBUG_TYPE "bar"
608 DEBUG(errs() << "'bar' debug type\n"));
610 #define DEBUG_TYPE ""
611 DEBUG(errs() << "No debug type (2)\n");
615 <p>Then you can run your pass like this:</p>
617 <div class="doc_code">
619 $ opt < a.bc > /dev/null -mypass
620 <i><no output></i>
621 $ opt < a.bc > /dev/null -mypass -debug
626 $ opt < a.bc > /dev/null -mypass -debug-only=foo
628 $ opt < a.bc > /dev/null -mypass -debug-only=bar
633 <p>Of course, in practice, you should only set <tt>DEBUG_TYPE</tt> at the top of
634 a file, to specify the debug type for the entire module (if you do this before
635 you <tt>#include "llvm/Support/Debug.h"</tt>, you don't have to insert the ugly
636 <tt>#undef</tt>'s). Also, you should use names more meaningful than "foo" and
637 "bar", because there is no system in place to ensure that names do not
638 conflict. If two different modules use the same string, they will all be turned
639 on when the name is specified. This allows, for example, all debug information
640 for instruction scheduling to be enabled with <tt>-debug-type=InstrSched</tt>,
641 even if the source lives in multiple files.</p>
643 <p>The <tt>DEBUG_WITH_TYPE</tt> macro is also available for situations where you
644 would like to set <tt>DEBUG_TYPE</tt>, but only for one specific <tt>DEBUG</tt>
645 statement. It takes an additional first parameter, which is the type to use. For
646 example, the preceding example could be written as:</p>
649 <div class="doc_code">
651 DEBUG_WITH_TYPE("", errs() << "No debug type\n");
652 DEBUG_WITH_TYPE("foo", errs() << "'foo' debug type\n");
653 DEBUG_WITH_TYPE("bar", errs() << "'bar' debug type\n"));
654 DEBUG_WITH_TYPE("", errs() << "No debug type (2)\n");
662 <!-- ======================================================================= -->
664 <a name="Statistic">The <tt>Statistic</tt> class & <tt>-stats</tt>
671 href="/doxygen/Statistic_8h-source.html">llvm/ADT/Statistic.h</a></tt>" file
672 provides a class named <tt>Statistic</tt> that is used as a unified way to
673 keep track of what the LLVM compiler is doing and how effective various
674 optimizations are. It is useful to see what optimizations are contributing to
675 making a particular program run faster.</p>
677 <p>Often you may run your pass on some big program, and you're interested to see
678 how many times it makes a certain transformation. Although you can do this with
679 hand inspection, or some ad-hoc method, this is a real pain and not very useful
680 for big programs. Using the <tt>Statistic</tt> class makes it very easy to
681 keep track of this information, and the calculated information is presented in a
682 uniform manner with the rest of the passes being executed.</p>
684 <p>There are many examples of <tt>Statistic</tt> uses, but the basics of using
685 it are as follows:</p>
688 <li><p>Define your statistic like this:</p>
690 <div class="doc_code">
692 #define <a href="#DEBUG_TYPE">DEBUG_TYPE</a> "mypassname" <i>// This goes before any #includes.</i>
693 STATISTIC(NumXForms, "The # of times I did stuff");
697 <p>The <tt>STATISTIC</tt> macro defines a static variable, whose name is
698 specified by the first argument. The pass name is taken from the DEBUG_TYPE
699 macro, and the description is taken from the second argument. The variable
700 defined ("NumXForms" in this case) acts like an unsigned integer.</p></li>
702 <li><p>Whenever you make a transformation, bump the counter:</p>
704 <div class="doc_code">
706 ++NumXForms; // <i>I did stuff!</i>
713 <p>That's all you have to do. To get '<tt>opt</tt>' to print out the
714 statistics gathered, use the '<tt>-stats</tt>' option:</p>
716 <div class="doc_code">
718 $ opt -stats -mypassname < program.bc > /dev/null
719 <i>... statistics output ...</i>
723 <p> When running <tt>opt</tt> on a C file from the SPEC benchmark
724 suite, it gives a report that looks like this:</p>
726 <div class="doc_code">
728 7646 bitcodewriter - Number of normal instructions
729 725 bitcodewriter - Number of oversized instructions
730 129996 bitcodewriter - Number of bitcode bytes written
731 2817 raise - Number of insts DCEd or constprop'd
732 3213 raise - Number of cast-of-self removed
733 5046 raise - Number of expression trees converted
734 75 raise - Number of other getelementptr's formed
735 138 raise - Number of load/store peepholes
736 42 deadtypeelim - Number of unused typenames removed from symtab
737 392 funcresolve - Number of varargs functions resolved
738 27 globaldce - Number of global variables removed
739 2 adce - Number of basic blocks removed
740 134 cee - Number of branches revectored
741 49 cee - Number of setcc instruction eliminated
742 532 gcse - Number of loads removed
743 2919 gcse - Number of instructions removed
744 86 indvars - Number of canonical indvars added
745 87 indvars - Number of aux indvars removed
746 25 instcombine - Number of dead inst eliminate
747 434 instcombine - Number of insts combined
748 248 licm - Number of load insts hoisted
749 1298 licm - Number of insts hoisted to a loop pre-header
750 3 licm - Number of insts hoisted to multiple loop preds (bad, no loop pre-header)
751 75 mem2reg - Number of alloca's promoted
752 1444 cfgsimplify - Number of blocks simplified
756 <p>Obviously, with so many optimizations, having a unified framework for this
757 stuff is very nice. Making your pass fit well into the framework makes it more
758 maintainable and useful.</p>
762 <!-- ======================================================================= -->
764 <a name="ViewGraph">Viewing graphs while debugging code</a>
769 <p>Several of the important data structures in LLVM are graphs: for example
770 CFGs made out of LLVM <a href="#BasicBlock">BasicBlock</a>s, CFGs made out of
771 LLVM <a href="CodeGenerator.html#machinebasicblock">MachineBasicBlock</a>s, and
772 <a href="CodeGenerator.html#selectiondag_intro">Instruction Selection
773 DAGs</a>. In many cases, while debugging various parts of the compiler, it is
774 nice to instantly visualize these graphs.</p>
776 <p>LLVM provides several callbacks that are available in a debug build to do
777 exactly that. If you call the <tt>Function::viewCFG()</tt> method, for example,
778 the current LLVM tool will pop up a window containing the CFG for the function
779 where each basic block is a node in the graph, and each node contains the
780 instructions in the block. Similarly, there also exists
781 <tt>Function::viewCFGOnly()</tt> (does not include the instructions), the
782 <tt>MachineFunction::viewCFG()</tt> and <tt>MachineFunction::viewCFGOnly()</tt>,
783 and the <tt>SelectionDAG::viewGraph()</tt> methods. Within GDB, for example,
784 you can usually use something like <tt>call DAG.viewGraph()</tt> to pop
785 up a window. Alternatively, you can sprinkle calls to these functions in your
786 code in places you want to debug.</p>
788 <p>Getting this to work requires a small amount of configuration. On Unix
789 systems with X11, install the <a href="http://www.graphviz.org">graphviz</a>
790 toolkit, and make sure 'dot' and 'gv' are in your path. If you are running on
791 Mac OS/X, download and install the Mac OS/X <a
792 href="http://www.pixelglow.com/graphviz/">Graphviz program</a>, and add
793 <tt>/Applications/Graphviz.app/Contents/MacOS/</tt> (or wherever you install
794 it) to your path. Once in your system and path are set up, rerun the LLVM
795 configure script and rebuild LLVM to enable this functionality.</p>
797 <p><tt>SelectionDAG</tt> has been extended to make it easier to locate
798 <i>interesting</i> nodes in large complex graphs. From gdb, if you
799 <tt>call DAG.setGraphColor(<i>node</i>, "<i>color</i>")</tt>, then the
800 next <tt>call DAG.viewGraph()</tt> would highlight the node in the
801 specified color (choices of colors can be found at <a
802 href="http://www.graphviz.org/doc/info/colors.html">colors</a>.) More
803 complex node attributes can be provided with <tt>call
804 DAG.setGraphAttrs(<i>node</i>, "<i>attributes</i>")</tt> (choices can be
805 found at <a href="http://www.graphviz.org/doc/info/attrs.html">Graph
806 Attributes</a>.) If you want to restart and clear all the current graph
807 attributes, then you can <tt>call DAG.clearGraphAttrs()</tt>. </p>
809 <p>Note that graph visualization features are compiled out of Release builds
810 to reduce file size. This means that you need a Debug+Asserts or
811 Release+Asserts build to use these features.</p>
817 <!-- *********************************************************************** -->
819 <a name="datastructure">Picking the Right Data Structure for a Task</a>
821 <!-- *********************************************************************** -->
825 <p>LLVM has a plethora of data structures in the <tt>llvm/ADT/</tt> directory,
826 and we commonly use STL data structures. This section describes the trade-offs
827 you should consider when you pick one.</p>
830 The first step is a choose your own adventure: do you want a sequential
831 container, a set-like container, or a map-like container? The most important
832 thing when choosing a container is the algorithmic properties of how you plan to
833 access the container. Based on that, you should use:</p>
836 <li>a <a href="#ds_map">map-like</a> container if you need efficient look-up
837 of an value based on another value. Map-like containers also support
838 efficient queries for containment (whether a key is in the map). Map-like
839 containers generally do not support efficient reverse mapping (values to
840 keys). If you need that, use two maps. Some map-like containers also
841 support efficient iteration through the keys in sorted order. Map-like
842 containers are the most expensive sort, only use them if you need one of
843 these capabilities.</li>
845 <li>a <a href="#ds_set">set-like</a> container if you need to put a bunch of
846 stuff into a container that automatically eliminates duplicates. Some
847 set-like containers support efficient iteration through the elements in
848 sorted order. Set-like containers are more expensive than sequential
852 <li>a <a href="#ds_sequential">sequential</a> container provides
853 the most efficient way to add elements and keeps track of the order they are
854 added to the collection. They permit duplicates and support efficient
855 iteration, but do not support efficient look-up based on a key.
858 <li>a <a href="#ds_string">string</a> container is a specialized sequential
859 container or reference structure that is used for character or byte
862 <li>a <a href="#ds_bit">bit</a> container provides an efficient way to store and
863 perform set operations on sets of numeric id's, while automatically
864 eliminating duplicates. Bit containers require a maximum of 1 bit for each
865 identifier you want to store.
870 Once the proper category of container is determined, you can fine tune the
871 memory use, constant factors, and cache behaviors of access by intelligently
872 picking a member of the category. Note that constant factors and cache behavior
873 can be a big deal. If you have a vector that usually only contains a few
874 elements (but could contain many), for example, it's much better to use
875 <a href="#dss_smallvector">SmallVector</a> than <a href="#dss_vector">vector</a>
876 . Doing so avoids (relatively) expensive malloc/free calls, which dwarf the
877 cost of adding the elements to the container. </p>
879 <!-- ======================================================================= -->
881 <a name="ds_sequential">Sequential Containers (std::vector, std::list, etc)</a>
885 There are a variety of sequential containers available for you, based on your
886 needs. Pick the first in this section that will do what you want.
888 <!-- _______________________________________________________________________ -->
890 <a name="dss_arrayref">llvm/ADT/ArrayRef.h</a>
894 <p>The llvm::ArrayRef class is the preferred class to use in an interface that
895 accepts a sequential list of elements in memory and just reads from them. By
896 taking an ArrayRef, the API can be passed a fixed size array, an std::vector,
897 an llvm::SmallVector and anything else that is contiguous in memory.
903 <!-- _______________________________________________________________________ -->
905 <a name="dss_fixedarrays">Fixed Size Arrays</a>
909 <p>Fixed size arrays are very simple and very fast. They are good if you know
910 exactly how many elements you have, or you have a (low) upper bound on how many
914 <!-- _______________________________________________________________________ -->
916 <a name="dss_heaparrays">Heap Allocated Arrays</a>
920 <p>Heap allocated arrays (new[] + delete[]) are also simple. They are good if
921 the number of elements is variable, if you know how many elements you will need
922 before the array is allocated, and if the array is usually large (if not,
923 consider a <a href="#dss_smallvector">SmallVector</a>). The cost of a heap
924 allocated array is the cost of the new/delete (aka malloc/free). Also note that
925 if you are allocating an array of a type with a constructor, the constructor and
926 destructors will be run for every element in the array (re-sizable vectors only
927 construct those elements actually used).</p>
930 <!-- _______________________________________________________________________ -->
932 <a name="dss_tinyptrvector">"llvm/ADT/TinyPtrVector.h"</a>
937 <p><tt>TinyPtrVector<Type></tt> is a highly specialized collection class
938 that is optimized to avoid allocation in the case when a vector has zero or one
939 elements. It has two major restrictions: 1) it can only hold values of pointer
940 type, and 2) it cannot hold a null pointer.</p>
942 <p>Since this container is highly specialized, it is rarely used.</p>
948 <!-- _______________________________________________________________________ -->
950 <a name="dss_smallvector">"llvm/ADT/SmallVector.h"</a>
954 <p><tt>SmallVector<Type, N></tt> is a simple class that looks and smells
955 just like <tt>vector<Type></tt>:
956 it supports efficient iteration, lays out elements in memory order (so you can
957 do pointer arithmetic between elements), supports efficient push_back/pop_back
958 operations, supports efficient random access to its elements, etc.</p>
960 <p>The advantage of SmallVector is that it allocates space for
961 some number of elements (N) <b>in the object itself</b>. Because of this, if
962 the SmallVector is dynamically smaller than N, no malloc is performed. This can
963 be a big win in cases where the malloc/free call is far more expensive than the
964 code that fiddles around with the elements.</p>
966 <p>This is good for vectors that are "usually small" (e.g. the number of
967 predecessors/successors of a block is usually less than 8). On the other hand,
968 this makes the size of the SmallVector itself large, so you don't want to
969 allocate lots of them (doing so will waste a lot of space). As such,
970 SmallVectors are most useful when on the stack.</p>
972 <p>SmallVector also provides a nice portable and efficient replacement for
977 <!-- _______________________________________________________________________ -->
979 <a name="dss_vector"><vector></a>
984 std::vector is well loved and respected. It is useful when SmallVector isn't:
985 when the size of the vector is often large (thus the small optimization will
986 rarely be a benefit) or if you will be allocating many instances of the vector
987 itself (which would waste space for elements that aren't in the container).
988 vector is also useful when interfacing with code that expects vectors :).
991 <p>One worthwhile note about std::vector: avoid code like this:</p>
993 <div class="doc_code">
996 std::vector<foo> V;
1002 <p>Instead, write this as:</p>
1004 <div class="doc_code">
1006 std::vector<foo> V;
1014 <p>Doing so will save (at least) one heap allocation and free per iteration of
1019 <!-- _______________________________________________________________________ -->
1021 <a name="dss_deque"><deque></a>
1025 <p>std::deque is, in some senses, a generalized version of std::vector. Like
1026 std::vector, it provides constant time random access and other similar
1027 properties, but it also provides efficient access to the front of the list. It
1028 does not guarantee continuity of elements within memory.</p>
1030 <p>In exchange for this extra flexibility, std::deque has significantly higher
1031 constant factor costs than std::vector. If possible, use std::vector or
1032 something cheaper.</p>
1035 <!-- _______________________________________________________________________ -->
1037 <a name="dss_list"><list></a>
1041 <p>std::list is an extremely inefficient class that is rarely useful.
