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="_static/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>
73 <li><a href="#dss_stringref">llvm/ADT/StringRef.h</a></li>
74 <li><a href="#dss_twine">llvm/ADT/Twine.h</a></li>
75 <li><a href="#dss_smallstring">llvm/ADT/SmallString.h</a></li>
76 <li><a href="#dss_stdstring">std::string</a></li>
78 <li><a href="#ds_set">Set-Like Containers (std::set, SmallSet, SetVector, etc)</a>
80 <li><a href="#dss_sortedvectorset">A sorted 'vector'</a></li>
81 <li><a href="#dss_smallset">"llvm/ADT/SmallSet.h"</a></li>
82 <li><a href="#dss_smallptrset">"llvm/ADT/SmallPtrSet.h"</a></li>
83 <li><a href="#dss_denseset">"llvm/ADT/DenseSet.h"</a></li>
84 <li><a href="#dss_sparseset">"llvm/ADT/SparseSet.h"</a></li>
85 <li><a href="#dss_FoldingSet">"llvm/ADT/FoldingSet.h"</a></li>
86 <li><a href="#dss_set"><set></a></li>
87 <li><a href="#dss_setvector">"llvm/ADT/SetVector.h"</a></li>
88 <li><a href="#dss_uniquevector">"llvm/ADT/UniqueVector.h"</a></li>
89 <li><a href="#dss_immutableset">"llvm/ADT/ImmutableSet.h"</a></li>
90 <li><a href="#dss_otherset">Other Set-Like Container Options</a></li>
92 <li><a href="#ds_map">Map-Like Containers (std::map, DenseMap, etc)</a>
94 <li><a href="#dss_sortedvectormap">A sorted 'vector'</a></li>
95 <li><a href="#dss_stringmap">"llvm/ADT/StringMap.h"</a></li>
96 <li><a href="#dss_indexedmap">"llvm/ADT/IndexedMap.h"</a></li>
97 <li><a href="#dss_densemap">"llvm/ADT/DenseMap.h"</a></li>
98 <li><a href="#dss_multiimplmap">"llvm/ADT/MultiImplMap.h"</a></li>
99 <li><a href="#dss_flatarraymap">"llvm/ADT/FlatArrayMap.h"</a></li>
100 <li><a href="#dss_smallmap">"llvm/ADT/SmallMap.h"</a></li>
101 <li><a href="#dss_valuemap">"llvm/ADT/ValueMap.h"</a></li>
102 <li><a href="#dss_intervalmap">"llvm/ADT/IntervalMap.h"</a></li>
103 <li><a href="#dss_map"><map></a></li>
104 <li><a href="#dss_inteqclasses">"llvm/ADT/IntEqClasses.h"</a></li>
105 <li><a href="#dss_immutablemap">"llvm/ADT/ImmutableMap.h"</a></li>
106 <li><a href="#dss_othermap">Other Map-Like Container Options</a></li>
108 <li><a href="#ds_bit">BitVector-like containers</a>
110 <li><a href="#dss_bitvector">A dense bitvector</a></li>
111 <li><a href="#dss_smallbitvector">A "small" dense bitvector</a></li>
112 <li><a href="#dss_sparsebitvector">A sparse bitvector</a></li>
116 <li><a href="#common">Helpful Hints for Common Operations</a>
118 <li><a href="#inspection">Basic Inspection and Traversal Routines</a>
120 <li><a href="#iterate_function">Iterating over the <tt>BasicBlock</tt>s
121 in a <tt>Function</tt></a> </li>
122 <li><a href="#iterate_basicblock">Iterating over the <tt>Instruction</tt>s
123 in a <tt>BasicBlock</tt></a> </li>
124 <li><a href="#iterate_institer">Iterating over the <tt>Instruction</tt>s
125 in a <tt>Function</tt></a> </li>
126 <li><a href="#iterate_convert">Turning an iterator into a
127 class pointer</a> </li>
128 <li><a href="#iterate_complex">Finding call sites: a more
129 complex example</a> </li>
130 <li><a href="#calls_and_invokes">Treating calls and invokes
131 the same way</a> </li>
132 <li><a href="#iterate_chains">Iterating over def-use &
133 use-def chains</a> </li>
134 <li><a href="#iterate_preds">Iterating over predecessors &
135 successors of blocks</a></li>
138 <li><a href="#simplechanges">Making simple changes</a>
140 <li><a href="#schanges_creating">Creating and inserting new
141 <tt>Instruction</tt>s</a> </li>
142 <li><a href="#schanges_deleting">Deleting <tt>Instruction</tt>s</a> </li>
143 <li><a href="#schanges_replacing">Replacing an <tt>Instruction</tt>
144 with another <tt>Value</tt></a> </li>
145 <li><a href="#schanges_deletingGV">Deleting <tt>GlobalVariable</tt>s</a> </li>
148 <li><a href="#create_types">How to Create Types</a></li>
150 <li>Working with the Control Flow Graph
152 <li>Accessing predecessors and successors of a <tt>BasicBlock</tt>
160 <li><a href="#threading">Threads and LLVM</a>
162 <li><a href="#startmultithreaded">Entering and Exiting Multithreaded Mode
164 <li><a href="#shutdown">Ending execution with <tt>llvm_shutdown()</tt></a></li>
165 <li><a href="#managedstatic">Lazy initialization with <tt>ManagedStatic</tt></a></li>
166 <li><a href="#llvmcontext">Achieving Isolation with <tt>LLVMContext</tt></a></li>
167 <li><a href="#jitthreading">Threads and the JIT</a></li>
171 <li><a href="#advanced">Advanced Topics</a>
174 <li><a href="#SymbolTable">The <tt>ValueSymbolTable</tt> class</a></li>
175 <li><a href="#UserLayout">The <tt>User</tt> and owned <tt>Use</tt> classes' memory layout</a></li>
178 <li><a href="#coreclasses">The Core LLVM Class Hierarchy Reference</a>
180 <li><a href="#Type">The <tt>Type</tt> class</a> </li>
181 <li><a href="#Module">The <tt>Module</tt> class</a></li>
182 <li><a href="#Value">The <tt>Value</tt> class</a>
184 <li><a href="#User">The <tt>User</tt> class</a>
186 <li><a href="#Instruction">The <tt>Instruction</tt> class</a></li>
187 <li><a href="#Constant">The <tt>Constant</tt> class</a>
189 <li><a href="#GlobalValue">The <tt>GlobalValue</tt> class</a>
191 <li><a href="#Function">The <tt>Function</tt> class</a></li>
192 <li><a href="#GlobalVariable">The <tt>GlobalVariable</tt> class</a></li>
199 <li><a href="#BasicBlock">The <tt>BasicBlock</tt> class</a></li>
200 <li><a href="#Argument">The <tt>Argument</tt> class</a></li>
207 <div class="doc_author">
208 <p>Written by <a href="mailto:sabre@nondot.org">Chris Lattner</a>,
209 <a href="mailto:dhurjati@cs.uiuc.edu">Dinakar Dhurjati</a>,
210 <a href="mailto:ggreif@gmail.com">Gabor Greif</a>,
211 <a href="mailto:jstanley@cs.uiuc.edu">Joel Stanley</a>,
212 <a href="mailto:rspencer@x10sys.com">Reid Spencer</a> and
213 <a href="mailto:owen@apple.com">Owen Anderson</a></p>
216 <!-- *********************************************************************** -->
218 <a name="introduction">Introduction </a>
220 <!-- *********************************************************************** -->
224 <p>This document is meant to highlight some of the important classes and
225 interfaces available in the LLVM source-base. This manual is not
226 intended to explain what LLVM is, how it works, and what LLVM code looks
227 like. It assumes that you know the basics of LLVM and are interested
228 in writing transformations or otherwise analyzing or manipulating the
231 <p>This document should get you oriented so that you can find your
232 way in the continuously growing source code that makes up the LLVM
233 infrastructure. Note that this manual is not intended to serve as a
234 replacement for reading the source code, so if you think there should be
235 a method in one of these classes to do something, but it's not listed,
236 check the source. Links to the <a href="/doxygen/">doxygen</a> sources
237 are provided to make this as easy as possible.</p>
239 <p>The first section of this document describes general information that is
240 useful to know when working in the LLVM infrastructure, and the second describes
241 the Core LLVM classes. In the future this manual will be extended with
242 information describing how to use extension libraries, such as dominator
243 information, CFG traversal routines, and useful utilities like the <tt><a
244 href="/doxygen/InstVisitor_8h-source.html">InstVisitor</a></tt> template.</p>
248 <!-- *********************************************************************** -->
250 <a name="general">General Information</a>
252 <!-- *********************************************************************** -->
256 <p>This section contains general information that is useful if you are working
257 in the LLVM source-base, but that isn't specific to any particular API.</p>
259 <!-- ======================================================================= -->
261 <a name="stl">The C++ Standard Template Library</a>
266 <p>LLVM makes heavy use of the C++ Standard Template Library (STL),
267 perhaps much more than you are used to, or have seen before. Because of
268 this, you might want to do a little background reading in the
269 techniques used and capabilities of the library. There are many good
270 pages that discuss the STL, and several books on the subject that you
271 can get, so it will not be discussed in this document.</p>
273 <p>Here are some useful links:</p>
277 <li><a href="http://www.dinkumware.com/manuals/#Standard C++ Library">Dinkumware
278 C++ Library reference</a> - an excellent reference for the STL and other parts
279 of the standard C++ library.</li>
281 <li><a href="http://www.tempest-sw.com/cpp/">C++ In a Nutshell</a> - This is an
282 O'Reilly book in the making. It has a decent Standard Library
283 Reference that rivals Dinkumware's, and is unfortunately no longer free since the
284 book has been published.</li>
286 <li><a href="http://www.parashift.com/c++-faq-lite/">C++ Frequently Asked
289 <li><a href="http://www.sgi.com/tech/stl/">SGI's STL Programmer's Guide</a> -
291 href="http://www.sgi.com/tech/stl/stl_introduction.html">Introduction to the
294 <li><a href="http://www.research.att.com/%7Ebs/C++.html">Bjarne Stroustrup's C++
297 <li><a href="http://64.78.49.204/">
298 Bruce Eckel's Thinking in C++, 2nd ed. Volume 2 Revision 4.0 (even better, get
303 <p>You are also encouraged to take a look at the <a
304 href="CodingStandards.html">LLVM Coding Standards</a> guide which focuses on how
305 to write maintainable code more than where to put your curly braces.</p>
309 <!-- ======================================================================= -->
311 <a name="stl">Other useful references</a>
317 <li><a href="http://www.fortran-2000.com/ArnaudRecipes/sharedlib.html">Using
318 static and shared libraries across platforms</a></li>
325 <!-- *********************************************************************** -->
327 <a name="apis">Important and useful LLVM APIs</a>
329 <!-- *********************************************************************** -->
333 <p>Here we highlight some LLVM APIs that are generally useful and good to
334 know about when writing transformations.</p>
336 <!-- ======================================================================= -->
338 <a name="isa">The <tt>isa<></tt>, <tt>cast<></tt> and
339 <tt>dyn_cast<></tt> templates</a>
344 <p>The LLVM source-base makes extensive use of a custom form of RTTI.
345 These templates have many similarities to the C++ <tt>dynamic_cast<></tt>
346 operator, but they don't have some drawbacks (primarily stemming from
347 the fact that <tt>dynamic_cast<></tt> only works on classes that
348 have a v-table). Because they are used so often, you must know what they
349 do and how they work. All of these templates are defined in the <a
350 href="/doxygen/Casting_8h-source.html"><tt>llvm/Support/Casting.h</tt></a>
351 file (note that you very rarely have to include this file directly).</p>
354 <dt><tt>isa<></tt>: </dt>
356 <dd><p>The <tt>isa<></tt> operator works exactly like the Java
357 "<tt>instanceof</tt>" operator. It returns true or false depending on whether
358 a reference or pointer points to an instance of the specified class. This can
359 be very useful for constraint checking of various sorts (example below).</p>
362 <dt><tt>cast<></tt>: </dt>
364 <dd><p>The <tt>cast<></tt> operator is a "checked cast" operation. It
365 converts a pointer or reference from a base class to a derived class, causing
366 an assertion failure if it is not really an instance of the right type. This
367 should be used in cases where you have some information that makes you believe
368 that something is of the right type. An example of the <tt>isa<></tt>
369 and <tt>cast<></tt> template is:</p>
371 <div class="doc_code">
373 static bool isLoopInvariant(const <a href="#Value">Value</a> *V, const Loop *L) {
374 if (isa<<a href="#Constant">Constant</a>>(V) || isa<<a href="#Argument">Argument</a>>(V) || isa<<a href="#GlobalValue">GlobalValue</a>>(V))
377 // <i>Otherwise, it must be an instruction...</i>
378 return !L->contains(cast<<a href="#Instruction">Instruction</a>>(V)->getParent());
383 <p>Note that you should <b>not</b> use an <tt>isa<></tt> test followed
384 by a <tt>cast<></tt>, for that use the <tt>dyn_cast<></tt>
389 <dt><tt>dyn_cast<></tt>:</dt>
391 <dd><p>The <tt>dyn_cast<></tt> operator is a "checking cast" operation.
392 It checks to see if the operand is of the specified type, and if so, returns a
393 pointer to it (this operator does not work with references). If the operand is
394 not of the correct type, a null pointer is returned. Thus, this works very
395 much like the <tt>dynamic_cast<></tt> operator in C++, and should be
396 used in the same circumstances. Typically, the <tt>dyn_cast<></tt>
397 operator is used in an <tt>if</tt> statement or some other flow control
398 statement like this:</p>
400 <div class="doc_code">
402 if (<a href="#AllocationInst">AllocationInst</a> *AI = dyn_cast<<a href="#AllocationInst">AllocationInst</a>>(Val)) {
408 <p>This form of the <tt>if</tt> statement effectively combines together a call
409 to <tt>isa<></tt> and a call to <tt>cast<></tt> into one
410 statement, which is very convenient.</p>
412 <p>Note that the <tt>dyn_cast<></tt> operator, like C++'s
413 <tt>dynamic_cast<></tt> or Java's <tt>instanceof</tt> operator, can be
414 abused. In particular, you should not use big chained <tt>if/then/else</tt>
415 blocks to check for lots of different variants of classes. If you find
416 yourself wanting to do this, it is much cleaner and more efficient to use the
417 <tt>InstVisitor</tt> class to dispatch over the instruction type directly.</p>
421 <dt><tt>cast_or_null<></tt>: </dt>
423 <dd><p>The <tt>cast_or_null<></tt> operator works just like the
424 <tt>cast<></tt> operator, except that it allows for a null pointer as an
425 argument (which it then propagates). This can sometimes be useful, allowing
426 you to combine several null checks into one.</p></dd>
428 <dt><tt>dyn_cast_or_null<></tt>: </dt>
430 <dd><p>The <tt>dyn_cast_or_null<></tt> operator works just like the
431 <tt>dyn_cast<></tt> operator, except that it allows for a null pointer
432 as an argument (which it then propagates). This can sometimes be useful,
433 allowing you to combine several null checks into one.</p></dd>
437 <p>These five templates can be used with any classes, whether they have a
438 v-table or not. To add support for these templates, you simply need to add
439 <tt>classof</tt> static methods to the class you are interested casting
440 to. Describing this is currently outside the scope of this document, but there
441 are lots of examples in the LLVM source base.</p>
446 <!-- ======================================================================= -->
448 <a name="string_apis">Passing strings (the <tt>StringRef</tt>
449 and <tt>Twine</tt> classes)</a>
454 <p>Although LLVM generally does not do much string manipulation, we do have
455 several important APIs which take strings. Two important examples are the
456 Value class -- which has names for instructions, functions, etc. -- and the
457 StringMap class which is used extensively in LLVM and Clang.</p>
459 <p>These are generic classes, and they need to be able to accept strings which
460 may have embedded null characters. Therefore, they cannot simply take
461 a <tt>const char *</tt>, and taking a <tt>const std::string&</tt> requires
462 clients to perform a heap allocation which is usually unnecessary. Instead,
463 many LLVM APIs use a <tt>StringRef</tt> or a <tt>const Twine&</tt> for
464 passing strings efficiently.</p>
466 <!-- _______________________________________________________________________ -->
468 <a name="StringRef">The <tt>StringRef</tt> class</a>
473 <p>The <tt>StringRef</tt> data type represents a reference to a constant string
474 (a character array and a length) and supports the common operations available
475 on <tt>std:string</tt>, but does not require heap allocation.</p>
477 <p>It can be implicitly constructed using a C style null-terminated string,
478 an <tt>std::string</tt>, or explicitly with a character pointer and length.
479 For example, the <tt>StringRef</tt> find function is declared as:</p>
481 <pre class="doc_code">
482 iterator find(StringRef Key);
485 <p>and clients can call it using any one of:</p>
487 <pre class="doc_code">
488 Map.find("foo"); <i>// Lookup "foo"</i>
489 Map.find(std::string("bar")); <i>// Lookup "bar"</i>
490 Map.find(StringRef("\0baz", 4)); <i>// Lookup "\0baz"</i>
493 <p>Similarly, APIs which need to return a string may return a <tt>StringRef</tt>
494 instance, which can be used directly or converted to an <tt>std::string</tt>
495 using the <tt>str</tt> member function. See
496 "<tt><a href="/doxygen/classllvm_1_1StringRef_8h-source.html">llvm/ADT/StringRef.h</a></tt>"
497 for more information.</p>
499 <p>You should rarely use the <tt>StringRef</tt> class directly, because it contains
500 pointers to external memory it is not generally safe to store an instance of the
501 class (unless you know that the external storage will not be freed). StringRef is
502 small and pervasive enough in LLVM that it should always be passed by value.</p>
506 <!-- _______________________________________________________________________ -->
508 <a name="Twine">The <tt>Twine</tt> class</a>
513 <p>The <tt>Twine</tt> class is an efficient way for APIs to accept concatenated
514 strings. For example, a common LLVM paradigm is to name one instruction based on
515 the name of another instruction with a suffix, for example:</p>
517 <div class="doc_code">
519 New = CmpInst::Create(<i>...</i>, SO->getName() + ".cmp");
523 <p>The <tt>Twine</tt> class is effectively a
524 lightweight <a href="http://en.wikipedia.org/wiki/Rope_(computer_science)">rope</a>
525 which points to temporary (stack allocated) objects. Twines can be implicitly
526 constructed as the result of the plus operator applied to strings (i.e., a C
527 strings, an <tt>std::string</tt>, or a <tt>StringRef</tt>). The twine delays the
528 actual concatenation of strings until it is actually required, at which point
529 it can be efficiently rendered directly into a character array. This avoids
530 unnecessary heap allocation involved in constructing the temporary results of
531 string concatenation. See
532 "<tt><a href="/doxygen/classllvm_1_1Twine_8h-source.html">llvm/ADT/Twine.h</a></tt>"
533 for more information.</p>
535 <p>As with a <tt>StringRef</tt>, <tt>Twine</tt> objects point to external memory
536 and should almost never be stored or mentioned directly. They are intended
537 solely for use when defining a function which should be able to efficiently
538 accept concatenated strings.</p>
544 <!-- ======================================================================= -->
546 <a name="DEBUG">The <tt>DEBUG()</tt> macro and <tt>-debug</tt> option</a>
551 <p>Often when working on your pass you will put a bunch of debugging printouts
552 and other code into your pass. After you get it working, you want to remove
553 it, but you may need it again in the future (to work out new bugs that you run
556 <p> Naturally, because of this, you don't want to delete the debug printouts,
557 but you don't want them to always be noisy. A standard compromise is to comment
558 them out, allowing you to enable them if you need them in the future.</p>
560 <p>The "<tt><a href="/doxygen/Debug_8h-source.html">llvm/Support/Debug.h</a></tt>"
561 file provides a macro named <tt>DEBUG()</tt> that is a much nicer solution to
562 this problem. Basically, you can put arbitrary code into the argument of the
563 <tt>DEBUG</tt> macro, and it is only executed if '<tt>opt</tt>' (or any other
564 tool) is run with the '<tt>-debug</tt>' command line argument:</p>
566 <div class="doc_code">
568 DEBUG(errs() << "I am here!\n");
572 <p>Then you can run your pass like this:</p>
574 <div class="doc_code">
576 $ opt < a.bc > /dev/null -mypass
577 <i><no output></i>
578 $ opt < a.bc > /dev/null -mypass -debug
583 <p>Using the <tt>DEBUG()</tt> macro instead of a home-brewed solution allows you
584 to not have to create "yet another" command line option for the debug output for
585 your pass. Note that <tt>DEBUG()</tt> macros are disabled for optimized builds,
586 so they do not cause a performance impact at all (for the same reason, they
587 should also not contain side-effects!).</p>
589 <p>One additional nice thing about the <tt>DEBUG()</tt> macro is that you can
590 enable or disable it directly in gdb. Just use "<tt>set DebugFlag=0</tt>" or
591 "<tt>set DebugFlag=1</tt>" from the gdb if the program is running. If the
592 program hasn't been started yet, you can always just run it with
595 <!-- _______________________________________________________________________ -->
597 <a name="DEBUG_TYPE">Fine grained debug info with <tt>DEBUG_TYPE</tt> and
598 the <tt>-debug-only</tt> option</a>
603 <p>Sometimes you may find yourself in a situation where enabling <tt>-debug</tt>
604 just turns on <b>too much</b> information (such as when working on the code
605 generator). If you want to enable debug information with more fine-grained
606 control, you define the <tt>DEBUG_TYPE</tt> macro and the <tt>-debug</tt> only
607 option as follows:</p>
609 <div class="doc_code">