1042 It performs a heap allocation for every element inserted into it, thus having an
1043 extremely high constant factor, particularly for small data types. std::list
1044 also only supports bidirectional iteration, not random access iteration.</p>
1046 <p>In exchange for this high cost, std::list supports efficient access to both
1047 ends of the list (like std::deque, but unlike std::vector or SmallVector). In
1048 addition, the iterator invalidation characteristics of std::list are stronger
1049 than that of a vector class: inserting or removing an element into the list does
1050 not invalidate iterator or pointers to other elements in the list.</p>
1053 <!-- _______________________________________________________________________ -->
1055 <a name="dss_ilist">llvm/ADT/ilist.h</a>
1059 <p><tt>ilist<T></tt> implements an 'intrusive' doubly-linked list. It is
1060 intrusive, because it requires the element to store and provide access to the
1061 prev/next pointers for the list.</p>
1063 <p><tt>ilist</tt> has the same drawbacks as <tt>std::list</tt>, and additionally
1064 requires an <tt>ilist_traits</tt> implementation for the element type, but it
1065 provides some novel characteristics. In particular, it can efficiently store
1066 polymorphic objects, the traits class is informed when an element is inserted or
1067 removed from the list, and <tt>ilist</tt>s are guaranteed to support a
1068 constant-time splice operation.</p>
1070 <p>These properties are exactly what we want for things like
1071 <tt>Instruction</tt>s and basic blocks, which is why these are implemented with
1072 <tt>ilist</tt>s.</p>
1074 Related classes of interest are explained in the following subsections:
1076 <li><a href="#dss_ilist_traits">ilist_traits</a></li>
1077 <li><a href="#dss_iplist">iplist</a></li>
1078 <li><a href="#dss_ilist_node">llvm/ADT/ilist_node.h</a></li>
1079 <li><a href="#dss_ilist_sentinel">Sentinels</a></li>
1083 <!-- _______________________________________________________________________ -->
1085 <a name="dss_packedvector">llvm/ADT/PackedVector.h</a>
1090 Useful for storing a vector of values using only a few number of bits for each
1091 value. Apart from the standard operations of a vector-like container, it can
1092 also perform an 'or' set operation.
1097 <div class="doc_code">
1101 FirstCondition = 0x1,
1102 SecondCondition = 0x2,
1107 PackedVector<State, 2> Vec1;
1108 Vec1.push_back(FirstCondition);
1110 PackedVector<State, 2> Vec2;
1111 Vec2.push_back(SecondCondition);
1114 return Vec1[0]; // returns 'Both'.
1121 <!-- _______________________________________________________________________ -->
1123 <a name="dss_ilist_traits">ilist_traits</a>
1127 <p><tt>ilist_traits<T></tt> is <tt>ilist<T></tt>'s customization
1128 mechanism. <tt>iplist<T></tt> (and consequently <tt>ilist<T></tt>)
1129 publicly derive from this traits class.</p>
1132 <!-- _______________________________________________________________________ -->
1134 <a name="dss_iplist">iplist</a>
1138 <p><tt>iplist<T></tt> is <tt>ilist<T></tt>'s base and as such
1139 supports a slightly narrower interface. Notably, inserters from
1140 <tt>T&</tt> are absent.</p>
1142 <p><tt>ilist_traits<T></tt> is a public base of this class and can be
1143 used for a wide variety of customizations.</p>
1146 <!-- _______________________________________________________________________ -->
1148 <a name="dss_ilist_node">llvm/ADT/ilist_node.h</a>
1152 <p><tt>ilist_node<T></tt> implements a the forward and backward links
1153 that are expected by the <tt>ilist<T></tt> (and analogous containers)
1154 in the default manner.</p>
1156 <p><tt>ilist_node<T></tt>s are meant to be embedded in the node type
1157 <tt>T</tt>, usually <tt>T</tt> publicly derives from
1158 <tt>ilist_node<T></tt>.</p>
1161 <!-- _______________________________________________________________________ -->
1163 <a name="dss_ilist_sentinel">Sentinels</a>
1167 <p><tt>ilist</tt>s have another specialty that must be considered. To be a good
1168 citizen in the C++ ecosystem, it needs to support the standard container
1169 operations, such as <tt>begin</tt> and <tt>end</tt> iterators, etc. Also, the
1170 <tt>operator--</tt> must work correctly on the <tt>end</tt> iterator in the
1171 case of non-empty <tt>ilist</tt>s.</p>
1173 <p>The only sensible solution to this problem is to allocate a so-called
1174 <i>sentinel</i> along with the intrusive list, which serves as the <tt>end</tt>
1175 iterator, providing the back-link to the last element. However conforming to the
1176 C++ convention it is illegal to <tt>operator++</tt> beyond the sentinel and it
1177 also must not be dereferenced.</p>
1179 <p>These constraints allow for some implementation freedom to the <tt>ilist</tt>
1180 how to allocate and store the sentinel. The corresponding policy is dictated
1181 by <tt>ilist_traits<T></tt>. By default a <tt>T</tt> gets heap-allocated
1182 whenever the need for a sentinel arises.</p>
1184 <p>While the default policy is sufficient in most cases, it may break down when
1185 <tt>T</tt> does not provide a default constructor. Also, in the case of many
1186 instances of <tt>ilist</tt>s, the memory overhead of the associated sentinels
1187 is wasted. To alleviate the situation with numerous and voluminous
1188 <tt>T</tt>-sentinels, sometimes a trick is employed, leading to <i>ghostly
1191 <p>Ghostly sentinels are obtained by specially-crafted <tt>ilist_traits<T></tt>
1192 which superpose the sentinel with the <tt>ilist</tt> instance in memory. Pointer
1193 arithmetic is used to obtain the sentinel, which is relative to the
1194 <tt>ilist</tt>'s <tt>this</tt> pointer. The <tt>ilist</tt> is augmented by an
1195 extra pointer, which serves as the back-link of the sentinel. This is the only
1196 field in the ghostly sentinel which can be legally accessed.</p>
1199 <!-- _______________________________________________________________________ -->
1201 <a name="dss_other">Other Sequential Container options</a>
1205 <p>Other STL containers are available, such as std::string.</p>
1207 <p>There are also various STL adapter classes such as std::queue,
1208 std::priority_queue, std::stack, etc. These provide simplified access to an
1209 underlying container but don't affect the cost of the container itself.</p>
1215 <!-- ======================================================================= -->
1217 <a name="ds_string">String-like containers</a>
1223 TODO: const char* vs stringref vs smallstring vs std::string. Describe twine,
1224 xref to #string_apis.
1230 <!-- ======================================================================= -->
1232 <a name="ds_set">Set-Like Containers (std::set, SmallSet, SetVector, etc)</a>
1237 <p>Set-like containers are useful when you need to canonicalize multiple values
1238 into a single representation. There are several different choices for how to do
1239 this, providing various trade-offs.</p>
1241 <!-- _______________________________________________________________________ -->
1243 <a name="dss_sortedvectorset">A sorted 'vector'</a>
1248 <p>If you intend to insert a lot of elements, then do a lot of queries, a
1249 great approach is to use a vector (or other sequential container) with
1250 std::sort+std::unique to remove duplicates. This approach works really well if
1251 your usage pattern has these two distinct phases (insert then query), and can be
1252 coupled with a good choice of <a href="#ds_sequential">sequential container</a>.
1256 This combination provides the several nice properties: the result data is
1257 contiguous in memory (good for cache locality), has few allocations, is easy to
1258 address (iterators in the final vector are just indices or pointers), and can be
1259 efficiently queried with a standard binary or radix search.</p>
1263 <!-- _______________________________________________________________________ -->
1265 <a name="dss_smallset">"llvm/ADT/SmallSet.h"</a>
1270 <p>If you have a set-like data structure that is usually small and whose elements
1271 are reasonably small, a <tt>SmallSet<Type, N></tt> is a good choice. This set
1272 has space for N elements in place (thus, if the set is dynamically smaller than
1273 N, no malloc traffic is required) and accesses them with a simple linear search.
1274 When the set grows beyond 'N' elements, it allocates a more expensive representation that
1275 guarantees efficient access (for most types, it falls back to std::set, but for
1276 pointers it uses something far better, <a
1277 href="#dss_smallptrset">SmallPtrSet</a>).</p>
1279 <p>The magic of this class is that it handles small sets extremely efficiently,
1280 but gracefully handles extremely large sets without loss of efficiency. The
1281 drawback is that the interface is quite small: it supports insertion, queries
1282 and erasing, but does not support iteration.</p>
1286 <!-- _______________________________________________________________________ -->
1288 <a name="dss_smallptrset">"llvm/ADT/SmallPtrSet.h"</a>
1293 <p>SmallPtrSet has all the advantages of <tt>SmallSet</tt> (and a <tt>SmallSet</tt> of pointers is
1294 transparently implemented with a <tt>SmallPtrSet</tt>), but also supports iterators. If
1295 more than 'N' insertions are performed, a single quadratically
1296 probed hash table is allocated and grows as needed, providing extremely
1297 efficient access (constant time insertion/deleting/queries with low constant
1298 factors) and is very stingy with malloc traffic.</p>
1300 <p>Note that, unlike <tt>std::set</tt>, the iterators of <tt>SmallPtrSet</tt> are invalidated
1301 whenever an insertion occurs. Also, the values visited by the iterators are not
1302 visited in sorted order.</p>
1306 <!-- _______________________________________________________________________ -->
1308 <a name="dss_denseset">"llvm/ADT/DenseSet.h"</a>
1314 DenseSet is a simple quadratically probed hash table. It excels at supporting
1315 small values: it uses a single allocation to hold all of the pairs that
1316 are currently inserted in the set. DenseSet is a great way to unique small
1317 values that are not simple pointers (use <a
1318 href="#dss_smallptrset">SmallPtrSet</a> for pointers). Note that DenseSet has
1319 the same requirements for the value type that <a
1320 href="#dss_densemap">DenseMap</a> has.
1325 <!-- _______________________________________________________________________ -->
1327 <a name="dss_FoldingSet">"llvm/ADT/FoldingSet.h"</a>
1333 FoldingSet is an aggregate class that is really good at uniquing
1334 expensive-to-create or polymorphic objects. It is a combination of a chained
1335 hash table with intrusive links (uniqued objects are required to inherit from
1336 FoldingSetNode) that uses <a href="#dss_smallvector">SmallVector</a> as part of
1339 <p>Consider a case where you want to implement a "getOrCreateFoo" method for
1340 a complex object (for example, a node in the code generator). The client has a
1341 description of *what* it wants to generate (it knows the opcode and all the
1342 operands), but we don't want to 'new' a node, then try inserting it into a set
1343 only to find out it already exists, at which point we would have to delete it
1344 and return the node that already exists.
1347 <p>To support this style of client, FoldingSet perform a query with a
1348 FoldingSetNodeID (which wraps SmallVector) that can be used to describe the
1349 element that we want to query for. The query either returns the element
1350 matching the ID or it returns an opaque ID that indicates where insertion should
1351 take place. Construction of the ID usually does not require heap traffic.</p>
1353 <p>Because FoldingSet uses intrusive links, it can support polymorphic objects
1354 in the set (for example, you can have SDNode instances mixed with LoadSDNodes).
1355 Because the elements are individually allocated, pointers to the elements are
1356 stable: inserting or removing elements does not invalidate any pointers to other
1362 <!-- _______________________________________________________________________ -->
1364 <a name="dss_set"><set></a>
1369 <p><tt>std::set</tt> is a reasonable all-around set class, which is decent at
1370 many things but great at nothing. std::set allocates memory for each element
1371 inserted (thus it is very malloc intensive) and typically stores three pointers
1372 per element in the set (thus adding a large amount of per-element space
1373 overhead). It offers guaranteed log(n) performance, which is not particularly
1374 fast from a complexity standpoint (particularly if the elements of the set are
1375 expensive to compare, like strings), and has extremely high constant factors for
1376 lookup, insertion and removal.</p>
1378 <p>The advantages of std::set are that its iterators are stable (deleting or
1379 inserting an element from the set does not affect iterators or pointers to other
1380 elements) and that iteration over the set is guaranteed to be in sorted order.
1381 If the elements in the set are large, then the relative overhead of the pointers
1382 and malloc traffic is not a big deal, but if the elements of the set are small,
1383 std::set is almost never a good choice.</p>
1387 <!-- _______________________________________________________________________ -->
1389 <a name="dss_setvector">"llvm/ADT/SetVector.h"</a>
1393 <p>LLVM's SetVector<Type> is an adapter class that combines your choice of
1394 a set-like container along with a <a href="#ds_sequential">Sequential
1395 Container</a>. The important property
1396 that this provides is efficient insertion with uniquing (duplicate elements are
1397 ignored) with iteration support. It implements this by inserting elements into
1398 both a set-like container and the sequential container, using the set-like
1399 container for uniquing and the sequential container for iteration.
1402 <p>The difference between SetVector and other sets is that the order of
1403 iteration is guaranteed to match the order of insertion into the SetVector.
1404 This property is really important for things like sets of pointers. Because
1405 pointer values are non-deterministic (e.g. vary across runs of the program on
1406 different machines), iterating over the pointers in the set will
1407 not be in a well-defined order.</p>
1410 The drawback of SetVector is that it requires twice as much space as a normal
1411 set and has the sum of constant factors from the set-like container and the
1412 sequential container that it uses. Use it *only* if you need to iterate over
1413 the elements in a deterministic order. SetVector is also expensive to delete
1414 elements out of (linear time), unless you use it's "pop_back" method, which is
1418 <p>SetVector is an adapter class that defaults to using std::vector and std::set
1419 for the underlying containers, so it is quite expensive. However,
1420 <tt>"llvm/ADT/SetVector.h"</tt> also provides a SmallSetVector class, which
1421 defaults to using a SmallVector and SmallSet of a specified size. If you use
1422 this, and if your sets are dynamically smaller than N, you will save a lot of
1427 <!-- _______________________________________________________________________ -->
1429 <a name="dss_uniquevector">"llvm/ADT/UniqueVector.h"</a>
1435 UniqueVector is similar to <a href="#dss_setvector">SetVector</a>, but it
1436 retains a unique ID for each element inserted into the set. It internally
1437 contains a map and a vector, and it assigns a unique ID for each value inserted
1440 <p>UniqueVector is very expensive: its cost is the sum of the cost of
1441 maintaining both the map and vector, it has high complexity, high constant
1442 factors, and produces a lot of malloc traffic. It should be avoided.</p>
1447 <!-- _______________________________________________________________________ -->
1449 <a name="dss_otherset">Other Set-Like Container Options</a>
1455 The STL provides several other options, such as std::multiset and the various
1456 "hash_set" like containers (whether from C++ TR1 or from the SGI library). We
1457 never use hash_set and unordered_set because they are generally very expensive
1458 (each insertion requires a malloc) and very non-portable.