612 DEBUG(errs() << "No debug type\n");
613 #define DEBUG_TYPE "foo"
614 DEBUG(errs() << "'foo' debug type\n");
616 #define DEBUG_TYPE "bar"
617 DEBUG(errs() << "'bar' debug type\n"));
619 #define DEBUG_TYPE ""
620 DEBUG(errs() << "No debug type (2)\n");
624 <p>Then you can run your pass like this:</p>
626 <div class="doc_code">
628 $ opt < a.bc > /dev/null -mypass
629 <i><no output></i>
630 $ opt < a.bc > /dev/null -mypass -debug
635 $ opt < a.bc > /dev/null -mypass -debug-only=foo
637 $ opt < a.bc > /dev/null -mypass -debug-only=bar
642 <p>Of course, in practice, you should only set <tt>DEBUG_TYPE</tt> at the top of
643 a file, to specify the debug type for the entire module (if you do this before
644 you <tt>#include "llvm/Support/Debug.h"</tt>, you don't have to insert the ugly
645 <tt>#undef</tt>'s). Also, you should use names more meaningful than "foo" and
646 "bar", because there is no system in place to ensure that names do not
647 conflict. If two different modules use the same string, they will all be turned
648 on when the name is specified. This allows, for example, all debug information
649 for instruction scheduling to be enabled with <tt>-debug-type=InstrSched</tt>,
650 even if the source lives in multiple files.</p>
652 <p>The <tt>DEBUG_WITH_TYPE</tt> macro is also available for situations where you
653 would like to set <tt>DEBUG_TYPE</tt>, but only for one specific <tt>DEBUG</tt>
654 statement. It takes an additional first parameter, which is the type to use. For
655 example, the preceding example could be written as:</p>
658 <div class="doc_code">
660 DEBUG_WITH_TYPE("", errs() << "No debug type\n");
661 DEBUG_WITH_TYPE("foo", errs() << "'foo' debug type\n");
662 DEBUG_WITH_TYPE("bar", errs() << "'bar' debug type\n"));
663 DEBUG_WITH_TYPE("", errs() << "No debug type (2)\n");
671 <!-- ======================================================================= -->
673 <a name="Statistic">The <tt>Statistic</tt> class & <tt>-stats</tt>
680 href="/doxygen/Statistic_8h-source.html">llvm/ADT/Statistic.h</a></tt>" file
681 provides a class named <tt>Statistic</tt> that is used as a unified way to
682 keep track of what the LLVM compiler is doing and how effective various
683 optimizations are. It is useful to see what optimizations are contributing to
684 making a particular program run faster.</p>
686 <p>Often you may run your pass on some big program, and you're interested to see
687 how many times it makes a certain transformation. Although you can do this with
688 hand inspection, or some ad-hoc method, this is a real pain and not very useful
689 for big programs. Using the <tt>Statistic</tt> class makes it very easy to
690 keep track of this information, and the calculated information is presented in a
691 uniform manner with the rest of the passes being executed.</p>
693 <p>There are many examples of <tt>Statistic</tt> uses, but the basics of using
694 it are as follows:</p>
697 <li><p>Define your statistic like this:</p>
699 <div class="doc_code">
701 #define <a href="#DEBUG_TYPE">DEBUG_TYPE</a> "mypassname" <i>// This goes before any #includes.</i>
702 STATISTIC(NumXForms, "The # of times I did stuff");
706 <p>The <tt>STATISTIC</tt> macro defines a static variable, whose name is
707 specified by the first argument. The pass name is taken from the DEBUG_TYPE
708 macro, and the description is taken from the second argument. The variable
709 defined ("NumXForms" in this case) acts like an unsigned integer.</p></li>
711 <li><p>Whenever you make a transformation, bump the counter:</p>
713 <div class="doc_code">
715 ++NumXForms; // <i>I did stuff!</i>
722 <p>That's all you have to do. To get '<tt>opt</tt>' to print out the
723 statistics gathered, use the '<tt>-stats</tt>' option:</p>
725 <div class="doc_code">
727 $ opt -stats -mypassname < program.bc > /dev/null
728 <i>... statistics output ...</i>
732 <p> When running <tt>opt</tt> on a C file from the SPEC benchmark
733 suite, it gives a report that looks like this:</p>
735 <div class="doc_code">
737 7646 bitcodewriter - Number of normal instructions
738 725 bitcodewriter - Number of oversized instructions
739 129996 bitcodewriter - Number of bitcode bytes written
740 2817 raise - Number of insts DCEd or constprop'd
741 3213 raise - Number of cast-of-self removed
742 5046 raise - Number of expression trees converted
743 75 raise - Number of other getelementptr's formed
744 138 raise - Number of load/store peepholes
745 42 deadtypeelim - Number of unused typenames removed from symtab
746 392 funcresolve - Number of varargs functions resolved
747 27 globaldce - Number of global variables removed
748 2 adce - Number of basic blocks removed
749 134 cee - Number of branches revectored
750 49 cee - Number of setcc instruction eliminated
751 532 gcse - Number of loads removed
752 2919 gcse - Number of instructions removed
753 86 indvars - Number of canonical indvars added
754 87 indvars - Number of aux indvars removed
755 25 instcombine - Number of dead inst eliminate
756 434 instcombine - Number of insts combined
757 248 licm - Number of load insts hoisted
758 1298 licm - Number of insts hoisted to a loop pre-header
759 3 licm - Number of insts hoisted to multiple loop preds (bad, no loop pre-header)
760 75 mem2reg - Number of alloca's promoted
761 1444 cfgsimplify - Number of blocks simplified
765 <p>Obviously, with so many optimizations, having a unified framework for this
766 stuff is very nice. Making your pass fit well into the framework makes it more
767 maintainable and useful.</p>
771 <!-- ======================================================================= -->
773 <a name="ViewGraph">Viewing graphs while debugging code</a>
778 <p>Several of the important data structures in LLVM are graphs: for example
779 CFGs made out of LLVM <a href="#BasicBlock">BasicBlock</a>s, CFGs made out of
780 LLVM <a href="CodeGenerator.html#machinebasicblock">MachineBasicBlock</a>s, and
781 <a href="CodeGenerator.html#selectiondag_intro">Instruction Selection
782 DAGs</a>. In many cases, while debugging various parts of the compiler, it is
783 nice to instantly visualize these graphs.</p>
785 <p>LLVM provides several callbacks that are available in a debug build to do
786 exactly that. If you call the <tt>Function::viewCFG()</tt> method, for example,
787 the current LLVM tool will pop up a window containing the CFG for the function
788 where each basic block is a node in the graph, and each node contains the
789 instructions in the block. Similarly, there also exists
790 <tt>Function::viewCFGOnly()</tt> (does not include the instructions), the
791 <tt>MachineFunction::viewCFG()</tt> and <tt>MachineFunction::viewCFGOnly()</tt>,
792 and the <tt>SelectionDAG::viewGraph()</tt> methods. Within GDB, for example,
793 you can usually use something like <tt>call DAG.viewGraph()</tt> to pop
794 up a window. Alternatively, you can sprinkle calls to these functions in your
795 code in places you want to debug.</p>
797 <p>Getting this to work requires a small amount of configuration. On Unix
798 systems with X11, install the <a href="http://www.graphviz.org">graphviz</a>
799 toolkit, and make sure 'dot' and 'gv' are in your path. If you are running on
800 Mac OS/X, download and install the Mac OS/X <a
801 href="http://www.pixelglow.com/graphviz/">Graphviz program</a>, and add
802 <tt>/Applications/Graphviz.app/Contents/MacOS/</tt> (or wherever you install
803 it) to your path. Once in your system and path are set up, rerun the LLVM
804 configure script and rebuild LLVM to enable this functionality.</p>
806 <p><tt>SelectionDAG</tt> has been extended to make it easier to locate
807 <i>interesting</i> nodes in large complex graphs. From gdb, if you
808 <tt>call DAG.setGraphColor(<i>node</i>, "<i>color</i>")</tt>, then the
809 next <tt>call DAG.viewGraph()</tt> would highlight the node in the
810 specified color (choices of colors can be found at <a
811 href="http://www.graphviz.org/doc/info/colors.html">colors</a>.) More
812 complex node attributes can be provided with <tt>call
813 DAG.setGraphAttrs(<i>node</i>, "<i>attributes</i>")</tt> (choices can be
814 found at <a href="http://www.graphviz.org/doc/info/attrs.html">Graph
815 Attributes</a>.) If you want to restart and clear all the current graph
816 attributes, then you can <tt>call DAG.clearGraphAttrs()</tt>. </p>
818 <p>Note that graph visualization features are compiled out of Release builds
819 to reduce file size. This means that you need a Debug+Asserts or
820 Release+Asserts build to use these features.</p>
826 <!-- *********************************************************************** -->
828 <a name="datastructure">Picking the Right Data Structure for a Task</a>
830 <!-- *********************************************************************** -->
834 <p>LLVM has a plethora of data structures in the <tt>llvm/ADT/</tt> directory,
835 and we commonly use STL data structures. This section describes the trade-offs
836 you should consider when you pick one.</p>
839 The first step is a choose your own adventure: do you want a sequential
840 container, a set-like container, or a map-like container? The most important
841 thing when choosing a container is the algorithmic properties of how you plan to
842 access the container. Based on that, you should use:</p>
845 <li>a <a href="#ds_map">map-like</a> container if you need efficient look-up
846 of an value based on another value. Map-like containers also support
847 efficient queries for containment (whether a key is in the map). Map-like
848 containers generally do not support efficient reverse mapping (values to
849 keys). If you need that, use two maps. Some map-like containers also
850 support efficient iteration through the keys in sorted order. Map-like
851 containers are the most expensive sort, only use them if you need one of
852 these capabilities.</li>
854 <li>a <a href="#ds_set">set-like</a> container if you need to put a bunch of
855 stuff into a container that automatically eliminates duplicates. Some
856 set-like containers support efficient iteration through the elements in
857 sorted order. Set-like containers are more expensive than sequential
861 <li>a <a href="#ds_sequential">sequential</a> container provides
862 the most efficient way to add elements and keeps track of the order they are
863 added to the collection. They permit duplicates and support efficient
864 iteration, but do not support efficient look-up based on a key.
867 <li>a <a href="#ds_string">string</a> container is a specialized sequential
868 container or reference structure that is used for character or byte
871 <li>a <a href="#ds_bit">bit</a> container provides an efficient way to store and
872 perform set operations on sets of numeric id's, while automatically
873 eliminating duplicates. Bit containers require a maximum of 1 bit for each
874 identifier you want to store.