1461 <p>std::multiset is useful if you're not interested in elimination of
1462 duplicates, but has all the drawbacks of std::set. A sorted vector (where you
1463 don't delete duplicate entries) or some other approach is almost always
1470 <!-- ======================================================================= -->
1472 <a name="ds_map">Map-Like Containers (std::map, DenseMap, etc)</a>
1476 Map-like containers are useful when you want to associate data to a key. As
1477 usual, there are a lot of different ways to do this. :)
1479 <!-- _______________________________________________________________________ -->
1481 <a name="dss_sortedvectormap">A sorted 'vector'</a>
1487 If your usage pattern follows a strict insert-then-query approach, you can
1488 trivially use the same approach as <a href="#dss_sortedvectorset">sorted vectors
1489 for set-like containers</a>. The only difference is that your query function
1490 (which uses std::lower_bound to get efficient log(n) lookup) should only compare
1491 the key, not both the key and value. This yields the same advantages as sorted
1496 <!-- _______________________________________________________________________ -->
1498 <a name="dss_stringmap">"llvm/ADT/StringMap.h"</a>
1504 Strings are commonly used as keys in maps, and they are difficult to support
1505 efficiently: they are variable length, inefficient to hash and compare when
1506 long, expensive to copy, etc. StringMap is a specialized container designed to
1507 cope with these issues. It supports mapping an arbitrary range of bytes to an
1508 arbitrary other object.</p>
1510 <p>The StringMap implementation uses a quadratically-probed hash table, where
1511 the buckets store a pointer to the heap allocated entries (and some other
1512 stuff). The entries in the map must be heap allocated because the strings are
1513 variable length. The string data (key) and the element object (value) are
1514 stored in the same allocation with the string data immediately after the element
1515 object. This container guarantees the "<tt>(char*)(&Value+1)</tt>" points
1516 to the key string for a value.</p>
1518 <p>The StringMap is very fast for several reasons: quadratic probing is very
1519 cache efficient for lookups, the hash value of strings in buckets is not
1520 recomputed when looking up an element, StringMap rarely has to touch the
1521 memory for unrelated objects when looking up a value (even when hash collisions
1522 happen), hash table growth does not recompute the hash values for strings
1523 already in the table, and each pair in the map is store in a single allocation
1524 (the string data is stored in the same allocation as the Value of a pair).</p>
1526 <p>StringMap also provides query methods that take byte ranges, so it only ever
1527 copies a string if a value is inserted into the table.</p>
1530 <!-- _______________________________________________________________________ -->
1532 <a name="dss_indexedmap">"llvm/ADT/IndexedMap.h"</a>
1537 IndexedMap is a specialized container for mapping small dense integers (or
1538 values that can be mapped to small dense integers) to some other type. It is
1539 internally implemented as a vector with a mapping function that maps the keys to
1540 the dense integer range.
1544 This is useful for cases like virtual registers in the LLVM code generator: they
1545 have a dense mapping that is offset by a compile-time constant (the first
1546 virtual register ID).</p>
1550 <!-- _______________________________________________________________________ -->
1552 <a name="dss_densemap">"llvm/ADT/DenseMap.h"</a>
1558 DenseMap is a simple quadratically probed hash table. It excels at supporting
1559 small keys and values: it uses a single allocation to hold all of the pairs that
1560 are currently inserted in the map. DenseMap is a great way to map pointers to
1561 pointers, or map other small types to each other.
1565 There are several aspects of DenseMap that you should be aware of, however. The
1566 iterators in a densemap are invalidated whenever an insertion occurs, unlike
1567 map. Also, because DenseMap allocates space for a large number of key/value
1568 pairs (it starts with 64 by default), it will waste a lot of space if your keys
1569 or values are large. Finally, you must implement a partial specialization of
1570 DenseMapInfo for the key that you want, if it isn't already supported. This
1571 is required to tell DenseMap about two special marker values (which can never be
1572 inserted into the map) that it needs internally.</p>
1576 <!-- _______________________________________________________________________ -->
1578 <a name="dss_valuemap">"llvm/ADT/ValueMap.h"</a>
1584 ValueMap is a wrapper around a <a href="#dss_densemap">DenseMap</a> mapping
1585 Value*s (or subclasses) to another type. When a Value is deleted or RAUW'ed,
1586 ValueMap will update itself so the new version of the key is mapped to the same
1587 value, just as if the key were a WeakVH. You can configure exactly how this
1588 happens, and what else happens on these two events, by passing
1589 a <code>Config</code> parameter to the ValueMap template.</p>
1593 <!-- _______________________________________________________________________ -->
1595 <a name="dss_intervalmap">"llvm/ADT/IntervalMap.h"</a>
1600 <p> IntervalMap is a compact map for small keys and values. It maps key
1601 intervals instead of single keys, and it will automatically coalesce adjacent
1602 intervals. When then map only contains a few intervals, they are stored in the
1603 map object itself to avoid allocations.</p>
1605 <p> The IntervalMap iterators are quite big, so they should not be passed around
1606 as STL iterators. The heavyweight iterators allow a smaller data structure.</p>
1610 <!-- _______________________________________________________________________ -->
1612 <a name="dss_map"><map></a>
1618 std::map has similar characteristics to <a href="#dss_set">std::set</a>: it uses
1619 a single allocation per pair inserted into the map, it offers log(n) lookup with
1620 an extremely large constant factor, imposes a space penalty of 3 pointers per
1621 pair in the map, etc.</p>
1623 <p>std::map is most useful when your keys or values are very large, if you need
1624 to iterate over the collection in sorted order, or if you need stable iterators
1625 into the map (i.e. they don't get invalidated if an insertion or deletion of
1626 another element takes place).</p>
1630 <!-- _______________________________________________________________________ -->
1632 <a name="dss_inteqclasses">"llvm/ADT/IntEqClasses.h"</a>
1637 <p>IntEqClasses provides a compact representation of equivalence classes of
1638 small integers. Initially, each integer in the range 0..n-1 has its own
1639 equivalence class. Classes can be joined by passing two class representatives to
1640 the join(a, b) method. Two integers are in the same class when findLeader()
1641 returns the same representative.</p>
1643 <p>Once all equivalence classes are formed, the map can be compressed so each
1644 integer 0..n-1 maps to an equivalence class number in the range 0..m-1, where m
1645 is the total number of equivalence classes. The map must be uncompressed before
1646 it can be edited again.</p>
1650 <!-- _______________________________________________________________________ -->
1652 <a name="dss_othermap">Other Map-Like Container Options</a>
1658 The STL provides several other options, such as std::multimap and the various
1659 "hash_map" like containers (whether from C++ TR1 or from the SGI library). We
1660 never use hash_set and unordered_set because they are generally very expensive
1661 (each insertion requires a malloc) and very non-portable.</p>
1663 <p>std::multimap is useful if you want to map a key to multiple values, but has
1664 all the drawbacks of std::map. A sorted vector or some other approach is almost
1671 <!-- ======================================================================= -->
1673 <a name="ds_bit">Bit storage containers (BitVector, SparseBitVector)</a>
1677 <p>Unlike the other containers, there are only two bit storage containers, and
1678 choosing when to use each is relatively straightforward.</p>
1680 <p>One additional option is
1681 <tt>std::vector<bool></tt>: we discourage its use for two reasons 1) the
1682 implementation in many common compilers (e.g. commonly available versions of
1683 GCC) is extremely inefficient and 2) the C++ standards committee is likely to
1684 deprecate this container and/or change it significantly somehow. In any case,
1685 please don't use it.</p>
1687 <!-- _______________________________________________________________________ -->
1689 <a name="dss_bitvector">BitVector</a>
1693 <p> The BitVector container provides a dynamic size set of bits for manipulation.
1694 It supports individual bit setting/testing, as well as set operations. The set
1695 operations take time O(size of bitvector), but operations are performed one word
1696 at a time, instead of one bit at a time. This makes the BitVector very fast for
1697 set operations compared to other containers. Use the BitVector when you expect
1698 the number of set bits to be high (IE a dense set).
1702 <!-- _______________________________________________________________________ -->
1704 <a name="dss_smallbitvector">SmallBitVector</a>
1708 <p> The SmallBitVector container provides the same interface as BitVector, but
1709 it is optimized for the case where only a small number of bits, less than
1710 25 or so, are needed. It also transparently supports larger bit counts, but
1711 slightly less efficiently than a plain BitVector, so SmallBitVector should
1712 only be used when larger counts are rare.
1716 At this time, SmallBitVector does not support set operations (and, or, xor),
1717 and its operator[] does not provide an assignable lvalue.
1721 <!-- _______________________________________________________________________ -->
1723 <a name="dss_sparsebitvector">SparseBitVector</a>
1727 <p> The SparseBitVector container is much like BitVector, with one major
1728 difference: Only the bits that are set, are stored. This makes the
1729 SparseBitVector much more space efficient than BitVector when the set is sparse,
1730 as well as making set operations O(number of set bits) instead of O(size of
1731 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
1732 (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).
1740 <!-- *********************************************************************** -->
1742 <a name="common">Helpful Hints for Common Operations</a>
1744 <!-- *********************************************************************** -->
1748 <p>This section describes how to perform some very simple transformations of
1749 LLVM code. This is meant to give examples of common idioms used, showing the
1750 practical side of LLVM transformations. <p> Because this is a "how-to" section,
1751 you should also read about the main classes that you will be working with. The
1752 <a href="#coreclasses">Core LLVM Class Hierarchy Reference</a> contains details
1753 and descriptions of the main classes that you should know about.</p>
1755 <!-- NOTE: this section should be heavy on example code -->
1756 <!-- ======================================================================= -->
1758 <a name="inspection">Basic Inspection and Traversal Routines</a>
1763 <p>The LLVM compiler infrastructure have many different data structures that may
1764 be traversed. Following the example of the C++ standard template library, the
1765 techniques used to traverse these various data structures are all basically the
1766 same. For a enumerable sequence of values, the <tt>XXXbegin()</tt> function (or
1767 method) returns an iterator to the start of the sequence, the <tt>XXXend()</tt>
1768 function returns an iterator pointing to one past the last valid element of the
1769 sequence, and there is some <tt>XXXiterator</tt> data type that is common
1770 between the two operations.</p>
1772 <p>Because the pattern for iteration is common across many different aspects of
1773 the program representation, the standard template library algorithms may be used
1774 on them, and it is easier to remember how to iterate. First we show a few common
1775 examples of the data structures that need to be traversed. Other data
1776 structures are traversed in very similar ways.</p>
1778 <!-- _______________________________________________________________________ -->
1780 <a name="iterate_function">Iterating over the </a><a
1781 href="#BasicBlock"><tt>BasicBlock</tt></a>s in a <a
1782 href="#Function"><tt>Function</tt></a>
1787 <p>It's quite common to have a <tt>Function</tt> instance that you'd like to
1788 transform in some way; in particular, you'd like to manipulate its
1789 <tt>BasicBlock</tt>s. To facilitate this, you'll need to iterate over all of
1790 the <tt>BasicBlock</tt>s that constitute the <tt>Function</tt>. The following is
1791 an example that prints the name of a <tt>BasicBlock</tt> and the number of
1792 <tt>Instruction</tt>s it contains:</p>
1794 <div class="doc_code">
1796 // <i>func is a pointer to a Function instance</i>
1797 for (Function::iterator i = func->begin(), e = func->end(); i != e; ++i)
1798 // <i>Print out the name of the basic block if it has one, and then the</i>
1799 // <i>number of instructions that it contains</i>
1800 errs() << "Basic block (name=" << i->getName() << ") has "
1801 << i->size() << " instructions.\n";
1805 <p>Note that i can be used as if it were a pointer for the purposes of
1806 invoking member functions of the <tt>Instruction</tt> class. This is
1807 because the indirection operator is overloaded for the iterator
1808 classes. In the above code, the expression <tt>i->size()</tt> is
1809 exactly equivalent to <tt>(*i).size()</tt> just like you'd expect.</p>
1813 <!-- _______________________________________________________________________ -->
1815 <a name="iterate_basicblock">Iterating over the </a><a
1816 href="#Instruction"><tt>Instruction</tt></a>s in a <a
1817 href="#BasicBlock"><tt>BasicBlock</tt></a>
1822 <p>Just like when dealing with <tt>BasicBlock</tt>s in <tt>Function</tt>s, it's
1823 easy to iterate over the individual instructions that make up
1824 <tt>BasicBlock</tt>s. Here's a code snippet that prints out each instruction in
1825 a <tt>BasicBlock</tt>:</p>
1827 <div class="doc_code">
1829 // <i>blk is a pointer to a BasicBlock instance</i>
1830 for (BasicBlock::iterator i = blk->begin(), e = blk->end(); i != e; ++i)
1831 // <i>The next statement works since operator<<(ostream&,...)</i>
1832 // <i>is overloaded for Instruction&</i>
1833 errs() << *i << "\n";
1837 <p>However, this isn't really the best way to print out the contents of a
1838 <tt>BasicBlock</tt>! Since the ostream operators are overloaded for virtually
1839 anything you'll care about, you could have just invoked the print routine on the
1840 basic block itself: <tt>errs() << *blk << "\n";</tt>.</p>
1844 <!-- _______________________________________________________________________ -->
1846 <a name="iterate_institer">Iterating over the </a><a
1847 href="#Instruction"><tt>Instruction</tt></a>s in a <a
1848 href="#Function"><tt>Function</tt></a>
1853 <p>If you're finding that you commonly iterate over a <tt>Function</tt>'s
1854 <tt>BasicBlock</tt>s and then that <tt>BasicBlock</tt>'s <tt>Instruction</tt>s,
1855 <tt>InstIterator</tt> should be used instead. You'll need to include <a
1856 href="/doxygen/InstIterator_8h-source.html"><tt>llvm/Support/InstIterator.h</tt></a>,
1857 and then instantiate <tt>InstIterator</tt>s explicitly in your code. Here's a
1858 small example that shows how to dump all instructions in a function to the standard error stream:<p>
1860 <div class="doc_code">
1862 #include "<a href="/doxygen/InstIterator_8h-source.html">llvm/Support/InstIterator.h</a>"
1864 // <i>F is a pointer to a Function instance</i>
1865 for (inst_iterator I = inst_begin(F), E = inst_end(F); I != E; ++I)
1866 errs() << *I << "\n";
1870 <p>Easy, isn't it? You can also use <tt>InstIterator</tt>s to fill a
1871 work list with its initial contents. For example, if you wanted to
1872 initialize a work list to contain all instructions in a <tt>Function</tt>
1873 F, all you would need to do is something like:</p>
1875 <div class="doc_code">
1877 std::set<Instruction*> worklist;
1878 // or better yet, SmallPtrSet<Instruction*, 64> worklist;
1880 for (inst_iterator I = inst_begin(F), E = inst_end(F); I != E; ++I)
1881 worklist.insert(&*I);
1885 <p>The STL set <tt>worklist</tt> would now contain all instructions in the
1886 <tt>Function</tt> pointed to by F.</p>
1890 <!-- _______________________________________________________________________ -->
1892 <a name="iterate_convert">Turning an iterator into a class pointer (and
1898 <p>Sometimes, it'll be useful to grab a reference (or pointer) to a class
1899 instance when all you've got at hand is an iterator. Well, extracting
1900 a reference or a pointer from an iterator is very straight-forward.
1901 Assuming that <tt>i</tt> is a <tt>BasicBlock::iterator</tt> and <tt>j</tt>
1902 is a <tt>BasicBlock::const_iterator</tt>:</p>
1904 <div class="doc_code">
1906 Instruction& inst = *i; // <i>Grab reference to instruction reference</i>
1907 Instruction* pinst = &*i; // <i>Grab pointer to instruction reference</i>
1908 const Instruction& inst = *j;
1912 <p>However, the iterators you'll be working with in the LLVM framework are
1913 special: they will automatically convert to a ptr-to-instance type whenever they
1914 need to. Instead of dereferencing the iterator and then taking the address of
1915 the result, you can simply assign the iterator to the proper pointer type and
1916 you get the dereference and address-of operation as a result of the assignment
1917 (behind the scenes, this is a result of overloading casting mechanisms). Thus
1918 the last line of the last example,</p>
1920 <div class="doc_code">
1922 Instruction *pinst = &*i;
1926 <p>is semantically equivalent to</p>
1928 <div class="doc_code">
1930 Instruction *pinst = i;
1934 <p>It's also possible to turn a class pointer into the corresponding iterator,
1935 and this is a constant time operation (very efficient). The following code
1936 snippet illustrates use of the conversion constructors provided by LLVM
1937 iterators. By using these, you can explicitly grab the iterator of something
1938 without actually obtaining it via iteration over some structure:</p>
1940 <div class="doc_code">
1942 void printNextInstruction(Instruction* inst) {
1943 BasicBlock::iterator it(inst);
1944 ++it; // <i>After this line, it refers to the instruction after *inst</i>
1945 if (it != inst->getParent()->end()) errs() << *it << "\n";
1950 <p>Unfortunately, these implicit conversions come at a cost; they prevent
1951 these iterators from conforming to standard iterator conventions, and thus
1952 from being usable with standard algorithms and containers. For example, they
1953 prevent the following code, where <tt>B</tt> is a <tt>BasicBlock</tt>,
1956 <div class="doc_code">
1958 llvm::SmallVector<llvm::Instruction *, 16>(B->begin(), B->end());
1962 <p>Because of this, these implicit conversions may be removed some day,
1963 and <tt>operator*</tt> changed to return a pointer instead of a reference.</p>
1967 <!--_______________________________________________________________________-->
1969 <a name="iterate_complex">Finding call sites: a slightly more complex
1975 <p>Say that you're writing a FunctionPass and would like to count all the
1976 locations in the entire module (that is, across every <tt>Function</tt>) where a
1977 certain function (i.e., some <tt>Function</tt>*) is already in scope. As you'll
1978 learn later, you may want to use an <tt>InstVisitor</tt> to accomplish this in a
1979 much more straight-forward manner, but this example will allow us to explore how
1980 you'd do it if you didn't have <tt>InstVisitor</tt> around. In pseudo-code, this
1981 is what we want to do:</p>
1983 <div class="doc_code">
1985 initialize callCounter to zero
1986 for each Function f in the Module
1987 for each BasicBlock b in f
1988 for each Instruction i in b
1989 if (i is a CallInst and calls the given function)
1990 increment callCounter
1994 <p>And the actual code is (remember, because we're writing a
1995 <tt>FunctionPass</tt>, our <tt>FunctionPass</tt>-derived class simply has to
1996 override the <tt>runOnFunction</tt> method):</p>
1998 <div class="doc_code">
2000 Function* targetFunc = ...;
2002 class OurFunctionPass : public FunctionPass {
2004 OurFunctionPass(): callCounter(0) { }
2006 virtual runOnFunction(Function& F) {
2007 for (Function::iterator b = F.begin(), be = F.end(); b != be; ++b) {
2008 for (BasicBlock::iterator i = b->begin(), ie = b->end(); i != ie; ++i) {
2009 if (<a href="#CallInst">CallInst</a>* callInst = <a href="#isa">dyn_cast</a><<a
2010 href="#CallInst">CallInst</a>>(&*i)) {
2011 // <i>We know we've encountered a call instruction, so we</i>
2012 // <i>need to determine if it's a call to the</i>
2013 // <i>function pointed to by m_func or not.</i>
2014 if (callInst->getCalledFunction() == targetFunc)
2022 unsigned callCounter;
2029 <!--_______________________________________________________________________-->
2031 <a name="calls_and_invokes">Treating calls and invokes the same way</a>
2036 <p>You may have noticed that the previous example was a bit oversimplified in
2037 that it did not deal with call sites generated by 'invoke' instructions. In
2038 this, and in other situations, you may find that you want to treat
2039 <tt>CallInst</tt>s and <tt>InvokeInst</tt>s the same way, even though their
2040 most-specific common base class is <tt>Instruction</tt>, which includes lots of
2041 less closely-related things. For these cases, LLVM provides a handy wrapper
2043 href="http://llvm.org/doxygen/classllvm_1_1CallSite.html"><tt>CallSite</tt></a>.
2044 It is essentially a wrapper around an <tt>Instruction</tt> pointer, with some
2045 methods that provide functionality common to <tt>CallInst</tt>s and
2046 <tt>InvokeInst</tt>s.</p>
2048 <p>This class has "value semantics": it should be passed by value, not by
2049 reference and it should not be dynamically allocated or deallocated using
2050 <tt>operator new</tt> or <tt>operator delete</tt>. It is efficiently copyable,
2051 assignable and constructable, with costs equivalents to that of a bare pointer.
2052 If you look at its definition, it has only a single pointer member.</p>
2056 <!--_______________________________________________________________________-->
2058 <a name="iterate_chains">Iterating over def-use & use-def chains</a>
2063 <p>Frequently, we might have an instance of the <a
2064 href="/doxygen/classllvm_1_1Value.html">Value Class</a> and we want to
2065 determine which <tt>User</tt>s use the <tt>Value</tt>. The list of all
2066 <tt>User</tt>s of a particular <tt>Value</tt> is called a <i>def-use</i> chain.
2067 For example, let's say we have a <tt>Function*</tt> named <tt>F</tt> to a
2068 particular function <tt>foo</tt>. Finding all of the instructions that
2069 <i>use</i> <tt>foo</tt> is as simple as iterating over the <i>def-use</i> chain
2072 <div class="doc_code">
2076 for (Value::use_iterator i = F->use_begin(), e = F->use_end(); i != e; ++i)
2077 if (Instruction *Inst = dyn_cast<Instruction>(*i)) {
2078 errs() << "F is used in instruction:\n";
2079 errs() << *Inst << "\n";
2084 <p>Note that dereferencing a <tt>Value::use_iterator</tt> is not a very cheap
2085 operation. Instead of performing <tt>*i</tt> above several times, consider
2086 doing it only once in the loop body and reusing its result.</p>
2088 <p>Alternatively, it's common to have an instance of the <a
2089 href="/doxygen/classllvm_1_1User.html">User Class</a> and need to know what
2090 <tt>Value</tt>s are used by it. The list of all <tt>Value</tt>s used by a
2091 <tt>User</tt> is known as a <i>use-def</i> chain. Instances of class
2092 <tt>Instruction</tt> are common <tt>User</tt>s, so we might want to iterate over
2093 all of the values that a particular instruction uses (that is, the operands of
2094 the particular <tt>Instruction</tt>):</p>
2096 <div class="doc_code">
2098 Instruction *pi = ...;
2100 for (User::op_iterator i = pi->op_begin(), e = pi->op_end(); i != e; ++i) {
2107 <p>Declaring objects as <tt>const</tt> is an important tool of enforcing
2108 mutation free algorithms (such as analyses, etc.). For this purpose above
2109 iterators come in constant flavors as <tt>Value::const_use_iterator</tt>
2110 and <tt>Value::const_op_iterator</tt>. They automatically arise when
2111 calling <tt>use/op_begin()</tt> on <tt>const Value*</tt>s or
2112 <tt>const User*</tt>s respectively. Upon dereferencing, they return
2113 <tt>const Use*</tt>s. Otherwise the above patterns remain unchanged.</p>
2117 <!--_______________________________________________________________________-->
2119 <a name="iterate_preds">Iterating over predecessors &
2120 successors of blocks</a>
2125 <p>Iterating over the predecessors and successors of a block is quite easy
2126 with the routines defined in <tt>"llvm/Support/CFG.h"</tt>. Just use code like
2127 this to iterate over all predecessors of BB:</p>
2129 <div class="doc_code">
2131 #include "llvm/Support/CFG.h"
2132 BasicBlock *BB = ...;
2134 for (pred_iterator PI = pred_begin(BB), E = pred_end(BB); PI != E; ++PI) {
2135 BasicBlock *Pred = *PI;
2141 <p>Similarly, to iterate over successors use
2142 succ_iterator/succ_begin/succ_end.</p>
2148 <!-- ======================================================================= -->
2150 <a name="simplechanges">Making simple changes</a>
2155 <p>There are some primitive transformation operations present in the LLVM
2156 infrastructure that are worth knowing about. When performing
2157 transformations, it's fairly common to manipulate the contents of basic
2158 blocks. This section describes some of the common methods for doing so
2159 and gives example code.</p>
2161 <!--_______________________________________________________________________-->
2163 <a name="schanges_creating">Creating and inserting new
2164 <tt>Instruction</tt>s</a>
2169 <p><i>Instantiating Instructions</i></p>
2171 <p>Creation of <tt>Instruction</tt>s is straight-forward: simply call the
2172 constructor for the kind of instruction to instantiate and provide the necessary
2173 parameters. For example, an <tt>AllocaInst</tt> only <i>requires</i> a
2174 (const-ptr-to) <tt>Type</tt>. Thus:</p>
2176 <div class="doc_code">
2178 AllocaInst* ai = new AllocaInst(Type::Int32Ty);
2182 <p>will create an <tt>AllocaInst</tt> instance that represents the allocation of
2183 one integer in the current stack frame, at run time. Each <tt>Instruction</tt>
2184 subclass is likely to have varying default parameters which change the semantics
2185 of the instruction, so refer to the <a
2186 href="/doxygen/classllvm_1_1Instruction.html">doxygen documentation for the subclass of
2187 Instruction</a> that you're interested in instantiating.</p>
2189 <p><i>Naming values</i></p>
2191 <p>It is very useful to name the values of instructions when you're able to, as
2192 this facilitates the debugging of your transformations. If you end up looking
2193 at generated LLVM machine code, you definitely want to have logical names
2194 associated with the results of instructions! By supplying a value for the
2195 <tt>Name</tt> (default) parameter of the <tt>Instruction</tt> constructor, you
2196 associate a logical name with the result of the instruction's execution at
2197 run time. For example, say that I'm writing a transformation that dynamically
2198 allocates space for an integer on the stack, and that integer is going to be
2199 used as some kind of index by some other code. To accomplish this, I place an
2200 <tt>AllocaInst</tt> at the first point in the first <tt>BasicBlock</tt> of some
2201 <tt>Function</tt>, and I'm intending to use it within the same
2202 <tt>Function</tt>. I might do:</p>
2204 <div class="doc_code">
2206 AllocaInst* pa = new AllocaInst(Type::Int32Ty, 0, "indexLoc");
2210 <p>where <tt>indexLoc</tt> is now the logical name of the instruction's
2211 execution value, which is a pointer to an integer on the run time stack.</p>
2213 <p><i>Inserting instructions</i></p>
2215 <p>There are essentially two ways to insert an <tt>Instruction</tt>
2216 into an existing sequence of instructions that form a <tt>BasicBlock</tt>:</p>
2219 <li>Insertion into an explicit instruction list
2221 <p>Given a <tt>BasicBlock* pb</tt>, an <tt>Instruction* pi</tt> within that
2222 <tt>BasicBlock</tt>, and a newly-created instruction we wish to insert
2223 before <tt>*pi</tt>, we do the following: </p>
2225 <div class="doc_code">
2227 BasicBlock *pb = ...;
2228 Instruction *pi = ...;
2229 Instruction *newInst = new Instruction(...);
2231 pb->getInstList().insert(pi, newInst); // <i>Inserts newInst before pi in pb</i>
2235 <p>Appending to the end of a <tt>BasicBlock</tt> is so common that
2236 the <tt>Instruction</tt> class and <tt>Instruction</tt>-derived
2237 classes provide constructors which take a pointer to a
2238 <tt>BasicBlock</tt> to be appended to. For example code that
2241 <div class="doc_code">
2243 BasicBlock *pb = ...;
2244 Instruction *newInst = new Instruction(...);
2246 pb->getInstList().push_back(newInst); // <i>Appends newInst to pb</i>
2252 <div class="doc_code">
2254 BasicBlock *pb = ...;
2255 Instruction *newInst = new Instruction(..., pb);
2259 <p>which is much cleaner, especially if you are creating
2260 long instruction streams.</p></li>
2262 <li>Insertion into an implicit instruction list
2264 <p><tt>Instruction</tt> instances that are already in <tt>BasicBlock</tt>s
2265 are implicitly associated with an existing instruction list: the instruction
2266 list of the enclosing basic block. Thus, we could have accomplished the same
2267 thing as the above code without being given a <tt>BasicBlock</tt> by doing:
2270 <div class="doc_code">
2272 Instruction *pi = ...;
2273 Instruction *newInst = new Instruction(...);
2275 pi->getParent()->getInstList().insert(pi, newInst);
2279 <p>In fact, this sequence of steps occurs so frequently that the
2280 <tt>Instruction</tt> class and <tt>Instruction</tt>-derived classes provide
2281 constructors which take (as a default parameter) a pointer to an
2282 <tt>Instruction</tt> which the newly-created <tt>Instruction</tt> should
2283 precede. That is, <tt>Instruction</tt> constructors are capable of
2284 inserting the newly-created instance into the <tt>BasicBlock</tt> of a
2285 provided instruction, immediately before that instruction. Using an
2286 <tt>Instruction</tt> constructor with a <tt>insertBefore</tt> (default)
2287 parameter, the above code becomes:</p>
2289 <div class="doc_code">
2291 Instruction* pi = ...;
2292 Instruction* newInst = new Instruction(..., pi);
2296 <p>which is much cleaner, especially if you're creating a lot of
2297 instructions and adding them to <tt>BasicBlock</tt>s.</p></li>
2302 <!--_______________________________________________________________________-->
2304 <a name="schanges_deleting">Deleting <tt>Instruction</tt>s</a>
2309 <p>Deleting an instruction from an existing sequence of instructions that form a
2310 <a href="#BasicBlock"><tt>BasicBlock</tt></a> is very straight-forward: just
2311 call the instruction's eraseFromParent() method. For example:</p>
2313 <div class="doc_code">
2315 <a href="#Instruction">Instruction</a> *I = .. ;
2316 I->eraseFromParent();
2320 <p>This unlinks the instruction from its containing basic block and deletes
2321 it. If you'd just like to unlink the instruction from its containing basic
2322 block but not delete it, you can use the <tt>removeFromParent()</tt> method.</p>
2326 <!--_______________________________________________________________________-->
2328 <a name="schanges_replacing">Replacing an <tt>Instruction</tt> with another
2334 <p><i>Replacing individual instructions</i></p>
2336 <p>Including "<a href="/doxygen/BasicBlockUtils_8h-source.html">llvm/Transforms/Utils/BasicBlockUtils.h</a>"
2337 permits use of two very useful replace functions: <tt>ReplaceInstWithValue</tt>
2338 and <tt>ReplaceInstWithInst</tt>.</p>
2340 <h5><a name="schanges_deleting">Deleting <tt>Instruction</tt>s</a></h5>
2343 <li><tt>ReplaceInstWithValue</tt>
2345 <p>This function replaces all uses of a given instruction with a value,
2346 and then removes the original instruction. The following example
2347 illustrates the replacement of the result of a particular
2348 <tt>AllocaInst</tt> that allocates memory for a single integer with a null
2349 pointer to an integer.</p>
2351 <div class="doc_code">
2353 AllocaInst* instToReplace = ...;
2354 BasicBlock::iterator ii(instToReplace);
2356 ReplaceInstWithValue(instToReplace->getParent()->getInstList(), ii,
2357 Constant::getNullValue(PointerType::getUnqual(Type::Int32Ty)));
2360 <li><tt>ReplaceInstWithInst</tt>
2362 <p>This function replaces a particular instruction with another
2363 instruction, inserting the new instruction into the basic block at the
2364 location where the old instruction was, and replacing any uses of the old
2365 instruction with the new instruction. The following example illustrates
2366 the replacement of one <tt>AllocaInst</tt> with another.</p>
2368 <div class="doc_code">
2370 AllocaInst* instToReplace = ...;
2371 BasicBlock::iterator ii(instToReplace);
2373 ReplaceInstWithInst(instToReplace->getParent()->getInstList(), ii,
2374 new AllocaInst(Type::Int32Ty, 0, "ptrToReplacedInt"));
2378 <p><i>Replacing multiple uses of <tt>User</tt>s and <tt>Value</tt>s</i></p>
2380 <p>You can use <tt>Value::replaceAllUsesWith</tt> and
2381 <tt>User::replaceUsesOfWith</tt> to change more than one use at a time. See the
2382 doxygen documentation for the <a href="/doxygen/classllvm_1_1Value.html">Value Class</a>
2383 and <a href="/doxygen/classllvm_1_1User.html">User Class</a>, respectively, for more
2386 <!-- Value::replaceAllUsesWith User::replaceUsesOfWith Point out:
2387 include/llvm/Transforms/Utils/ especially BasicBlockUtils.h with:
2388 ReplaceInstWithValue, ReplaceInstWithInst -->
2392 <!--_______________________________________________________________________-->
2394 <a name="schanges_deletingGV">Deleting <tt>GlobalVariable</tt>s</a>
2399 <p>Deleting a global variable from a module is just as easy as deleting an
2400 Instruction. First, you must have a pointer to the global variable that you wish
2401 to delete. You use this pointer to erase it from its parent, the module.