879 Once the proper category of container is determined, you can fine tune the
880 memory use, constant factors, and cache behaviors of access by intelligently
881 picking a member of the category. Note that constant factors and cache behavior
882 can be a big deal. If you have a vector that usually only contains a few
883 elements (but could contain many), for example, it's much better to use
884 <a href="#dss_smallvector">SmallVector</a> than <a href="#dss_vector">vector</a>
885 . Doing so avoids (relatively) expensive malloc/free calls, which dwarf the
886 cost of adding the elements to the container. </p>
888 <!-- ======================================================================= -->
890 <a name="ds_sequential">Sequential Containers (std::vector, std::list, etc)</a>
894 There are a variety of sequential containers available for you, based on your
895 needs. Pick the first in this section that will do what you want.
897 <!-- _______________________________________________________________________ -->
899 <a name="dss_arrayref">llvm/ADT/ArrayRef.h</a>
903 <p>The llvm::ArrayRef class is the preferred class to use in an interface that
904 accepts a sequential list of elements in memory and just reads from them. By
905 taking an ArrayRef, the API can be passed a fixed size array, an std::vector,
906 an llvm::SmallVector and anything else that is contiguous in memory.
912 <!-- _______________________________________________________________________ -->
914 <a name="dss_fixedarrays">Fixed Size Arrays</a>
918 <p>Fixed size arrays are very simple and very fast. They are good if you know
919 exactly how many elements you have, or you have a (low) upper bound on how many
923 <!-- _______________________________________________________________________ -->
925 <a name="dss_heaparrays">Heap Allocated Arrays</a>
929 <p>Heap allocated arrays (new[] + delete[]) are also simple. They are good if
930 the number of elements is variable, if you know how many elements you will need
931 before the array is allocated, and if the array is usually large (if not,
932 consider a <a href="#dss_smallvector">SmallVector</a>). The cost of a heap
933 allocated array is the cost of the new/delete (aka malloc/free). Also note that
934 if you are allocating an array of a type with a constructor, the constructor and
935 destructors will be run for every element in the array (re-sizable vectors only
936 construct those elements actually used).</p>
939 <!-- _______________________________________________________________________ -->
941 <a name="dss_tinyptrvector">"llvm/ADT/TinyPtrVector.h"</a>
946 <p><tt>TinyPtrVector<Type></tt> is a highly specialized collection class
947 that is optimized to avoid allocation in the case when a vector has zero or one
948 elements. It has two major restrictions: 1) it can only hold values of pointer
949 type, and 2) it cannot hold a null pointer.</p>
951 <p>Since this container is highly specialized, it is rarely used.</p>
955 <!-- _______________________________________________________________________ -->
957 <a name="dss_smallvector">"llvm/ADT/SmallVector.h"</a>
961 <p><tt>SmallVector<Type, N></tt> is a simple class that looks and smells
962 just like <tt>vector<Type></tt>:
963 it supports efficient iteration, lays out elements in memory order (so you can
964 do pointer arithmetic between elements), supports efficient push_back/pop_back
965 operations, supports efficient random access to its elements, etc.</p>
967 <p>The advantage of SmallVector is that it allocates space for
968 some number of elements (N) <b>in the object itself</b>. Because of this, if
969 the SmallVector is dynamically smaller than N, no malloc is performed. This can
970 be a big win in cases where the malloc/free call is far more expensive than the
971 code that fiddles around with the elements.</p>
973 <p>This is good for vectors that are "usually small" (e.g. the number of
974 predecessors/successors of a block is usually less than 8). On the other hand,
975 this makes the size of the SmallVector itself large, so you don't want to
976 allocate lots of them (doing so will waste a lot of space). As such,
977 SmallVectors are most useful when on the stack.</p>
979 <p>SmallVector also provides a nice portable and efficient replacement for
984 <!-- _______________________________________________________________________ -->
986 <a name="dss_vector"><vector></a>
991 std::vector is well loved and respected. It is useful when SmallVector isn't:
992 when the size of the vector is often large (thus the small optimization will
993 rarely be a benefit) or if you will be allocating many instances of the vector
994 itself (which would waste space for elements that aren't in the container).
995 vector is also useful when interfacing with code that expects vectors :).
998 <p>One worthwhile note about std::vector: avoid code like this:</p>
1000 <div class="doc_code">
1003 std::vector<foo> V;
1009 <p>Instead, write this as:</p>
1011 <div class="doc_code">
1013 std::vector<foo> V;
1021 <p>Doing so will save (at least) one heap allocation and free per iteration of
1026 <!-- _______________________________________________________________________ -->
1028 <a name="dss_deque"><deque></a>
1032 <p>std::deque is, in some senses, a generalized version of std::vector. Like
1033 std::vector, it provides constant time random access and other similar
1034 properties, but it also provides efficient access to the front of the list. It
1035 does not guarantee continuity of elements within memory.</p>
1037 <p>In exchange for this extra flexibility, std::deque has significantly higher
1038 constant factor costs than std::vector. If possible, use std::vector or
1039 something cheaper.</p>
1042 <!-- _______________________________________________________________________ -->
1044 <a name="dss_list"><list></a>
1048 <p>std::list is an extremely inefficient class that is rarely useful.
1049 It performs a heap allocation for every element inserted into it, thus having an
1050 extremely high constant factor, particularly for small data types. std::list
1051 also only supports bidirectional iteration, not random access iteration.</p>
1053 <p>In exchange for this high cost, std::list supports efficient access to both
1054 ends of the list (like std::deque, but unlike std::vector or SmallVector). In
1055 addition, the iterator invalidation characteristics of std::list are stronger
1056 than that of a vector class: inserting or removing an element into the list does
1057 not invalidate iterator or pointers to other elements in the list.</p>
1060 <!-- _______________________________________________________________________ -->
1062 <a name="dss_ilist">llvm/ADT/ilist.h</a>
1066 <p><tt>ilist<T></tt> implements an 'intrusive' doubly-linked list. It is
1067 intrusive, because it requires the element to store and provide access to the
1068 prev/next pointers for the list.</p>
1070 <p><tt>ilist</tt> has the same drawbacks as <tt>std::list</tt>, and additionally
1071 requires an <tt>ilist_traits</tt> implementation for the element type, but it
1072 provides some novel characteristics. In particular, it can efficiently store
1073 polymorphic objects, the traits class is informed when an element is inserted or
1074 removed from the list, and <tt>ilist</tt>s are guaranteed to support a
1075 constant-time splice operation.</p>
1077 <p>These properties are exactly what we want for things like
1078 <tt>Instruction</tt>s and basic blocks, which is why these are implemented with
1079 <tt>ilist</tt>s.</p>
1081 Related classes of interest are explained in the following subsections:
1083 <li><a href="#dss_ilist_traits">ilist_traits</a></li>
1084 <li><a href="#dss_iplist">iplist</a></li>
1085 <li><a href="#dss_ilist_node">llvm/ADT/ilist_node.h</a></li>
1086 <li><a href="#dss_ilist_sentinel">Sentinels</a></li>
1090 <!-- _______________________________________________________________________ -->
1092 <a name="dss_packedvector">llvm/ADT/PackedVector.h</a>
1097 Useful for storing a vector of values using only a few number of bits for each
1098 value. Apart from the standard operations of a vector-like container, it can
1099 also perform an 'or' set operation.
1104 <div class="doc_code">
1108 FirstCondition = 0x1,
1109 SecondCondition = 0x2,
1114 PackedVector<State, 2> Vec1;
1115 Vec1.push_back(FirstCondition);
1117 PackedVector<State, 2> Vec2;
1118 Vec2.push_back(SecondCondition);
1121 return Vec1[0]; // returns 'Both'.
1128 <!-- _______________________________________________________________________ -->
1130 <a name="dss_ilist_traits">ilist_traits</a>
1134 <p><tt>ilist_traits<T></tt> is <tt>ilist<T></tt>'s customization
1135 mechanism. <tt>iplist<T></tt> (and consequently <tt>ilist<T></tt>)
1136 publicly derive from this traits class.</p>
1139 <!-- _______________________________________________________________________ -->
1141 <a name="dss_iplist">iplist</a>
1145 <p><tt>iplist<T></tt> is <tt>ilist<T></tt>'s base and as such
1146 supports a slightly narrower interface. Notably, inserters from
1147 <tt>T&</tt> are absent.</p>
1149 <p><tt>ilist_traits<T></tt> is a public base of this class and can be
1150 used for a wide variety of customizations.</p>
1153 <!-- _______________________________________________________________________ -->
1155 <a name="dss_ilist_node">llvm/ADT/ilist_node.h</a>
1159 <p><tt>ilist_node<T></tt> implements a the forward and backward links
1160 that are expected by the <tt>ilist<T></tt> (and analogous containers)
1161 in the default manner.</p>
1163 <p><tt>ilist_node<T></tt>s are meant to be embedded in the node type
1164 <tt>T</tt>, usually <tt>T</tt> publicly derives from
1165 <tt>ilist_node<T></tt>.</p>
1168 <!-- _______________________________________________________________________ -->
1170 <a name="dss_ilist_sentinel">Sentinels</a>
1174 <p><tt>ilist</tt>s have another specialty that must be considered. To be a good
1175 citizen in the C++ ecosystem, it needs to support the standard container
1176 operations, such as <tt>begin</tt> and <tt>end</tt> iterators, etc. Also, the
1177 <tt>operator--</tt> must work correctly on the <tt>end</tt> iterator in the
1178 case of non-empty <tt>ilist</tt>s.</p>
1180 <p>The only sensible solution to this problem is to allocate a so-called
1181 <i>sentinel</i> along with the intrusive list, which serves as the <tt>end</tt>
1182 iterator, providing the back-link to the last element. However conforming to the
1183 C++ convention it is illegal to <tt>operator++</tt> beyond the sentinel and it
1184 also must not be dereferenced.</p>
1186 <p>These constraints allow for some implementation freedom to the <tt>ilist</tt>
1187 how to allocate and store the sentinel. The corresponding policy is dictated
1188 by <tt>ilist_traits<T></tt>. By default a <tt>T</tt> gets heap-allocated
1189 whenever the need for a sentinel arises.</p>
1191 <p>While the default policy is sufficient in most cases, it may break down when
1192 <tt>T</tt> does not provide a default constructor. Also, in the case of many
1193 instances of <tt>ilist</tt>s, the memory overhead of the associated sentinels
1194 is wasted. To alleviate the situation with numerous and voluminous
1195 <tt>T</tt>-sentinels, sometimes a trick is employed, leading to <i>ghostly
1198 <p>Ghostly sentinels are obtained by specially-crafted <tt>ilist_traits<T></tt>
1199 which superpose the sentinel with the <tt>ilist</tt> instance in memory. Pointer
1200 arithmetic is used to obtain the sentinel, which is relative to the
1201 <tt>ilist</tt>'s <tt>this</tt> pointer. The <tt>ilist</tt> is augmented by an
1202 extra pointer, which serves as the back-link of the sentinel. This is the only
1203 field in the ghostly sentinel which can be legally accessed.</p>
1206 <!-- _______________________________________________________________________ -->
1208 <a name="dss_other">Other Sequential Container options</a>
1212 <p>Other STL containers are available, such as std::string.</p>
1214 <p>There are also various STL adapter classes such as std::queue,
1215 std::priority_queue, std::stack, etc. These provide simplified access to an
1216 underlying container but don't affect the cost of the container itself.</p>
1221 <!-- ======================================================================= -->
1223 <a name="ds_string">String-like containers</a>
1229 There are a variety of ways to pass around and use strings in C and C++, and
1230 LLVM adds a few new options to choose from. Pick the first option on this list
1231 that will do what you need, they are ordered according to their relative cost.
1234 Note that is is generally preferred to <em>not</em> pass strings around as
1235 "<tt>const char*</tt>"'s. These have a number of problems, including the fact
1236 that they cannot represent embedded nul ("\0") characters, and do not have a
1237 length available efficiently. The general replacement for '<tt>const
1238 char*</tt>' is StringRef.
1241 <p>For more information on choosing string containers for APIs, please see
1242 <a href="#string_apis">Passing strings</a>.</p>
1245 <!-- _______________________________________________________________________ -->
1247 <a name="dss_stringref">llvm/ADT/StringRef.h</a>
1252 The StringRef class is a simple value class that contains a pointer to a
1253 character and a length, and is quite related to the <a
1254 href="#dss_arrayref">ArrayRef</a> class (but specialized for arrays of
1255 characters). Because StringRef carries a length with it, it safely handles
1256 strings with embedded nul characters in it, getting the length does not require
1257 a strlen call, and it even has very convenient APIs for slicing and dicing the
1258 character range that it represents.
1262 StringRef is ideal for passing simple strings around that are known to be live,
1263 either because they are C string literals, std::string, a C array, or a
1264 SmallVector. Each of these cases has an efficient implicit conversion to
1265 StringRef, which doesn't result in a dynamic strlen being executed.
1268 <p>StringRef has a few major limitations which make more powerful string
1269 containers useful:</p>
1272 <li>You cannot directly convert a StringRef to a 'const char*' because there is
1273 no way to add a trailing nul (unlike the .c_str() method on various stronger
1277 <li>StringRef doesn't own or keep alive the underlying string bytes.
1278 As such it can easily lead to dangling pointers, and is not suitable for
1279 embedding in datastructures in most cases (instead, use an std::string or
1280 something like that).</li>
1282 <li>For the same reason, StringRef cannot be used as the return value of a
1283 method if the method "computes" the result string. Instead, use
1286 <li>StringRef's do not allow you to mutate the pointed-to string bytes and it
1287 doesn't allow you to insert or remove bytes from the range. For editing
1288 operations like this, it interoperates with the <a
1289 href="#dss_twine">Twine</a> class.</li>
1292 <p>Because of its strengths and limitations, it is very common for a function to
1293 take a StringRef and for a method on an object to return a StringRef that
1294 points into some string that it owns.</p>
1298 <!-- _______________________________________________________________________ -->
1300 <a name="dss_twine">llvm/ADT/Twine.h</a>
1305 The Twine class is used as an intermediary datatype for APIs that want to take
1306 a string that can be constructed inline with a series of concatenations.
1307 Twine works by forming recursive instances of the Twine datatype (a simple
1308 value object) on the stack as temporary objects, linking them together into a
1309 tree which is then linearized when the Twine is consumed. Twine is only safe
1310 to use as the argument to a function, and should always be a const reference,
1315 void foo(const Twine &T);
1319 foo(X + "." + Twine(i));
1322 <p>This example forms a string like "blarg.42" by concatenating the values
1323 together, and does not form intermediate strings containing "blarg" or
1327 <p>Because Twine is constructed with temporary objects on the stack, and
1328 because these instances are destroyed at the end of the current statement,
1329 it is an inherently dangerous API. For example, this simple variant contains
1330 undefined behavior and will probably crash:</p>
1333 void foo(const Twine &T);
1337 const Twine &Tmp = X + "." + Twine(i);
1341 <p>... because the temporaries are destroyed before the call. That said,
1342 Twine's are much more efficient than intermediate std::string temporaries, and
1343 they work really well with StringRef. Just be aware of their limitations.</p>
1348 <!-- _______________________________________________________________________ -->
1350 <a name="dss_smallstring">llvm/ADT/SmallString.h</a>
1355 <p>SmallString is a subclass of <a href="#dss_smallvector">SmallVector</a> that
1356 adds some convenience APIs like += that takes StringRef's. SmallString avoids
1357 allocating memory in the case when the preallocated space is enough to hold its
1358 data, and it calls back to general heap allocation when required. Since it owns
1359 its data, it is very safe to use and supports full mutation of the string.</p>
1361 <p>Like SmallVector's, the big downside to SmallString is their sizeof. While
1362 they are optimized for small strings, they themselves are not particularly
1363 small. This means that they work great for temporary scratch buffers on the
1364 stack, but should not generally be put into the heap: it is very rare to
1365 see a SmallString as the member of a frequently-allocated heap data structure
1366 or returned by-value.
1371 <!-- _______________________________________________________________________ -->
1373 <a name="dss_stdstring">std::string</a>
1378 <p>The standard C++ std::string class is a very general class that (like
1379 SmallString) owns its underlying data. sizeof(std::string) is very reasonable
1380 so it can be embedded into heap data structures and returned by-value.
1381 On the other hand, std::string is highly inefficient for inline editing (e.g.
1382 concatenating a bunch of stuff together) and because it is provided by the
1383 standard library, its performance characteristics depend a lot of the host
1384 standard library (e.g. libc++ and MSVC provide a highly optimized string
1385 class, GCC contains a really slow implementation).
1388 <p>The major disadvantage of std::string is that almost every operation that
1389 makes them larger can allocate memory, which is slow. As such, it is better
1390 to use SmallVector or Twine as a scratch buffer, but then use std::string to
1391 persist the result.</p>
1396 <!-- end of strings -->
1400 <!-- ======================================================================= -->
1402 <a name="ds_set">Set-Like Containers (std::set, SmallSet, SetVector, etc)</a>
1407 <p>Set-like containers are useful when you need to canonicalize multiple values
1408 into a single representation. There are several different choices for how to do
1409 this, providing various trade-offs.</p>
1411 <!-- _______________________________________________________________________ -->
1413 <a name="dss_sortedvectorset">A sorted 'vector'</a>
1418 <p>If you intend to insert a lot of elements, then do a lot of queries, a
1419 great approach is to use a vector (or other sequential container) with
1420 std::sort+std::unique to remove duplicates. This approach works really well if
1421 your usage pattern has these two distinct phases (insert then query), and can be
1422 coupled with a good choice of <a href="#ds_sequential">sequential container</a>.
1426 This combination provides the several nice properties: the result data is
1427 contiguous in memory (good for cache locality), has few allocations, is easy to
1428 address (iterators in the final vector are just indices or pointers), and can be
1429 efficiently queried with a standard binary or radix search.</p>
1433 <!-- _______________________________________________________________________ -->
1435 <a name="dss_smallset">"llvm/ADT/SmallSet.h"</a>
1440 <p>If you have a set-like data structure that is usually small and whose elements
1441 are reasonably small, a <tt>SmallSet<Type, N></tt> is a good choice. This set
1442 has space for N elements in place (thus, if the set is dynamically smaller than
1443 N, no malloc traffic is required) and accesses them with a simple linear search.
1444 When the set grows beyond 'N' elements, it allocates a more expensive representation that
1445 guarantees efficient access (for most types, it falls back to std::set, but for
1446 pointers it uses something far better, <a
1447 href="#dss_smallptrset">SmallPtrSet</a>).</p>
1449 <p>The magic of this class is that it handles small sets extremely efficiently,
1450 but gracefully handles extremely large sets without loss of efficiency. The
1451 drawback is that the interface is quite small: it supports insertion, queries
1452 and erasing, but does not support iteration.</p>
1456 <!-- _______________________________________________________________________ -->
1458 <a name="dss_smallptrset">"llvm/ADT/SmallPtrSet.h"</a>
1463 <p>SmallPtrSet has all the advantages of <tt>SmallSet</tt> (and a <tt>SmallSet</tt> of pointers is
1464 transparently implemented with a <tt>SmallPtrSet</tt>), but also supports iterators. If
1465 more than 'N' insertions are performed, a single quadratically
1466 probed hash table is allocated and grows as needed, providing extremely
1467 efficient access (constant time insertion/deleting/queries with low constant
1468 factors) and is very stingy with malloc traffic.</p>
1470 <p>Note that, unlike <tt>std::set</tt>, the iterators of <tt>SmallPtrSet</tt> are invalidated
1471 whenever an insertion occurs. Also, the values visited by the iterators are not
1472 visited in sorted order.</p>
1476 <!-- _______________________________________________________________________ -->
1478 <a name="dss_denseset">"llvm/ADT/DenseSet.h"</a>
1484 DenseSet is a simple quadratically probed hash table. It excels at supporting
1485 small values: it uses a single allocation to hold all of the pairs that
1486 are currently inserted in the set. DenseSet is a great way to unique small
1487 values that are not simple pointers (use <a
1488 href="#dss_smallptrset">SmallPtrSet</a> for pointers). Note that DenseSet has
1489 the same requirements for the value type that <a
1490 href="#dss_densemap">DenseMap</a> has.
1495 <!-- _______________________________________________________________________ -->
1497 <a name="dss_sparseset">"llvm/ADT/SparseSet.h"</a>
1502 <p>SparseSet holds a small number of objects identified by unsigned keys of
1503 moderate size. It uses a lot of memory, but provides operations that are
1504 almost as fast as a vector. Typical keys are physical registers, virtual
1505 registers, or numbered basic blocks.</p>
1507 <p>SparseSet is useful for algorithms that need very fast clear/find/insert/erase
1508 and fast iteration over small sets. It is not intended for building composite
1509 data structures.</p>
1513 <!-- _______________________________________________________________________ -->
1515 <a name="dss_FoldingSet">"llvm/ADT/FoldingSet.h"</a>
1521 FoldingSet is an aggregate class that is really good at uniquing
1522 expensive-to-create or polymorphic objects. It is a combination of a chained
1523 hash table with intrusive links (uniqued objects are required to inherit from
1524 FoldingSetNode) that uses <a href="#dss_smallvector">SmallVector</a> as part of
1527 <p>Consider a case where you want to implement a "getOrCreateFoo" method for
1528 a complex object (for example, a node in the code generator). The client has a
1529 description of *what* it wants to generate (it knows the opcode and all the
1530 operands), but we don't want to 'new' a node, then try inserting it into a set
1531 only to find out it already exists, at which point we would have to delete it
1532 and return the node that already exists.