2404 <div class="doc_code">
2406 <a href="#GlobalVariable">GlobalVariable</a> *GV = .. ;
2408 GV->eraseFromParent();
2416 <!-- ======================================================================= -->
2418 <a name="create_types">How to Create Types</a>
2423 <p>In generating IR, you may need some complex types. If you know these types
2424 statically, you can use <tt>TypeBuilder<...>::get()</tt>, defined
2425 in <tt>llvm/Support/TypeBuilder.h</tt>, to retrieve them. <tt>TypeBuilder</tt>
2426 has two forms depending on whether you're building types for cross-compilation
2427 or native library use. <tt>TypeBuilder<T, true></tt> requires
2428 that <tt>T</tt> be independent of the host environment, meaning that it's built
2430 the <a href="/doxygen/namespacellvm_1_1types.html"><tt>llvm::types</tt></a>
2431 namespace and pointers, functions, arrays, etc. built of
2432 those. <tt>TypeBuilder<T, false></tt> additionally allows native C types
2433 whose size may depend on the host compiler. For example,</p>
2435 <div class="doc_code">
2437 FunctionType *ft = TypeBuilder<types::i<8>(types::i<32>*), true>::get();
2441 <p>is easier to read and write than the equivalent</p>
2443 <div class="doc_code">
2445 std::vector<const Type*> params;
2446 params.push_back(PointerType::getUnqual(Type::Int32Ty));
2447 FunctionType *ft = FunctionType::get(Type::Int8Ty, params, false);
2451 <p>See the <a href="/doxygen/TypeBuilder_8h-source.html#l00001">class
2452 comment</a> for more details.</p>
2458 <!-- *********************************************************************** -->
2460 <a name="threading">Threads and LLVM</a>
2462 <!-- *********************************************************************** -->
2466 This section describes the interaction of the LLVM APIs with multithreading,
2467 both on the part of client applications, and in the JIT, in the hosted
2472 Note that LLVM's support for multithreading is still relatively young. Up
2473 through version 2.5, the execution of threaded hosted applications was
2474 supported, but not threaded client access to the APIs. While this use case is
2475 now supported, clients <em>must</em> adhere to the guidelines specified below to
2476 ensure proper operation in multithreaded mode.
2480 Note that, on Unix-like platforms, LLVM requires the presence of GCC's atomic
2481 intrinsics in order to support threaded operation. If you need a
2482 multhreading-capable LLVM on a platform without a suitably modern system
2483 compiler, consider compiling LLVM and LLVM-GCC in single-threaded mode, and
2484 using the resultant compiler to build a copy of LLVM with multithreading
2488 <!-- ======================================================================= -->
2490 <a name="startmultithreaded">Entering and Exiting Multithreaded Mode</a>
2496 In order to properly protect its internal data structures while avoiding
2497 excessive locking overhead in the single-threaded case, the LLVM must intialize
2498 certain data structures necessary to provide guards around its internals. To do
2499 so, the client program must invoke <tt>llvm_start_multithreaded()</tt> before
2500 making any concurrent LLVM API calls. To subsequently tear down these
2501 structures, use the <tt>llvm_stop_multithreaded()</tt> call. You can also use
2502 the <tt>llvm_is_multithreaded()</tt> call to check the status of multithreaded
2507 Note that both of these calls must be made <em>in isolation</em>. That is to
2508 say that no other LLVM API calls may be executing at any time during the
2509 execution of <tt>llvm_start_multithreaded()</tt> or <tt>llvm_stop_multithreaded
2510 </tt>. It's is the client's responsibility to enforce this isolation.
2514 The return value of <tt>llvm_start_multithreaded()</tt> indicates the success or
2515 failure of the initialization. Failure typically indicates that your copy of
2516 LLVM was built without multithreading support, typically because GCC atomic
2517 intrinsics were not found in your system compiler. In this case, the LLVM API
2518 will not be safe for concurrent calls. However, it <em>will</em> be safe for
2519 hosting threaded applications in the JIT, though <a href="#jitthreading">care
2520 must be taken</a> to ensure that side exits and the like do not accidentally
2521 result in concurrent LLVM API calls.
2525 <!-- ======================================================================= -->
2527 <a name="shutdown">Ending Execution with <tt>llvm_shutdown()</tt></a>
2532 When you are done using the LLVM APIs, you should call <tt>llvm_shutdown()</tt>
2533 to deallocate memory used for internal structures. This will also invoke
2534 <tt>llvm_stop_multithreaded()</tt> if LLVM is operating in multithreaded mode.
2535 As such, <tt>llvm_shutdown()</tt> requires the same isolation guarantees as
2536 <tt>llvm_stop_multithreaded()</tt>.
2540 Note that, if you use scope-based shutdown, you can use the
2541 <tt>llvm_shutdown_obj</tt> class, which calls <tt>llvm_shutdown()</tt> in its
2545 <!-- ======================================================================= -->
2547 <a name="managedstatic">Lazy Initialization with <tt>ManagedStatic</tt></a>
2552 <tt>ManagedStatic</tt> is a utility class in LLVM used to implement static
2553 initialization of static resources, such as the global type tables. Before the
2554 invocation of <tt>llvm_shutdown()</tt>, it implements a simple lazy
2555 initialization scheme. Once <tt>llvm_start_multithreaded()</tt> returns,
2556 however, it uses double-checked locking to implement thread-safe lazy
2561 Note that, because no other threads are allowed to issue LLVM API calls before
2562 <tt>llvm_start_multithreaded()</tt> returns, it is possible to have
2563 <tt>ManagedStatic</tt>s of <tt>llvm::sys::Mutex</tt>s.
2567 The <tt>llvm_acquire_global_lock()</tt> and <tt>llvm_release_global_lock</tt>
2568 APIs provide access to the global lock used to implement the double-checked
2569 locking for lazy initialization. These should only be used internally to LLVM,
2570 and only if you know what you're doing!
2574 <!-- ======================================================================= -->
2576 <a name="llvmcontext">Achieving Isolation with <tt>LLVMContext</tt></a>
2581 <tt>LLVMContext</tt> is an opaque class in the LLVM API which clients can use
2582 to operate multiple, isolated instances of LLVM concurrently within the same
2583 address space. For instance, in a hypothetical compile-server, the compilation
2584 of an individual translation unit is conceptually independent from all the
2585 others, and it would be desirable to be able to compile incoming translation
2586 units concurrently on independent server threads. Fortunately,
2587 <tt>LLVMContext</tt> exists to enable just this kind of scenario!
2591 Conceptually, <tt>LLVMContext</tt> provides isolation. Every LLVM entity
2592 (<tt>Module</tt>s, <tt>Value</tt>s, <tt>Type</tt>s, <tt>Constant</tt>s, etc.)
2593 in LLVM's in-memory IR belongs to an <tt>LLVMContext</tt>. Entities in
2594 different contexts <em>cannot</em> interact with each other: <tt>Module</tt>s in
2595 different contexts cannot be linked together, <tt>Function</tt>s cannot be added
2596 to <tt>Module</tt>s in different contexts, etc. What this means is that is is
2597 safe to compile on multiple threads simultaneously, as long as no two threads
2598 operate on entities within the same context.
2602 In practice, very few places in the API require the explicit specification of a
2603 <tt>LLVMContext</tt>, other than the <tt>Type</tt> creation/lookup APIs.
2604 Because every <tt>Type</tt> carries a reference to its owning context, most
2605 other entities can determine what context they belong to by looking at their
2606 own <tt>Type</tt>. If you are adding new entities to LLVM IR, please try to
2607 maintain this interface design.
2611 For clients that do <em>not</em> require the benefits of isolation, LLVM
2612 provides a convenience API <tt>getGlobalContext()</tt>. This returns a global,
2613 lazily initialized <tt>LLVMContext</tt> that may be used in situations where
2614 isolation is not a concern.
2618 <!-- ======================================================================= -->
2620 <a name="jitthreading">Threads and the JIT</a>
2625 LLVM's "eager" JIT compiler is safe to use in threaded programs. Multiple
2626 threads can call <tt>ExecutionEngine::getPointerToFunction()</tt> or
2627 <tt>ExecutionEngine::runFunction()</tt> concurrently, and multiple threads can
2628 run code output by the JIT concurrently. The user must still ensure that only
2629 one thread accesses IR in a given <tt>LLVMContext</tt> while another thread
2630 might be modifying it. One way to do that is to always hold the JIT lock while
2631 accessing IR outside the JIT (the JIT <em>modifies</em> the IR by adding
2632 <tt>CallbackVH</tt>s). Another way is to only
2633 call <tt>getPointerToFunction()</tt> from the <tt>LLVMContext</tt>'s thread.
2636 <p>When the JIT is configured to compile lazily (using
2637 <tt>ExecutionEngine::DisableLazyCompilation(false)</tt>), there is currently a
2638 <a href="http://llvm.org/bugs/show_bug.cgi?id=5184">race condition</a> in
2639 updating call sites after a function is lazily-jitted. It's still possible to
2640 use the lazy JIT in a threaded program if you ensure that only one thread at a
2641 time can call any particular lazy stub and that the JIT lock guards any IR
2642 access, but we suggest using only the eager JIT in threaded programs.
2648 <!-- *********************************************************************** -->
2650 <a name="advanced">Advanced Topics</a>
2652 <!-- *********************************************************************** -->
2656 This section describes some of the advanced or obscure API's that most clients
2657 do not need to be aware of. These API's tend manage the inner workings of the
2658 LLVM system, and only need to be accessed in unusual circumstances.
2662 <!-- ======================================================================= -->
2664 <a name="SymbolTable">The <tt>ValueSymbolTable</tt> class</a>
2668 <p>The <tt><a href="http://llvm.org/doxygen/classllvm_1_1ValueSymbolTable.html">
2669 ValueSymbolTable</a></tt> class provides a symbol table that the <a
2670 href="#Function"><tt>Function</tt></a> and <a href="#Module">
2671 <tt>Module</tt></a> classes use for naming value definitions. The symbol table
2672 can provide a name for any <a href="#Value"><tt>Value</tt></a>.
2675 <p>Note that the <tt>SymbolTable</tt> class should not be directly accessed
2676 by most clients. It should only be used when iteration over the symbol table
2677 names themselves are required, which is very special purpose. Note that not
2679 <tt><a href="#Value">Value</a></tt>s have names, and those without names (i.e. they have
2680 an empty name) do not exist in the symbol table.
2683 <p>Symbol tables support iteration over the values in the symbol
2684 table with <tt>begin/end/iterator</tt> and supports querying to see if a
2685 specific name is in the symbol table (with <tt>lookup</tt>). The
2686 <tt>ValueSymbolTable</tt> class exposes no public mutator methods, instead,
2687 simply call <tt>setName</tt> on a value, which will autoinsert it into the
2688 appropriate symbol table.</p>
2694 <!-- ======================================================================= -->
2696 <a name="UserLayout">The <tt>User</tt> and owned <tt>Use</tt> classes' memory layout</a>
2700 <p>The <tt><a href="http://llvm.org/doxygen/classllvm_1_1User.html">
2701 User</a></tt> class provides a basis for expressing the ownership of <tt>User</tt>
2702 towards other <tt><a href="http://llvm.org/doxygen/classllvm_1_1Value.html">
2703 Value</a></tt>s. The <tt><a href="http://llvm.org/doxygen/classllvm_1_1Use.html">
2704 Use</a></tt> helper class is employed to do the bookkeeping and to facilitate <i>O(1)</i>
2705 addition and removal.</p>
2707 <!-- ______________________________________________________________________ -->
2710 Interaction and relationship between <tt>User</tt> and <tt>Use</tt> objects
2716 A subclass of <tt>User</tt> can choose between incorporating its <tt>Use</tt> objects
2717 or refer to them out-of-line by means of a pointer. A mixed variant
2718 (some <tt>Use</tt>s inline others hung off) is impractical and breaks the invariant
2719 that the <tt>Use</tt> objects belonging to the same <tt>User</tt> form a contiguous array.
2723 We have 2 different layouts in the <tt>User</tt> (sub)classes:
2726 The <tt>Use</tt> object(s) are inside (resp. at fixed offset) of the <tt>User</tt>
2727 object and there are a fixed number of them.</p>
2730 The <tt>Use</tt> object(s) are referenced by a pointer to an
2731 array from the <tt>User</tt> object and there may be a variable
2735 As of v2.4 each layout still possesses a direct pointer to the
2736 start of the array of <tt>Use</tt>s. Though not mandatory for layout a),
2737 we stick to this redundancy for the sake of simplicity.
2738 The <tt>User</tt> object also stores the number of <tt>Use</tt> objects it
2739 has. (Theoretically this information can also be calculated
2740 given the scheme presented below.)</p>
2742 Special forms of allocation operators (<tt>operator new</tt>)
2743 enforce the following memory layouts:</p>
2746 <li><p>Layout a) is modelled by prepending the <tt>User</tt> object by the <tt>Use[]</tt> array.</p>
2749 ...---.---.---.---.-------...