1535 <p>To support this style of client, FoldingSet perform a query with a
1536 FoldingSetNodeID (which wraps SmallVector) that can be used to describe the
1537 element that we want to query for. The query either returns the element
1538 matching the ID or it returns an opaque ID that indicates where insertion should
1539 take place. Construction of the ID usually does not require heap traffic.</p>
1541 <p>Because FoldingSet uses intrusive links, it can support polymorphic objects
1542 in the set (for example, you can have SDNode instances mixed with LoadSDNodes).
1543 Because the elements are individually allocated, pointers to the elements are
1544 stable: inserting or removing elements does not invalidate any pointers to other
1550 <!-- _______________________________________________________________________ -->
1552 <a name="dss_set"><set></a>
1557 <p><tt>std::set</tt> is a reasonable all-around set class, which is decent at
1558 many things but great at nothing. std::set allocates memory for each element
1559 inserted (thus it is very malloc intensive) and typically stores three pointers
1560 per element in the set (thus adding a large amount of per-element space
1561 overhead). It offers guaranteed log(n) performance, which is not particularly
1562 fast from a complexity standpoint (particularly if the elements of the set are
1563 expensive to compare, like strings), and has extremely high constant factors for
1564 lookup, insertion and removal.</p>
1566 <p>The advantages of std::set are that its iterators are stable (deleting or
1567 inserting an element from the set does not affect iterators or pointers to other
1568 elements) and that iteration over the set is guaranteed to be in sorted order.
1569 If the elements in the set are large, then the relative overhead of the pointers
1570 and malloc traffic is not a big deal, but if the elements of the set are small,
1571 std::set is almost never a good choice.</p>
1575 <!-- _______________________________________________________________________ -->
1577 <a name="dss_setvector">"llvm/ADT/SetVector.h"</a>
1581 <p>LLVM's SetVector<Type> is an adapter class that combines your choice of
1582 a set-like container along with a <a href="#ds_sequential">Sequential
1583 Container</a>. The important property
1584 that this provides is efficient insertion with uniquing (duplicate elements are
1585 ignored) with iteration support. It implements this by inserting elements into
1586 both a set-like container and the sequential container, using the set-like
1587 container for uniquing and the sequential container for iteration.
1590 <p>The difference between SetVector and other sets is that the order of
1591 iteration is guaranteed to match the order of insertion into the SetVector.
1592 This property is really important for things like sets of pointers. Because
1593 pointer values are non-deterministic (e.g. vary across runs of the program on
1594 different machines), iterating over the pointers in the set will
1595 not be in a well-defined order.</p>
1598 The drawback of SetVector is that it requires twice as much space as a normal
1599 set and has the sum of constant factors from the set-like container and the
1600 sequential container that it uses. Use it *only* if you need to iterate over
1601 the elements in a deterministic order. SetVector is also expensive to delete
1602 elements out of (linear time), unless you use it's "pop_back" method, which is
1606 <p><tt>SetVector</tt> is an adapter class that defaults to
1607 using <tt>std::vector</tt> and a size 16 <tt>SmallSet</tt> for the underlying
1608 containers, so it is quite expensive. However,
1609 <tt>"llvm/ADT/SetVector.h"</tt> also provides a <tt>SmallSetVector</tt>
1610 class, which defaults to using a <tt>SmallVector</tt> and <tt>SmallSet</tt>
1611 of a specified size. If you use this, and if your sets are dynamically
1612 smaller than <tt>N</tt>, you will save a lot of heap traffic.</p>
1616 <!-- _______________________________________________________________________ -->
1618 <a name="dss_uniquevector">"llvm/ADT/UniqueVector.h"</a>
1624 UniqueVector is similar to <a href="#dss_setvector">SetVector</a>, but it
1625 retains a unique ID for each element inserted into the set. It internally
1626 contains a map and a vector, and it assigns a unique ID for each value inserted
1629 <p>UniqueVector is very expensive: its cost is the sum of the cost of
1630 maintaining both the map and vector, it has high complexity, high constant
1631 factors, and produces a lot of malloc traffic. It should be avoided.</p>
1635 <!-- _______________________________________________________________________ -->
1637 <a name="dss_immutableset">"llvm/ADT/ImmutableSet.h"</a>
1643 ImmutableSet is an immutable (functional) set implementation based on an AVL
1645 Adding or removing elements is done through a Factory object and results in the
1646 creation of a new ImmutableSet object.
1647 If an ImmutableSet already exists with the given contents, then the existing one
1648 is returned; equality is compared with a FoldingSetNodeID.
1649 The time and space complexity of add or remove operations is logarithmic in the
1650 size of the original set.
1653 There is no method for returning an element of the set, you can only check for
1659 <!-- _______________________________________________________________________ -->
1661 <a name="dss_otherset">Other Set-Like Container Options</a>
1667 The STL provides several other options, such as std::multiset and the various
1668 "hash_set" like containers (whether from C++ TR1 or from the SGI library). We
1669 never use hash_set and unordered_set because they are generally very expensive
1670 (each insertion requires a malloc) and very non-portable.
1673 <p>std::multiset is useful if you're not interested in elimination of
1674 duplicates, but has all the drawbacks of std::set. A sorted vector (where you
1675 don't delete duplicate entries) or some other approach is almost always
1682 <!-- ======================================================================= -->
1684 <a name="ds_map">Map-Like Containers (std::map, DenseMap, etc)</a>
1688 Map-like containers are useful when you want to associate data to a key. As
1689 usual, there are a lot of different ways to do this. :)
1691 <!-- _______________________________________________________________________ -->
1693 <a name="dss_sortedvectormap">A sorted 'vector'</a>
1699 If your usage pattern follows a strict insert-then-query approach, you can
1700 trivially use the same approach as <a href="#dss_sortedvectorset">sorted vectors
1701 for set-like containers</a>. The only difference is that your query function
1702 (which uses std::lower_bound to get efficient log(n) lookup) should only compare
1703 the key, not both the key and value. This yields the same advantages as sorted
1708 <!-- _______________________________________________________________________ -->
1710 <a name="dss_stringmap">"llvm/ADT/StringMap.h"</a>
1716 Strings are commonly used as keys in maps, and they are difficult to support
1717 efficiently: they are variable length, inefficient to hash and compare when
1718 long, expensive to copy, etc. StringMap is a specialized container designed to
1719 cope with these issues. It supports mapping an arbitrary range of bytes to an
1720 arbitrary other object.</p>
1722 <p>The StringMap implementation uses a quadratically-probed hash table, where
1723 the buckets store a pointer to the heap allocated entries (and some other
1724 stuff). The entries in the map must be heap allocated because the strings are
1725 variable length. The string data (key) and the element object (value) are
1726 stored in the same allocation with the string data immediately after the element
1727 object. This container guarantees the "<tt>(char*)(&Value+1)</tt>" points
1728 to the key string for a value.</p>
1730 <p>The StringMap is very fast for several reasons: quadratic probing is very
1731 cache efficient for lookups, the hash value of strings in buckets is not
1732 recomputed when looking up an element, StringMap rarely has to touch the
1733 memory for unrelated objects when looking up a value (even when hash collisions
1734 happen), hash table growth does not recompute the hash values for strings
1735 already in the table, and each pair in the map is store in a single allocation
1736 (the string data is stored in the same allocation as the Value of a pair).</p>
1738 <p>StringMap also provides query methods that take byte ranges, so it only ever
1739 copies a string if a value is inserted into the table.</p>
1741 <p>StringMap iteratation order, however, is not guaranteed to be deterministic,
1742 so any uses which require that should instead use a std::map.</p>
1745 <!-- _______________________________________________________________________ -->
1747 <a name="dss_indexedmap">"llvm/ADT/IndexedMap.h"</a>
1752 IndexedMap is a specialized container for mapping small dense integers (or
1753 values that can be mapped to small dense integers) to some other type. It is
1754 internally implemented as a vector with a mapping function that maps the keys to
1755 the dense integer range.
1759 This is useful for cases like virtual registers in the LLVM code generator: they
1760 have a dense mapping that is offset by a compile-time constant (the first
1761 virtual register ID).</p>
1765 <!-- _______________________________________________________________________ -->
1767 <a name="dss_densemap">"llvm/ADT/DenseMap.h"</a>
1773 DenseMap is a simple quadratically probed hash table. It excels at supporting
1774 small keys and values: it uses a single allocation to hold all of the pairs that
1775 are currently inserted in the map. DenseMap is a great way to map pointers to
1776 pointers, or map other small types to each other.
1780 There are several aspects of DenseMap that you should be aware of, however. The
1781 iterators in a DenseMap are invalidated whenever an insertion occurs, unlike
1782 map. Also, because DenseMap allocates space for a large number of key/value
1783 pairs (it starts with 64 by default), it will waste a lot of space if your keys
1784 or values are large. Finally, you must implement a partial specialization of
1785 DenseMapInfo for the key that you want, if it isn't already supported. This
1786 is required to tell DenseMap about two special marker values (which can never be
1787 inserted into the map) that it needs internally.</p>
1790 DenseMap's find_as() method supports lookup operations using an alternate key
1791 type. This is useful in cases where the normal key type is expensive to
1792 construct, but cheap to compare against. The DenseMapInfo is responsible for
1793 defining the appropriate comparison and hashing methods for each alternate
1799 <!-- _______________________________________________________________________ -->
1801 <a name="dss_valuemap">"llvm/ADT/ValueMap.h"</a>
1807 ValueMap is a wrapper around a <a href="#dss_densemap">DenseMap</a> mapping
1808 Value*s (or subclasses) to another type. When a Value is deleted or RAUW'ed,
1809 ValueMap will update itself so the new version of the key is mapped to the same
1810 value, just as if the key were a WeakVH. You can configure exactly how this
1811 happens, and what else happens on these two events, by passing
1812 a <code>Config</code> parameter to the ValueMap template.</p>
1816 <!-- _______________________________________________________________________ -->
1818 <a name="dss_multiimplmap">"llvm/ADT/MultiImplMap.h"</a>
1824 MultiImplMap is map that has two modes, one for small amount of elements and
1825 one for big amount. User should set map implementation for both of them.