2750 | P | P | P | P | User
2751 '''---'---'---'---'-------'''
2754 <li><p>Layout b) is modelled by pointing at the <tt>Use[]</tt> array.</p>
2766 <i>(In the above figures '<tt>P</tt>' stands for the <tt>Use**</tt> that
2767 is stored in each <tt>Use</tt> object in the member <tt>Use::Prev</tt>)</i>
2771 <!-- ______________________________________________________________________ -->
2773 <a name="Waymarking">The waymarking algorithm</a>
2778 Since the <tt>Use</tt> objects are deprived of the direct (back)pointer to
2779 their <tt>User</tt> objects, there must be a fast and exact method to
2780 recover it. This is accomplished by the following scheme:</p>
2782 A bit-encoding in the 2 LSBits (least significant bits) of the <tt>Use::Prev</tt> allows to find the
2783 start of the <tt>User</tt> object:
2785 <li><tt>00</tt> —> binary digit 0</li>
2786 <li><tt>01</tt> —> binary digit 1</li>
2787 <li><tt>10</tt> —> stop and calculate (<tt>s</tt>)</li>
2788 <li><tt>11</tt> —> full stop (<tt>S</tt>)</li>
2791 Given a <tt>Use*</tt>, all we have to do is to walk till we get
2792 a stop and we either have a <tt>User</tt> immediately behind or
2793 we have to walk to the next stop picking up digits
2794 and calculating the offset:</p>
2796 .---.---.---.---.---.---.---.---.---.---.---.---.---.---.---.---.----------------
2797 | 1 | s | 1 | 0 | 1 | 0 | s | 1 | 1 | 0 | s | 1 | 1 | s | 1 | S | User (or User*)
2798 '---'---'---'---'---'---'---'---'---'---'---'---'---'---'---'---'----------------
2799 |+15 |+10 |+6 |+3 |+1
2802 | | |______________________>
2803 | |______________________________________>
2804 |__________________________________________________________>
2807 Only the significant number of bits need to be stored between the
2808 stops, so that the <i>worst case is 20 memory accesses</i> when there are
2809 1000 <tt>Use</tt> objects associated with a <tt>User</tt>.</p>
2813 <!-- ______________________________________________________________________ -->
2815 <a name="ReferenceImpl">Reference implementation</a>
2820 The following literate Haskell fragment demonstrates the concept:</p>
2822 <div class="doc_code">
2824 > import Test.QuickCheck
2826 > digits :: Int -> [Char] -> [Char]
2827 > digits 0 acc = '0' : acc
2828 > digits 1 acc = '1' : acc
2829 > digits n acc = digits (n `div` 2) $ digits (n `mod` 2) acc
2831 > dist :: Int -> [Char] -> [Char]
2834 > dist 1 acc = let r = dist 0 acc in 's' : digits (length r) r
2835 > dist n acc = dist (n - 1) $ dist 1 acc
2837 > takeLast n ss = reverse $ take n $ reverse ss
2839 > test = takeLast 40 $ dist 20 []
2844 Printing <test> gives: <tt>"1s100000s11010s10100s1111s1010s110s11s1S"</tt></p>
2846 The reverse algorithm computes the length of the string just by examining
2847 a certain prefix:</p>
2849 <div class="doc_code">
2851 > pref :: [Char] -> Int
2853 > pref ('s':'1':rest) = decode 2 1 rest
2854 > pref (_:rest) = 1 + pref rest
2856 > decode walk acc ('0':rest) = decode (walk + 1) (acc * 2) rest
2857 > decode walk acc ('1':rest) = decode (walk + 1) (acc * 2 + 1) rest
2858 > decode walk acc _ = walk + acc
2863 Now, as expected, printing <pref test> gives <tt>40</tt>.</p>
2865 We can <i>quickCheck</i> this with following property:</p>
2867 <div class="doc_code">
2869 > testcase = dist 2000 []
2870 > testcaseLength = length testcase
2872 > identityProp n = n > 0 && n <= testcaseLength ==> length arr == pref arr
2873 > where arr = takeLast n testcase
2878 As expected <quickCheck identityProp> gives:</p>
2881 *Main> quickCheck identityProp
2882 OK, passed 100 tests.
2885 Let's be a bit more exhaustive:</p>
2887 <div class="doc_code">
2890 > deepCheck p = check (defaultConfig { configMaxTest = 500 }) p
2895 And here is the result of <deepCheck identityProp>:</p>
2898 *Main> deepCheck identityProp
2899 OK, passed 500 tests.
2904 <!-- ______________________________________________________________________ -->
2906 <a name="Tagging">Tagging considerations</a>
2912 To maintain the invariant that the 2 LSBits of each <tt>Use**</tt> in <tt>Use</tt>
2913 never change after being set up, setters of <tt>Use::Prev</tt> must re-tag the
2914 new <tt>Use**</tt> on every modification. Accordingly getters must strip the
2917 For layout b) instead of the <tt>User</tt> we find a pointer (<tt>User*</tt> with LSBit set).
2918 Following this pointer brings us to the <tt>User</tt>. A portable trick ensures
2919 that the first bytes of <tt>User</tt> (if interpreted as a pointer) never has
2920 the LSBit set. (Portability is relying on the fact that all known compilers place the
2921 <tt>vptr</tt> in the first word of the instances.)</p>
2929 <!-- *********************************************************************** -->
2931 <a name="coreclasses">The Core LLVM Class Hierarchy Reference </a>
2933 <!-- *********************************************************************** -->
2936 <p><tt>#include "<a href="/doxygen/Type_8h-source.html">llvm/Type.h</a>"</tt>
2937 <br>doxygen info: <a href="/doxygen/classllvm_1_1Type.html">Type Class</a></p>
2939 <p>The Core LLVM classes are the primary means of representing the program
2940 being inspected or transformed. The core LLVM classes are defined in
2941 header files in the <tt>include/llvm/</tt> directory, and implemented in
2942 the <tt>lib/VMCore</tt> directory.</p>
2944 <!-- ======================================================================= -->
2946 <a name="Type">The <tt>Type</tt> class and Derived Types</a>
2951 <p><tt>Type</tt> is a superclass of all type classes. Every <tt>Value</tt> has
2952 a <tt>Type</tt>. <tt>Type</tt> cannot be instantiated directly but only
2953 through its subclasses. Certain primitive types (<tt>VoidType</tt>,
2954 <tt>LabelType</tt>, <tt>FloatType</tt> and <tt>DoubleType</tt>) have hidden
2955 subclasses. They are hidden because they offer no useful functionality beyond
2956 what the <tt>Type</tt> class offers except to distinguish themselves from
2957 other subclasses of <tt>Type</tt>.</p>
2958 <p>All other types are subclasses of <tt>DerivedType</tt>. Types can be
2959 named, but this is not a requirement. There exists exactly
2960 one instance of a given shape at any one time. This allows type equality to
2961 be performed with address equality of the Type Instance. That is, given two
2962 <tt>Type*</tt> values, the types are identical if the pointers are identical.
2965 <!-- _______________________________________________________________________ -->
2967 <a name="m_Type">Important Public Methods</a>
2973 <li><tt>bool isIntegerTy() const</tt>: Returns true for any integer type.</li>
2975 <li><tt>bool isFloatingPointTy()</tt>: Return true if this is one of the five
2976 floating point types.</li>
2978 <li><tt>bool isSized()</tt>: Return true if the type has known size. Things
2979 that don't have a size are abstract types, labels and void.</li>
2984 <!-- _______________________________________________________________________ -->
2986 <a name="derivedtypes">Important Derived Types</a>
2990 <dt><tt>IntegerType</tt></dt>
2991 <dd>Subclass of DerivedType that represents integer types of any bit width.
2992 Any bit width between <tt>IntegerType::MIN_INT_BITS</tt> (1) and
2993 <tt>IntegerType::MAX_INT_BITS</tt> (~8 million) can be represented.
2995 <li><tt>static const IntegerType* get(unsigned NumBits)</tt>: get an integer
2996 type of a specific bit width.</li>
2997 <li><tt>unsigned getBitWidth() const</tt>: Get the bit width of an integer
3001 <dt><tt>SequentialType</tt></dt>
3002 <dd>This is subclassed by ArrayType, PointerType and VectorType.
3004 <li><tt>const Type * getElementType() const</tt>: Returns the type of each
3005 of the elements in the sequential type. </li>
3008 <dt><tt>ArrayType</tt></dt>
3009 <dd>This is a subclass of SequentialType and defines the interface for array
3012 <li><tt>unsigned getNumElements() const</tt>: Returns the number of
3013 elements in the array. </li>
3016 <dt><tt>PointerType</tt></dt>
3017 <dd>Subclass of SequentialType for pointer types.</dd>
3018 <dt><tt>VectorType</tt></dt>
3019 <dd>Subclass of SequentialType for vector types. A
3020 vector type is similar to an ArrayType but is distinguished because it is
3021 a first class type whereas ArrayType is not. Vector types are used for
3022 vector operations and are usually small vectors of of an integer or floating
3024 <dt><tt>StructType</tt></dt>
3025 <dd>Subclass of DerivedTypes for struct types.</dd>
3026 <dt><tt><a name="FunctionType">FunctionType</a></tt></dt>
3027 <dd>Subclass of DerivedTypes for function types.
3029 <li><tt>bool isVarArg() const</tt>: Returns true if it's a vararg
3031 <li><tt> const Type * getReturnType() const</tt>: Returns the
3032 return type of the function.</li>
3033 <li><tt>const Type * getParamType (unsigned i)</tt>: Returns
3034 the type of the ith parameter.</li>
3035 <li><tt> const unsigned getNumParams() const</tt>: Returns the
3036 number of formal parameters.</li>
3044 <!-- ======================================================================= -->
3046 <a name="Module">The <tt>Module</tt> class</a>
3052 href="/doxygen/Module_8h-source.html">llvm/Module.h</a>"</tt><br> doxygen info:
3053 <a href="/doxygen/classllvm_1_1Module.html">Module Class</a></p>
3055 <p>The <tt>Module</tt> class represents the top level structure present in LLVM
3056 programs. An LLVM module is effectively either a translation unit of the
3057 original program or a combination of several translation units merged by the
3058 linker. The <tt>Module</tt> class keeps track of a list of <a
3059 href="#Function"><tt>Function</tt></a>s, a list of <a
3060 href="#GlobalVariable"><tt>GlobalVariable</tt></a>s, and a <a
3061 href="#SymbolTable"><tt>SymbolTable</tt></a>. Additionally, it contains a few
3062 helpful member functions that try to make common operations easy.</p>
3064 <!-- _______________________________________________________________________ -->
3066 <a name="m_Module">Important Public Members of the <tt>Module</tt> class</a>
3072 <li><tt>Module::Module(std::string name = "")</tt></li>
3075 <p>Constructing a <a href="#Module">Module</a> is easy. You can optionally
3076 provide a name for it (probably based on the name of the translation unit).</p>
3079 <li><tt>Module::iterator</tt> - Typedef for function list iterator<br>
3080 <tt>Module::const_iterator</tt> - Typedef for const_iterator.<br>
3082 <tt>begin()</tt>, <tt>end()</tt>
3083 <tt>size()</tt>, <tt>empty()</tt>
3085 <p>These are forwarding methods that make it easy to access the contents of
3086 a <tt>Module</tt> object's <a href="#Function"><tt>Function</tt></a>
3089 <li><tt>Module::FunctionListType &getFunctionList()</tt>
3091 <p> Returns the list of <a href="#Function"><tt>Function</tt></a>s. This is
3092 necessary to use when you need to update the list or perform a complex
3093 action that doesn't have a forwarding method.</p>
3095 <p><!-- Global Variable --></p></li>
3101 <li><tt>Module::global_iterator</tt> - Typedef for global variable list iterator<br>
3103 <tt>Module::const_global_iterator</tt> - Typedef for const_iterator.<br>
3105 <tt>global_begin()</tt>, <tt>global_end()</tt>
3106 <tt>global_size()</tt>, <tt>global_empty()</tt>
3108 <p> These are forwarding methods that make it easy to access the contents of
3109 a <tt>Module</tt> object's <a
3110 href="#GlobalVariable"><tt>GlobalVariable</tt></a> list.</p></li>
3112 <li><tt>Module::GlobalListType &getGlobalList()</tt>
3114 <p>Returns the list of <a
3115 href="#GlobalVariable"><tt>GlobalVariable</tt></a>s. This is necessary to
3116 use when you need to update the list or perform a complex action that
3117 doesn't have a forwarding method.</p>
3119 <p><!-- Symbol table stuff --> </p></li>
3125 <li><tt><a href="#SymbolTable">SymbolTable</a> *getSymbolTable()</tt>
3127 <p>Return a reference to the <a href="#SymbolTable"><tt>SymbolTable</tt></a>
3128 for this <tt>Module</tt>.</p>
3130 <p><!-- Convenience methods --></p></li>
3136 <li><tt><a href="#Function">Function</a> *getFunction(const std::string
3137 &Name, const <a href="#FunctionType">FunctionType</a> *Ty)</tt>
3139 <p>Look up the specified function in the <tt>Module</tt> <a
3140 href="#SymbolTable"><tt>SymbolTable</tt></a>. If it does not exist, return
3141 <tt>null</tt>.</p></li>
3143 <li><tt><a href="#Function">Function</a> *getOrInsertFunction(const
3144 std::string &Name, const <a href="#FunctionType">FunctionType</a> *T)</tt>
3146 <p>Look up the specified function in the <tt>Module</tt> <a
3147 href="#SymbolTable"><tt>SymbolTable</tt></a>. If it does not exist, add an
3148 external declaration for the function and return it.</p></li>
3150 <li><tt>std::string getTypeName(const <a href="#Type">Type</a> *Ty)</tt>
3152 <p>If there is at least one entry in the <a
3153 href="#SymbolTable"><tt>SymbolTable</tt></a> for the specified <a
3154 href="#Type"><tt>Type</tt></a>, return it. Otherwise return the empty
3157 <li><tt>bool addTypeName(const std::string &Name, const <a
3158 href="#Type">Type</a> *Ty)</tt>
3160 <p>Insert an entry in the <a href="#SymbolTable"><tt>SymbolTable</tt></a>
3161 mapping <tt>Name</tt> to <tt>Ty</tt>. If there is already an entry for this
3162 name, true is returned and the <a
3163 href="#SymbolTable"><tt>SymbolTable</tt></a> is not modified.</p></li>
3170 <!-- ======================================================================= -->
3172 <a name="Value">The <tt>Value</tt> class</a>
3177 <p><tt>#include "<a href="/doxygen/Value_8h-source.html">llvm/Value.h</a>"</tt>
3179 doxygen info: <a href="/doxygen/classllvm_1_1Value.html">Value Class</a></p>
3181 <p>The <tt>Value</tt> class is the most important class in the LLVM Source
3182 base. It represents a typed value that may be used (among other things) as an
3183 operand to an instruction. There are many different types of <tt>Value</tt>s,
3184 such as <a href="#Constant"><tt>Constant</tt></a>s,<a
3185 href="#Argument"><tt>Argument</tt></a>s. Even <a
3186 href="#Instruction"><tt>Instruction</tt></a>s and <a
3187 href="#Function"><tt>Function</tt></a>s are <tt>Value</tt>s.</p>
3189 <p>A particular <tt>Value</tt> may be used many times in the LLVM representation
3190 for a program. For example, an incoming argument to a function (represented
3191 with an instance of the <a href="#Argument">Argument</a> class) is "used" by
3192 every instruction in the function that references the argument. To keep track
3193 of this relationship, the <tt>Value</tt> class keeps a list of all of the <a
3194 href="#User"><tt>User</tt></a>s that is using it (the <a
3195 href="#User"><tt>User</tt></a> class is a base class for all nodes in the LLVM
3196 graph that can refer to <tt>Value</tt>s). This use list is how LLVM represents
3197 def-use information in the program, and is accessible through the <tt>use_</tt>*
3198 methods, shown below.</p>
3200 <p>Because LLVM is a typed representation, every LLVM <tt>Value</tt> is typed,
3201 and this <a href="#Type">Type</a> is available through the <tt>getType()</tt>
3202 method. In addition, all LLVM values can be named. The "name" of the
3203 <tt>Value</tt> is a symbolic string printed in the LLVM code:</p>
3205 <div class="doc_code">
3207 %<b>foo</b> = add i32 1, 2
3211 <p><a name="nameWarning">The name of this instruction is "foo".</a> <b>NOTE</b>
3212 that the name of any value may be missing (an empty string), so names should
3213 <b>ONLY</b> be used for debugging (making the source code easier to read,
3214 debugging printouts), they should not be used to keep track of values or map
3215 between them. For this purpose, use a <tt>std::map</tt> of pointers to the
3216 <tt>Value</tt> itself instead.</p>
3218 <p>One important aspect of LLVM is that there is no distinction between an SSA
3219 variable and the operation that produces it. Because of this, any reference to
3220 the value produced by an instruction (or the value available as an incoming
3221 argument, for example) is represented as a direct pointer to the instance of
3223 represents this value. Although this may take some getting used to, it
3224 simplifies the representation and makes it easier to manipulate.</p>
3226 <!-- _______________________________________________________________________ -->
3228 <a name="m_Value">Important Public Members of the <tt>Value</tt> class</a>
3234 <li><tt>Value::use_iterator</tt> - Typedef for iterator over the
3236 <tt>Value::const_use_iterator</tt> - Typedef for const_iterator over
3238 <tt>unsigned use_size()</tt> - Returns the number of users of the
3240 <tt>bool use_empty()</tt> - Returns true if there are no users.<br>
3241 <tt>use_iterator use_begin()</tt> - Get an iterator to the start of
3243 <tt>use_iterator use_end()</tt> - Get an iterator to the end of the
3245 <tt><a href="#User">User</a> *use_back()</tt> - Returns the last
3246 element in the list.
3247 <p> These methods are the interface to access the def-use
3248 information in LLVM. As with all other iterators in LLVM, the naming
3249 conventions follow the conventions defined by the <a href="#stl">STL</a>.</p>
3251 <li><tt><a href="#Type">Type</a> *getType() const</tt>
3252 <p>This method returns the Type of the Value.</p>
3254 <li><tt>bool hasName() const</tt><br>
3255 <tt>std::string getName() const</tt><br>
3256 <tt>void setName(const std::string &Name)</tt>
3257 <p> This family of methods is used to access and assign a name to a <tt>Value</tt>,
3258 be aware of the <a href="#nameWarning">precaution above</a>.</p>
3260 <li><tt>void replaceAllUsesWith(Value *V)</tt>
3262 <p>This method traverses the use list of a <tt>Value</tt> changing all <a
3263 href="#User"><tt>User</tt>s</a> of the current value to refer to
3264 "<tt>V</tt>" instead. For example, if you detect that an instruction always
3265 produces a constant value (for example through constant folding), you can
3266 replace all uses of the instruction with the constant like this:</p>
3268 <div class="doc_code">
3270 Inst->replaceAllUsesWith(ConstVal);
3280 <!-- ======================================================================= -->
3282 <a name="User">The <tt>User</tt> class</a>
3288 <tt>#include "<a href="/doxygen/User_8h-source.html">llvm/User.h</a>"</tt><br>
3289 doxygen info: <a href="/doxygen/classllvm_1_1User.html">User Class</a><br>
3290 Superclass: <a href="#Value"><tt>Value</tt></a></p>
3292 <p>The <tt>User</tt> class is the common base class of all LLVM nodes that may
3293 refer to <a href="#Value"><tt>Value</tt></a>s. It exposes a list of "Operands"
3294 that are all of the <a href="#Value"><tt>Value</tt></a>s that the User is
3295 referring to. The <tt>User</tt> class itself is a subclass of
3298 <p>The operands of a <tt>User</tt> point directly to the LLVM <a
3299 href="#Value"><tt>Value</tt></a> that it refers to. Because LLVM uses Static
3300 Single Assignment (SSA) form, there can only be one definition referred to,
3301 allowing this direct connection. This connection provides the use-def
3302 information in LLVM.</p>
3304 <!-- _______________________________________________________________________ -->
3306 <a name="m_User">Important Public Members of the <tt>User</tt> class</a>
3311 <p>The <tt>User</tt> class exposes the operand list in two ways: through
3312 an index access interface and through an iterator based interface.</p>
3315 <li><tt>Value *getOperand(unsigned i)</tt><br>
3316 <tt>unsigned getNumOperands()</tt>
3317 <p> These two methods expose the operands of the <tt>User</tt> in a
3318 convenient form for direct access.</p></li>
3320 <li><tt>User::op_iterator</tt> - Typedef for iterator over the operand
3322 <tt>op_iterator op_begin()</tt> - Get an iterator to the start of
3323 the operand list.<br>
3324 <tt>op_iterator op_end()</tt> - Get an iterator to the end of the
3326 <p> Together, these methods make up the iterator based interface to
3327 the operands of a <tt>User</tt>.</p></li>
3334 <!-- ======================================================================= -->
3336 <a name="Instruction">The <tt>Instruction</tt> class</a>
3341 <p><tt>#include "</tt><tt><a
3342 href="/doxygen/Instruction_8h-source.html">llvm/Instruction.h</a>"</tt><br>
3343 doxygen info: <a href="/doxygen/classllvm_1_1Instruction.html">Instruction Class</a><br>
3344 Superclasses: <a href="#User"><tt>User</tt></a>, <a
3345 href="#Value"><tt>Value</tt></a></p>
3347 <p>The <tt>Instruction</tt> class is the common base class for all LLVM
3348 instructions. It provides only a few methods, but is a very commonly used
3349 class. The primary data tracked by the <tt>Instruction</tt> class itself is the
3350 opcode (instruction type) and the parent <a
3351 href="#BasicBlock"><tt>BasicBlock</tt></a> the <tt>Instruction</tt> is embedded
3352 into. To represent a specific type of instruction, one of many subclasses of
3353 <tt>Instruction</tt> are used.</p>
3355 <p> Because the <tt>Instruction</tt> class subclasses the <a
3356 href="#User"><tt>User</tt></a> class, its operands can be accessed in the same
3357 way as for other <a href="#User"><tt>User</tt></a>s (with the
3358 <tt>getOperand()</tt>/<tt>getNumOperands()</tt> and
3359 <tt>op_begin()</tt>/<tt>op_end()</tt> methods).</p> <p> An important file for
3360 the <tt>Instruction</tt> class is the <tt>llvm/Instruction.def</tt> file. This
3361 file contains some meta-data about the various different types of instructions
3362 in LLVM. It describes the enum values that are used as opcodes (for example
3363 <tt>Instruction::Add</tt> and <tt>Instruction::ICmp</tt>), as well as the
3364 concrete sub-classes of <tt>Instruction</tt> that implement the instruction (for
3365 example <tt><a href="#BinaryOperator">BinaryOperator</a></tt> and <tt><a
3366 href="#CmpInst">CmpInst</a></tt>). Unfortunately, the use of macros in
3367 this file confuses doxygen, so these enum values don't show up correctly in the
3368 <a href="/doxygen/classllvm_1_1Instruction.html">doxygen output</a>.</p>
3370 <!-- _______________________________________________________________________ -->
3372 <a name="s_Instruction">
3373 Important Subclasses of the <tt>Instruction</tt> class
3378 <li><tt><a name="BinaryOperator">BinaryOperator</a></tt>
3379 <p>This subclasses represents all two operand instructions whose operands
3380 must be the same type, except for the comparison instructions.</p></li>
3381 <li><tt><a name="CastInst">CastInst</a></tt>
3382 <p>This subclass is the parent of the 12 casting instructions. It provides
3383 common operations on cast instructions.</p>
3384 <li><tt><a name="CmpInst">CmpInst</a></tt>
3385 <p>This subclass respresents the two comparison instructions,
3386 <a href="LangRef.html#i_icmp">ICmpInst</a> (integer opreands), and
3387 <a href="LangRef.html#i_fcmp">FCmpInst</a> (floating point operands).</p>
3388 <li><tt><a name="TerminatorInst">TerminatorInst</a></tt>
3389 <p>This subclass is the parent of all terminator instructions (those which
3390 can terminate a block).</p>
3394 <!-- _______________________________________________________________________ -->
3396 <a name="m_Instruction">
3397 Important Public Members of the <tt>Instruction</tt> class
3404 <li><tt><a href="#BasicBlock">BasicBlock</a> *getParent()</tt>
3405 <p>Returns the <a href="#BasicBlock"><tt>BasicBlock</tt></a> that
3406 this <tt>Instruction</tt> is embedded into.</p></li>
3407 <li><tt>bool mayWriteToMemory()</tt>
3408 <p>Returns true if the instruction writes to memory, i.e. it is a
3409 <tt>call</tt>,<tt>free</tt>,<tt>invoke</tt>, or <tt>store</tt>.</p></li>
3410 <li><tt>unsigned getOpcode()</tt>
3411 <p>Returns the opcode for the <tt>Instruction</tt>.</p></li>
3412 <li><tt><a href="#Instruction">Instruction</a> *clone() const</tt>
3413 <p>Returns another instance of the specified instruction, identical
3414 in all ways to the original except that the instruction has no parent
3415 (ie it's not embedded into a <a href="#BasicBlock"><tt>BasicBlock</tt></a>),
3416 and it has no name</p></li>
3423 <!-- ======================================================================= -->
3425 <a name="Constant">The <tt>Constant</tt> class and subclasses</a>
3430 <p>Constant represents a base class for different types of constants. It
3431 is subclassed by ConstantInt, ConstantArray, etc. for representing
3432 the various types of Constants. <a href="#GlobalValue">GlobalValue</a> is also
3433 a subclass, which represents the address of a global variable or function.
3436 <!-- _______________________________________________________________________ -->
3437 <h4>Important Subclasses of Constant</h4>
3440 <li>ConstantInt : This subclass of Constant represents an integer constant of
3443 <li><tt>const APInt& getValue() const</tt>: Returns the underlying
3444 value of this constant, an APInt value.</li>
3445 <li><tt>int64_t getSExtValue() const</tt>: Converts the underlying APInt
3446 value to an int64_t via sign extension. If the value (not the bit width)
3447 of the APInt is too large to fit in an int64_t, an assertion will result.
3448 For this reason, use of this method is discouraged.</li>
3449 <li><tt>uint64_t getZExtValue() const</tt>: Converts the underlying APInt
3450 value to a uint64_t via zero extension. IF the value (not the bit width)
3451 of the APInt is too large to fit in a uint64_t, an assertion will result.
3452 For this reason, use of this method is discouraged.</li>
3453 <li><tt>static ConstantInt* get(const APInt& Val)</tt>: Returns the
3454 ConstantInt object that represents the value provided by <tt>Val</tt>.
3455 The type is implied as the IntegerType that corresponds to the bit width
3456 of <tt>Val</tt>.</li>
3457 <li><tt>static ConstantInt* get(const Type *Ty, uint64_t Val)</tt>:
3458 Returns the ConstantInt object that represents the value provided by
3459 <tt>Val</tt> for integer type <tt>Ty</tt>.</li>
3462 <li>ConstantFP : This class represents a floating point constant.
3464 <li><tt>double getValue() const</tt>: Returns the underlying value of
3465 this constant. </li>
3468 <li>ConstantArray : This represents a constant array.
3470 <li><tt>const std::vector<Use> &getValues() const</tt>: Returns
3471 a vector of component constants that makeup this array. </li>
3474 <li>ConstantStruct : This represents a constant struct.
3476 <li><tt>const std::vector<Use> &getValues() const</tt>: Returns
3477 a vector of component constants that makeup this array. </li>
3480 <li>GlobalValue : This represents either a global variable or a function. In
3481 either case, the value is a constant fixed address (after linking).
3488 <!-- ======================================================================= -->
3490 <a name="GlobalValue">The <tt>GlobalValue</tt> class</a>
3496 href="/doxygen/GlobalValue_8h-source.html">llvm/GlobalValue.h</a>"</tt><br>
3497 doxygen info: <a href="/doxygen/classllvm_1_1GlobalValue.html">GlobalValue
3499 Superclasses: <a href="#Constant"><tt>Constant</tt></a>,
3500 <a href="#User"><tt>User</tt></a>, <a href="#Value"><tt>Value</tt></a></p>
3502 <p>Global values (<a href="#GlobalVariable"><tt>GlobalVariable</tt></a>s or <a
3503 href="#Function"><tt>Function</tt></a>s) are the only LLVM values that are
3504 visible in the bodies of all <a href="#Function"><tt>Function</tt></a>s.