1826 User also should set the maximum possible number of elements for small mode.
1830 If user want to use MultiImplMap instead of
1831 <a href="#dss_densemap">DenseMap</a>, he should pass template parameter
1832 DenseMapCompatible = true. Note, that in this case map implementations
1833 should present additional DenseMap specific methods (see below):
1834 <code>isPointerIntoBucketsArray</code>, <code>getPointerIntoBucketsArray</code>
1835 and <code>FindAndConstruct</code>.
1839 Initially MultiImplMap uses small mode and small map implementation. It
1840 triggered to the big mode when the number of contained elements exceeds
1841 maximum possible elements for small mode.
1846 <!-- _______________________________________________________________________ -->
1848 <a name="dss_flatarraymap">"llvm/ADT/FlatArrayMap.h"</a>
1854 FlatArrayMap optimized for small amount of elements. It uses flat array
1855 implementation inside:
1857 <pre>[ key0, value0, key1, value1, ... keyN, valueN ]</pre>
1861 User should pass key type, mapped type (type of value), and maximum
1866 After maximum number of elements is reached, map declines any further
1867 attempts to insert new elements ("insert" method returns <end(),
1872 FlatArrayMap has interface that is compatible with
1873 <a href="#dss_densemap">DenseMap</a>, so user can replace it with DenseMap
1874 without any code changing and vice versa.
1879 <!-- _______________________________________________________________________ -->
1881 <a name="dss_smallmap">"llvm/ADT/SmallMap.h"</a>
1887 SmallMap is wrapper around <a href="#dss_multiimplmap">MultiImplMap</a>.
1888 It uses <a href="#dss_flatarraymap">FlatArrayMap</a> for small mode, and
1889 <a href="#dss_densemap">DenseMap</a> for big mode.
1894 <!-- _______________________________________________________________________ -->
1896 <a name="dss_intervalmap">"llvm/ADT/IntervalMap.h"</a>
1901 <p> IntervalMap is a compact map for small keys and values. It maps key
1902 intervals instead of single keys, and it will automatically coalesce adjacent
1903 intervals. When then map only contains a few intervals, they are stored in the
1904 map object itself to avoid allocations.</p>
1906 <p> The IntervalMap iterators are quite big, so they should not be passed around
1907 as STL iterators. The heavyweight iterators allow a smaller data structure.</p>
1911 <!-- _______________________________________________________________________ -->
1913 <a name="dss_map"><map></a>
1919 std::map has similar characteristics to <a href="#dss_set">std::set</a>: it uses
1920 a single allocation per pair inserted into the map, it offers log(n) lookup with
1921 an extremely large constant factor, imposes a space penalty of 3 pointers per
1922 pair in the map, etc.</p>
1924 <p>std::map is most useful when your keys or values are very large, if you need
1925 to iterate over the collection in sorted order, or if you need stable iterators
1926 into the map (i.e. they don't get invalidated if an insertion or deletion of
1927 another element takes place).</p>
1931 <!-- _______________________________________________________________________ -->
1933 <a name="dss_inteqclasses">"llvm/ADT/IntEqClasses.h"</a>
1938 <p>IntEqClasses provides a compact representation of equivalence classes of
1939 small integers. Initially, each integer in the range 0..n-1 has its own
1940 equivalence class. Classes can be joined by passing two class representatives to
1941 the join(a, b) method. Two integers are in the same class when findLeader()
1942 returns the same representative.</p>
1944 <p>Once all equivalence classes are formed, the map can be compressed so each
1945 integer 0..n-1 maps to an equivalence class number in the range 0..m-1, where m
1946 is the total number of equivalence classes. The map must be uncompressed before
1947 it can be edited again.</p>
1951 <!-- _______________________________________________________________________ -->
1953 <a name="dss_immutablemap">"llvm/ADT/ImmutableMap.h"</a>
1959 ImmutableMap is an immutable (functional) map implementation based on an AVL
1961 Adding or removing elements is done through a Factory object and results in the
1962 creation of a new ImmutableMap object.
1963 If an ImmutableMap already exists with the given key set, then the existing one
1964 is returned; equality is compared with a FoldingSetNodeID.
1965 The time and space complexity of add or remove operations is logarithmic in the
1966 size of the original map.
1970 <!-- _______________________________________________________________________ -->
1972 <a name="dss_othermap">Other Map-Like Container Options</a>
1978 The STL provides several other options, such as std::multimap and the various
1979 "hash_map" like containers (whether from C++ TR1 or from the SGI library). We
1980 never use hash_set and unordered_set because they are generally very expensive
1981 (each insertion requires a malloc) and very non-portable.</p>
1983 <p>std::multimap is useful if you want to map a key to multiple values, but has
1984 all the drawbacks of std::map. A sorted vector or some other approach is almost
1991 <!-- ======================================================================= -->
1993 <a name="ds_bit">Bit storage containers (BitVector, SparseBitVector)</a>
1997 <p>Unlike the other containers, there are only two bit storage containers, and
1998 choosing when to use each is relatively straightforward.</p>
2000 <p>One additional option is
2001 <tt>std::vector<bool></tt>: we discourage its use for two reasons 1) the
2002 implementation in many common compilers (e.g. commonly available versions of
2003 GCC) is extremely inefficient and 2) the C++ standards committee is likely to
2004 deprecate this container and/or change it significantly somehow. In any case,
2005 please don't use it.</p>
2007 <!-- _______________________________________________________________________ -->
2009 <a name="dss_bitvector">BitVector</a>
2013 <p> The BitVector container provides a dynamic size set of bits for manipulation.
2014 It supports individual bit setting/testing, as well as set operations. The set
2015 operations take time O(size of bitvector), but operations are performed one word
2016 at a time, instead of one bit at a time. This makes the BitVector very fast for
2017 set operations compared to other containers. Use the BitVector when you expect
2018 the number of set bits to be high (IE a dense set).
2022 <!-- _______________________________________________________________________ -->
2024 <a name="dss_smallbitvector">SmallBitVector</a>
2028 <p> The SmallBitVector container provides the same interface as BitVector, but
2029 it is optimized for the case where only a small number of bits, less than
2030 25 or so, are needed. It also transparently supports larger bit counts, but
2031 slightly less efficiently than a plain BitVector, so SmallBitVector should
2032 only be used when larger counts are rare.
2036 At this time, SmallBitVector does not support set operations (and, or, xor),
2037 and its operator[] does not provide an assignable lvalue.
2041 <!-- _______________________________________________________________________ -->
2043 <a name="dss_sparsebitvector">SparseBitVector</a>
2047 <p> The SparseBitVector container is much like BitVector, with one major
2048 difference: Only the bits that are set, are stored. This makes the
2049 SparseBitVector much more space efficient than BitVector when the set is sparse,
2050 as well as making set operations O(number of set bits) instead of O(size of
2051 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
2052 (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).
2060 <!-- *********************************************************************** -->
2062 <a name="common">Helpful Hints for Common Operations</a>
2064 <!-- *********************************************************************** -->
2068 <p>This section describes how to perform some very simple transformations of
2069 LLVM code. This is meant to give examples of common idioms used, showing the
2070 practical side of LLVM transformations. <p> Because this is a "how-to" section,
2071 you should also read about the main classes that you will be working with. The
2072 <a href="#coreclasses">Core LLVM Class Hierarchy Reference</a> contains details
2073 and descriptions of the main classes that you should know about.</p>
2075 <!-- NOTE: this section should be heavy on example code -->
2076 <!-- ======================================================================= -->
2078 <a name="inspection">Basic Inspection and Traversal Routines</a>
2083 <p>The LLVM compiler infrastructure have many different data structures that may
2084 be traversed. Following the example of the C++ standard template library, the
2085 techniques used to traverse these various data structures are all basically the
2086 same. For a enumerable sequence of values, the <tt>XXXbegin()</tt> function (or
2087 method) returns an iterator to the start of the sequence, the <tt>XXXend()</tt>
2088 function returns an iterator pointing to one past the last valid element of the
2089 sequence, and there is some <tt>XXXiterator</tt> data type that is common
2090 between the two operations.</p>
2092 <p>Because the pattern for iteration is common across many different aspects of
2093 the program representation, the standard template library algorithms may be used
2094 on them, and it is easier to remember how to iterate. First we show a few common
2095 examples of the data structures that need to be traversed. Other data
2096 structures are traversed in very similar ways.</p>
2098 <!-- _______________________________________________________________________ -->
2100 <a name="iterate_function">Iterating over the </a><a
2101 href="#BasicBlock"><tt>BasicBlock</tt></a>s in a <a
2102 href="#Function"><tt>Function</tt></a>
2107 <p>It's quite common to have a <tt>Function</tt> instance that you'd like to
2108 transform in some way; in particular, you'd like to manipulate its
2109 <tt>BasicBlock</tt>s. To facilitate this, you'll need to iterate over all of
2110 the <tt>BasicBlock</tt>s that constitute the <tt>Function</tt>. The following is
2111 an example that prints the name of a <tt>BasicBlock</tt> and the number of
2112 <tt>Instruction</tt>s it contains:</p>
2114 <div class="doc_code">
2116 // <i>func is a pointer to a Function instance</i>
2117 for (Function::iterator i = func->begin(), e = func->end(); i != e; ++i)
2118 // <i>Print out the name of the basic block if it has one, and then the</i>
2119 // <i>number of instructions that it contains</i>
2120 errs() << "Basic block (name=" << i->getName() << ") has "
2121 << i->size() << " instructions.\n";
2125 <p>Note that i can be used as if it were a pointer for the purposes of
2126 invoking member functions of the <tt>Instruction</tt> class. This is
2127 because the indirection operator is overloaded for the iterator
2128 classes. In the above code, the expression <tt>i->size()</tt> is
2129 exactly equivalent to <tt>(*i).size()</tt> just like you'd expect.</p>
2133 <!-- _______________________________________________________________________ -->
2135 <a name="iterate_basicblock">Iterating over the </a><a
2136 href="#Instruction"><tt>Instruction</tt></a>s in a <a
2137 href="#BasicBlock"><tt>BasicBlock</tt></a>
2142 <p>Just like when dealing with <tt>BasicBlock</tt>s in <tt>Function</tt>s, it's
2143 easy to iterate over the individual instructions that make up
2144 <tt>BasicBlock</tt>s. Here's a code snippet that prints out each instruction in
2145 a <tt>BasicBlock</tt>:</p>
2147 <div class="doc_code">
2149 // <i>blk is a pointer to a BasicBlock instance</i>
2150 for (BasicBlock::iterator i = blk->begin(), e = blk->end(); i != e; ++i)
2151 // <i>The next statement works since operator<<(ostream&,...)</i>
2152 // <i>is overloaded for Instruction&</i>
2153 errs() << *i << "\n";
2157 <p>However, this isn't really the best way to print out the contents of a
2158 <tt>BasicBlock</tt>! Since the ostream operators are overloaded for virtually
2159 anything you'll care about, you could have just invoked the print routine on the
2160 basic block itself: <tt>errs() << *blk << "\n";</tt>.</p>
2164 <!-- _______________________________________________________________________ -->
2166 <a name="iterate_institer">Iterating over the </a><a
2167 href="#Instruction"><tt>Instruction</tt></a>s in a <a
2168 href="#Function"><tt>Function</tt></a>
2173 <p>If you're finding that you commonly iterate over a <tt>Function</tt>'s
2174 <tt>BasicBlock</tt>s and then that <tt>BasicBlock</tt>'s <tt>Instruction</tt>s,
2175 <tt>InstIterator</tt> should be used instead. You'll need to include <a
2176 href="/doxygen/InstIterator_8h-source.html"><tt>llvm/Support/InstIterator.h</tt></a>,
2177 and then instantiate <tt>InstIterator</tt>s explicitly in your code. Here's a
2178 small example that shows how to dump all instructions in a function to the standard error stream:<p>
2180 <div class="doc_code">
2182 #include "<a href="/doxygen/InstIterator_8h-source.html">llvm/Support/InstIterator.h</a>"
2184 // <i>F is a pointer to a Function instance</i>
2185 for (inst_iterator I = inst_begin(F), E = inst_end(F); I != E; ++I)
2186 errs() << *I << "\n";
2190 <p>Easy, isn't it? You can also use <tt>InstIterator</tt>s to fill a
2191 work list with its initial contents. For example, if you wanted to
2192 initialize a work list to contain all instructions in a <tt>Function</tt>
2193 F, all you would need to do is something like:</p>
2195 <div class="doc_code">
2197 std::set<Instruction*> worklist;
2198 // or better yet, SmallPtrSet<Instruction*, 64> worklist;
2200 for (inst_iterator I = inst_begin(F), E = inst_end(F); I != E; ++I)
2201 worklist.insert(&*I);
2205 <p>The STL set <tt>worklist</tt> would now contain all instructions in the
2206 <tt>Function</tt> pointed to by F.</p>
2210 <!-- _______________________________________________________________________ -->
2212 <a name="iterate_convert">Turning an iterator into a class pointer (and
2218 <p>Sometimes, it'll be useful to grab a reference (or pointer) to a class
2219 instance when all you've got at hand is an iterator. Well, extracting
2220 a reference or a pointer from an iterator is very straight-forward.