3505 Because they are visible at global scope, they are also subject to linking with
3506 other globals defined in different translation units. To control the linking
3507 process, <tt>GlobalValue</tt>s know their linkage rules. Specifically,
3508 <tt>GlobalValue</tt>s know whether they have internal or external linkage, as
3509 defined by the <tt>LinkageTypes</tt> enumeration.</p>
3511 <p>If a <tt>GlobalValue</tt> has internal linkage (equivalent to being
3512 <tt>static</tt> in C), it is not visible to code outside the current translation
3513 unit, and does not participate in linking. If it has external linkage, it is
3514 visible to external code, and does participate in linking. In addition to
3515 linkage information, <tt>GlobalValue</tt>s keep track of which <a
3516 href="#Module"><tt>Module</tt></a> they are currently part of.</p>
3518 <p>Because <tt>GlobalValue</tt>s are memory objects, they are always referred to
3519 by their <b>address</b>. As such, the <a href="#Type"><tt>Type</tt></a> of a
3520 global is always a pointer to its contents. It is important to remember this
3521 when using the <tt>GetElementPtrInst</tt> instruction because this pointer must
3522 be dereferenced first. For example, if you have a <tt>GlobalVariable</tt> (a
3523 subclass of <tt>GlobalValue)</tt> that is an array of 24 ints, type <tt>[24 x
3524 i32]</tt>, then the <tt>GlobalVariable</tt> is a pointer to that array. Although
3525 the address of the first element of this array and the value of the
3526 <tt>GlobalVariable</tt> are the same, they have different types. The
3527 <tt>GlobalVariable</tt>'s type is <tt>[24 x i32]</tt>. The first element's type
3528 is <tt>i32.</tt> Because of this, accessing a global value requires you to
3529 dereference the pointer with <tt>GetElementPtrInst</tt> first, then its elements
3530 can be accessed. This is explained in the <a href="LangRef.html#globalvars">LLVM
3531 Language Reference Manual</a>.</p>
3533 <!-- _______________________________________________________________________ -->
3535 <a name="m_GlobalValue">
3536 Important Public Members of the <tt>GlobalValue</tt> class
3543 <li><tt>bool hasInternalLinkage() const</tt><br>
3544 <tt>bool hasExternalLinkage() const</tt><br>
3545 <tt>void setInternalLinkage(bool HasInternalLinkage)</tt>
3546 <p> These methods manipulate the linkage characteristics of the <tt>GlobalValue</tt>.</p>
3549 <li><tt><a href="#Module">Module</a> *getParent()</tt>
3550 <p> This returns the <a href="#Module"><tt>Module</tt></a> that the
3551 GlobalValue is currently embedded into.</p></li>
3558 <!-- ======================================================================= -->
3560 <a name="Function">The <tt>Function</tt> class</a>
3566 href="/doxygen/Function_8h-source.html">llvm/Function.h</a>"</tt><br> doxygen
3567 info: <a href="/doxygen/classllvm_1_1Function.html">Function Class</a><br>
3568 Superclasses: <a href="#GlobalValue"><tt>GlobalValue</tt></a>,
3569 <a href="#Constant"><tt>Constant</tt></a>,
3570 <a href="#User"><tt>User</tt></a>,
3571 <a href="#Value"><tt>Value</tt></a></p>
3573 <p>The <tt>Function</tt> class represents a single procedure in LLVM. It is
3574 actually one of the more complex classes in the LLVM hierarchy because it must
3575 keep track of a large amount of data. The <tt>Function</tt> class keeps track
3576 of a list of <a href="#BasicBlock"><tt>BasicBlock</tt></a>s, a list of formal
3577 <a href="#Argument"><tt>Argument</tt></a>s, and a
3578 <a href="#SymbolTable"><tt>SymbolTable</tt></a>.</p>
3580 <p>The list of <a href="#BasicBlock"><tt>BasicBlock</tt></a>s is the most
3581 commonly used part of <tt>Function</tt> objects. The list imposes an implicit
3582 ordering of the blocks in the function, which indicate how the code will be
3583 laid out by the backend. Additionally, the first <a
3584 href="#BasicBlock"><tt>BasicBlock</tt></a> is the implicit entry node for the
3585 <tt>Function</tt>. It is not legal in LLVM to explicitly branch to this initial
3586 block. There are no implicit exit nodes, and in fact there may be multiple exit
3587 nodes from a single <tt>Function</tt>. If the <a
3588 href="#BasicBlock"><tt>BasicBlock</tt></a> list is empty, this indicates that
3589 the <tt>Function</tt> is actually a function declaration: the actual body of the
3590 function hasn't been linked in yet.</p>
3592 <p>In addition to a list of <a href="#BasicBlock"><tt>BasicBlock</tt></a>s, the
3593 <tt>Function</tt> class also keeps track of the list of formal <a
3594 href="#Argument"><tt>Argument</tt></a>s that the function receives. This
3595 container manages the lifetime of the <a href="#Argument"><tt>Argument</tt></a>
3596 nodes, just like the <a href="#BasicBlock"><tt>BasicBlock</tt></a> list does for
3597 the <a href="#BasicBlock"><tt>BasicBlock</tt></a>s.</p>
3599 <p>The <a href="#SymbolTable"><tt>SymbolTable</tt></a> is a very rarely used
3600 LLVM feature that is only used when you have to look up a value by name. Aside
3601 from that, the <a href="#SymbolTable"><tt>SymbolTable</tt></a> is used
3602 internally to make sure that there are not conflicts between the names of <a
3603 href="#Instruction"><tt>Instruction</tt></a>s, <a
3604 href="#BasicBlock"><tt>BasicBlock</tt></a>s, or <a
3605 href="#Argument"><tt>Argument</tt></a>s in the function body.</p>
3607 <p>Note that <tt>Function</tt> is a <a href="#GlobalValue">GlobalValue</a>
3608 and therefore also a <a href="#Constant">Constant</a>. The value of the function
3609 is its address (after linking) which is guaranteed to be constant.</p>
3611 <!-- _______________________________________________________________________ -->
3613 <a name="m_Function">
3614 Important Public Members of the <tt>Function</tt> class
3621 <li><tt>Function(const </tt><tt><a href="#FunctionType">FunctionType</a>
3622 *Ty, LinkageTypes Linkage, const std::string &N = "", Module* Parent = 0)</tt>
3624 <p>Constructor used when you need to create new <tt>Function</tt>s to add
3625 the the program. The constructor must specify the type of the function to
3626 create and what type of linkage the function should have. The <a
3627 href="#FunctionType"><tt>FunctionType</tt></a> argument
3628 specifies the formal arguments and return value for the function. The same
3629 <a href="#FunctionType"><tt>FunctionType</tt></a> value can be used to
3630 create multiple functions. The <tt>Parent</tt> argument specifies the Module
3631 in which the function is defined. If this argument is provided, the function
3632 will automatically be inserted into that module's list of
3635 <li><tt>bool isDeclaration()</tt>
3637 <p>Return whether or not the <tt>Function</tt> has a body defined. If the
3638 function is "external", it does not have a body, and thus must be resolved
3639 by linking with a function defined in a different translation unit.</p></li>
3641 <li><tt>Function::iterator</tt> - Typedef for basic block list iterator<br>
3642 <tt>Function::const_iterator</tt> - Typedef for const_iterator.<br>
3644 <tt>begin()</tt>, <tt>end()</tt>
3645 <tt>size()</tt>, <tt>empty()</tt>
3647 <p>These are forwarding methods that make it easy to access the contents of
3648 a <tt>Function</tt> object's <a href="#BasicBlock"><tt>BasicBlock</tt></a>
3651 <li><tt>Function::BasicBlockListType &getBasicBlockList()</tt>
3653 <p>Returns the list of <a href="#BasicBlock"><tt>BasicBlock</tt></a>s. This
3654 is necessary to use when you need to update the list or perform a complex
3655 action that doesn't have a forwarding method.</p></li>
3657 <li><tt>Function::arg_iterator</tt> - Typedef for the argument list
3659 <tt>Function::const_arg_iterator</tt> - Typedef for const_iterator.<br>
3661 <tt>arg_begin()</tt>, <tt>arg_end()</tt>
3662 <tt>arg_size()</tt>, <tt>arg_empty()</tt>
3664 <p>These are forwarding methods that make it easy to access the contents of
3665 a <tt>Function</tt> object's <a href="#Argument"><tt>Argument</tt></a>
3668 <li><tt>Function::ArgumentListType &getArgumentList()</tt>
3670 <p>Returns the list of <a href="#Argument"><tt>Argument</tt></a>s. This is
3671 necessary to use when you need to update the list or perform a complex
3672 action that doesn't have a forwarding method.</p></li>
3674 <li><tt><a href="#BasicBlock">BasicBlock</a> &getEntryBlock()</tt>
3676 <p>Returns the entry <a href="#BasicBlock"><tt>BasicBlock</tt></a> for the
3677 function. Because the entry block for the function is always the first
3678 block, this returns the first block of the <tt>Function</tt>.</p></li>
3680 <li><tt><a href="#Type">Type</a> *getReturnType()</tt><br>
3681 <tt><a href="#FunctionType">FunctionType</a> *getFunctionType()</tt>
3683 <p>This traverses the <a href="#Type"><tt>Type</tt></a> of the
3684 <tt>Function</tt> and returns the return type of the function, or the <a
3685 href="#FunctionType"><tt>FunctionType</tt></a> of the actual
3688 <li><tt><a href="#SymbolTable">SymbolTable</a> *getSymbolTable()</tt>
3690 <p> Return a pointer to the <a href="#SymbolTable"><tt>SymbolTable</tt></a>
3691 for this <tt>Function</tt>.</p></li>
3698 <!-- ======================================================================= -->
3700 <a name="GlobalVariable">The <tt>GlobalVariable</tt> class</a>
3706 href="/doxygen/GlobalVariable_8h-source.html">llvm/GlobalVariable.h</a>"</tt>
3708 doxygen info: <a href="/doxygen/classllvm_1_1GlobalVariable.html">GlobalVariable
3710 Superclasses: <a href="#GlobalValue"><tt>GlobalValue</tt></a>,
3711 <a href="#Constant"><tt>Constant</tt></a>,
3712 <a href="#User"><tt>User</tt></a>,
3713 <a href="#Value"><tt>Value</tt></a></p>
3715 <p>Global variables are represented with the (surprise surprise)
3716 <tt>GlobalVariable</tt> class. Like functions, <tt>GlobalVariable</tt>s are also
3717 subclasses of <a href="#GlobalValue"><tt>GlobalValue</tt></a>, and as such are
3718 always referenced by their address (global values must live in memory, so their
3719 "name" refers to their constant address). See
3720 <a href="#GlobalValue"><tt>GlobalValue</tt></a> for more on this. Global
3721 variables may have an initial value (which must be a
3722 <a href="#Constant"><tt>Constant</tt></a>), and if they have an initializer,
3723 they may be marked as "constant" themselves (indicating that their contents
3724 never change at runtime).</p>
3726 <!-- _______________________________________________________________________ -->
3728 <a name="m_GlobalVariable">
3729 Important Public Members of the <tt>GlobalVariable</tt> class
3736 <li><tt>GlobalVariable(const </tt><tt><a href="#Type">Type</a> *Ty, bool
3737 isConstant, LinkageTypes& Linkage, <a href="#Constant">Constant</a>
3738 *Initializer = 0, const std::string &Name = "", Module* Parent = 0)</tt>
3740 <p>Create a new global variable of the specified type. If
3741 <tt>isConstant</tt> is true then the global variable will be marked as
3742 unchanging for the program. The Linkage parameter specifies the type of
3743 linkage (internal, external, weak, linkonce, appending) for the variable.
3744 If the linkage is InternalLinkage, WeakAnyLinkage, WeakODRLinkage,
3745 LinkOnceAnyLinkage or LinkOnceODRLinkage, then the resultant
3746 global variable will have internal linkage. AppendingLinkage concatenates
3747 together all instances (in different translation units) of the variable
3748 into a single variable but is only applicable to arrays. See
3749 the <a href="LangRef.html#modulestructure">LLVM Language Reference</a> for
3750 further details on linkage types. Optionally an initializer, a name, and the
3751 module to put the variable into may be specified for the global variable as
3754 <li><tt>bool isConstant() const</tt>
3756 <p>Returns true if this is a global variable that is known not to
3757 be modified at runtime.</p></li>
3759 <li><tt>bool hasInitializer()</tt>
3761 <p>Returns true if this <tt>GlobalVariable</tt> has an intializer.</p></li>
3763 <li><tt><a href="#Constant">Constant</a> *getInitializer()</tt>
3765 <p>Returns the initial value for a <tt>GlobalVariable</tt>. It is not legal
3766 to call this method if there is no initializer.</p></li>
3773 <!-- ======================================================================= -->
3775 <a name="BasicBlock">The <tt>BasicBlock</tt> class</a>
3781 href="/doxygen/BasicBlock_8h-source.html">llvm/BasicBlock.h</a>"</tt><br>
3782 doxygen info: <a href="/doxygen/classllvm_1_1BasicBlock.html">BasicBlock
3784 Superclass: <a href="#Value"><tt>Value</tt></a></p>
3786 <p>This class represents a single entry single exit section of the code,
3787 commonly known as a basic block by the compiler community. The
3788 <tt>BasicBlock</tt> class maintains a list of <a
3789 href="#Instruction"><tt>Instruction</tt></a>s, which form the body of the block.
3790 Matching the language definition, the last element of this list of instructions
3791 is always a terminator instruction (a subclass of the <a
3792 href="#TerminatorInst"><tt>TerminatorInst</tt></a> class).</p>
3794 <p>In addition to tracking the list of instructions that make up the block, the
3795 <tt>BasicBlock</tt> class also keeps track of the <a
3796 href="#Function"><tt>Function</tt></a> that it is embedded into.</p>
3798 <p>Note that <tt>BasicBlock</tt>s themselves are <a
3799 href="#Value"><tt>Value</tt></a>s, because they are referenced by instructions
3800 like branches and can go in the switch tables. <tt>BasicBlock</tt>s have type
3803 <!-- _______________________________________________________________________ -->
3805 <a name="m_BasicBlock">
3806 Important Public Members of the <tt>BasicBlock</tt> class
3813 <li><tt>BasicBlock(const std::string &Name = "", </tt><tt><a
3814 href="#Function">Function</a> *Parent = 0)</tt>
3816 <p>The <tt>BasicBlock</tt> constructor is used to create new basic blocks for
3817 insertion into a function. The constructor optionally takes a name for the new
3818 block, and a <a href="#Function"><tt>Function</tt></a> to insert it into. If
3819 the <tt>Parent</tt> parameter is specified, the new <tt>BasicBlock</tt> is
3820 automatically inserted at the end of the specified <a
3821 href="#Function"><tt>Function</tt></a>, if not specified, the BasicBlock must be
3822 manually inserted into the <a href="#Function"><tt>Function</tt></a>.</p></li>
3824 <li><tt>BasicBlock::iterator</tt> - Typedef for instruction list iterator<br>
3825 <tt>BasicBlock::const_iterator</tt> - Typedef for const_iterator.<br>
3826 <tt>begin()</tt>, <tt>end()</tt>, <tt>front()</tt>, <tt>back()</tt>,
3827 <tt>size()</tt>, <tt>empty()</tt>
3828 STL-style functions for accessing the instruction list.
3830 <p>These methods and typedefs are forwarding functions that have the same
3831 semantics as the standard library methods of the same names. These methods
3832 expose the underlying instruction list of a basic block in a way that is easy to
3833 manipulate. To get the full complement of container operations (including
3834 operations to update the list), you must use the <tt>getInstList()</tt>
3837 <li><tt>BasicBlock::InstListType &getInstList()</tt>
3839 <p>This method is used to get access to the underlying container that actually
3840 holds the Instructions. This method must be used when there isn't a forwarding
3841 function in the <tt>BasicBlock</tt> class for the operation that you would like
3842 to perform. Because there are no forwarding functions for "updating"
3843 operations, you need to use this if you want to update the contents of a
3844 <tt>BasicBlock</tt>.</p></li>
3846 <li><tt><a href="#Function">Function</a> *getParent()</tt>
3848 <p> Returns a pointer to <a href="#Function"><tt>Function</tt></a> the block is
3849 embedded into, or a null pointer if it is homeless.</p></li>
3851 <li><tt><a href="#TerminatorInst">TerminatorInst</a> *getTerminator()</tt>
3853 <p> Returns a pointer to the terminator instruction that appears at the end of
3854 the <tt>BasicBlock</tt>. If there is no terminator instruction, or if the last
3855 instruction in the block is not a terminator, then a null pointer is
3864 <!-- ======================================================================= -->
3866 <a name="Argument">The <tt>Argument</tt> class</a>
3871 <p>This subclass of Value defines the interface for incoming formal
3872 arguments to a function. A Function maintains a list of its formal
3873 arguments. An argument has a pointer to the parent Function.</p>
3879 <!-- *********************************************************************** -->
3882 <a href="http://jigsaw.w3.org/css-validator/check/referer"><img
3883 src="http://jigsaw.w3.org/css-validator/images/vcss-blue" alt="Valid CSS"></a>
3884 <a href="http://validator.w3.org/check/referer"><img
3885 src="http://www.w3.org/Icons/valid-html401" alt="Valid HTML 4.01 Strict"></a>
3887 <a href="mailto:dhurjati@cs.uiuc.edu">Dinakar Dhurjati</a> and
3888 <a href="mailto:sabre@nondot.org">Chris Lattner</a><br>
3889 <a href="http://llvm.org/">The LLVM Compiler Infrastructure</a><br>
3890 Last modified: $Date$