2221 Assuming that <tt>i</tt> is a <tt>BasicBlock::iterator</tt> and <tt>j</tt>
2222 is a <tt>BasicBlock::const_iterator</tt>:</p>
2224 <div class="doc_code">
2226 Instruction& inst = *i; // <i>Grab reference to instruction reference</i>
2227 Instruction* pinst = &*i; // <i>Grab pointer to instruction reference</i>
2228 const Instruction& inst = *j;
2232 <p>However, the iterators you'll be working with in the LLVM framework are
2233 special: they will automatically convert to a ptr-to-instance type whenever they
2234 need to. Instead of dereferencing the iterator and then taking the address of
2235 the result, you can simply assign the iterator to the proper pointer type and
2236 you get the dereference and address-of operation as a result of the assignment
2237 (behind the scenes, this is a result of overloading casting mechanisms). Thus
2238 the last line of the last example,</p>
2240 <div class="doc_code">
2242 Instruction *pinst = &*i;
2246 <p>is semantically equivalent to</p>
2248 <div class="doc_code">
2250 Instruction *pinst = i;
2254 <p>It's also possible to turn a class pointer into the corresponding iterator,
2255 and this is a constant time operation (very efficient). The following code
2256 snippet illustrates use of the conversion constructors provided by LLVM
2257 iterators. By using these, you can explicitly grab the iterator of something
2258 without actually obtaining it via iteration over some structure:</p>
2260 <div class="doc_code">
2262 void printNextInstruction(Instruction* inst) {
2263 BasicBlock::iterator it(inst);
2264 ++it; // <i>After this line, it refers to the instruction after *inst</i>
2265 if (it != inst->getParent()->end()) errs() << *it << "\n";
2270 <p>Unfortunately, these implicit conversions come at a cost; they prevent
2271 these iterators from conforming to standard iterator conventions, and thus
2272 from being usable with standard algorithms and containers. For example, they
2273 prevent the following code, where <tt>B</tt> is a <tt>BasicBlock</tt>,
2276 <div class="doc_code">
2278 llvm::SmallVector<llvm::Instruction *, 16>(B->begin(), B->end());
2282 <p>Because of this, these implicit conversions may be removed some day,
2283 and <tt>operator*</tt> changed to return a pointer instead of a reference.</p>
2287 <!--_______________________________________________________________________-->
2289 <a name="iterate_complex">Finding call sites: a slightly more complex
2295 <p>Say that you're writing a FunctionPass and would like to count all the
2296 locations in the entire module (that is, across every <tt>Function</tt>) where a
2297 certain function (i.e., some <tt>Function</tt>*) is already in scope. As you'll
2298 learn later, you may want to use an <tt>InstVisitor</tt> to accomplish this in a
2299 much more straight-forward manner, but this example will allow us to explore how
2300 you'd do it if you didn't have <tt>InstVisitor</tt> around. In pseudo-code, this
2301 is what we want to do:</p>
2303 <div class="doc_code">
2305 initialize callCounter to zero
2306 for each Function f in the Module
2307 for each BasicBlock b in f
2308 for each Instruction i in b
2309 if (i is a CallInst and calls the given function)
2310 increment callCounter
2314 <p>And the actual code is (remember, because we're writing a
2315 <tt>FunctionPass</tt>, our <tt>FunctionPass</tt>-derived class simply has to
2316 override the <tt>runOnFunction</tt> method):</p>
2318 <div class="doc_code">
2320 Function* targetFunc = ...;
2322 class OurFunctionPass : public FunctionPass {
2324 OurFunctionPass(): callCounter(0) { }
2326 virtual runOnFunction(Function& F) {
2327 for (Function::iterator b = F.begin(), be = F.end(); b != be; ++b) {
2328 for (BasicBlock::iterator i = b->begin(), ie = b->end(); i != ie; ++i) {
2329 if (<a href="#CallInst">CallInst</a>* callInst = <a href="#isa">dyn_cast</a><<a
2330 href="#CallInst">CallInst</a>>(&*i)) {
2331 // <i>We know we've encountered a call instruction, so we</i>
2332 // <i>need to determine if it's a call to the</i>
2333 // <i>function pointed to by m_func or not.</i>
2334 if (callInst->getCalledFunction() == targetFunc)
2342 unsigned callCounter;
2349 <!--_______________________________________________________________________-->
2351 <a name="calls_and_invokes">Treating calls and invokes the same way</a>
2356 <p>You may have noticed that the previous example was a bit oversimplified in
2357 that it did not deal with call sites generated by 'invoke' instructions. In
2358 this, and in other situations, you may find that you want to treat
2359 <tt>CallInst</tt>s and <tt>InvokeInst</tt>s the same way, even though their
2360 most-specific common base class is <tt>Instruction</tt>, which includes lots of
2361 less closely-related things. For these cases, LLVM provides a handy wrapper
2363 href="http://llvm.org/doxygen/classllvm_1_1CallSite.html"><tt>CallSite</tt></a>.
2364 It is essentially a wrapper around an <tt>Instruction</tt> pointer, with some
2365 methods that provide functionality common to <tt>CallInst</tt>s and
2366 <tt>InvokeInst</tt>s.</p>
2368 <p>This class has "value semantics": it should be passed by value, not by
2369 reference and it should not be dynamically allocated or deallocated using
2370 <tt>operator new</tt> or <tt>operator delete</tt>. It is efficiently copyable,
2371 assignable and constructable, with costs equivalents to that of a bare pointer.
2372 If you look at its definition, it has only a single pointer member.</p>
2376 <!--_______________________________________________________________________-->
2378 <a name="iterate_chains">Iterating over def-use & use-def chains</a>
2383 <p>Frequently, we might have an instance of the <a
2384 href="/doxygen/classllvm_1_1Value.html">Value Class</a> and we want to
2385 determine which <tt>User</tt>s use the <tt>Value</tt>. The list of all
2386 <tt>User</tt>s of a particular <tt>Value</tt> is called a <i>def-use</i> chain.
2387 For example, let's say we have a <tt>Function*</tt> named <tt>F</tt> to a
2388 particular function <tt>foo</tt>. Finding all of the instructions that
2389 <i>use</i> <tt>foo</tt> is as simple as iterating over the <i>def-use</i> chain
2392 <div class="doc_code">
2396 for (Value::use_iterator i = F->use_begin(), e = F->use_end(); i != e; ++i)
2397 if (Instruction *Inst = dyn_cast<Instruction>(*i)) {
2398 errs() << "F is used in instruction:\n";
2399 errs() << *Inst << "\n";
2404 <p>Note that dereferencing a <tt>Value::use_iterator</tt> is not a very cheap
2405 operation. Instead of performing <tt>*i</tt> above several times, consider
2406 doing it only once in the loop body and reusing its result.</p>
2408 <p>Alternatively, it's common to have an instance of the <a
2409 href="/doxygen/classllvm_1_1User.html">User Class</a> and need to know what
2410 <tt>Value</tt>s are used by it. The list of all <tt>Value</tt>s used by a
2411 <tt>User</tt> is known as a <i>use-def</i> chain. Instances of class
2412 <tt>Instruction</tt> are common <tt>User</tt>s, so we might want to iterate over
2413 all of the values that a particular instruction uses (that is, the operands of
2414 the particular <tt>Instruction</tt>):</p>
2416 <div class="doc_code">
2418 Instruction *pi = ...;
2420 for (User::op_iterator i = pi->op_begin(), e = pi->op_end(); i != e; ++i) {
2427 <p>Declaring objects as <tt>const</tt> is an important tool of enforcing
2428 mutation free algorithms (such as analyses, etc.). For this purpose above
2429 iterators come in constant flavors as <tt>Value::const_use_iterator</tt>
2430 and <tt>Value::const_op_iterator</tt>. They automatically arise when
2431 calling <tt>use/op_begin()</tt> on <tt>const Value*</tt>s or
2432 <tt>const User*</tt>s respectively. Upon dereferencing, they return
2433 <tt>const Use*</tt>s. Otherwise the above patterns remain unchanged.</p>
2437 <!--_______________________________________________________________________-->
2439 <a name="iterate_preds">Iterating over predecessors &
2440 successors of blocks</a>
2445 <p>Iterating over the predecessors and successors of a block is quite easy
2446 with the routines defined in <tt>"llvm/Support/CFG.h"</tt>. Just use code like
2447 this to iterate over all predecessors of BB:</p>
2449 <div class="doc_code">
2451 #include "llvm/Support/CFG.h"
2452 BasicBlock *BB = ...;
2454 for (pred_iterator PI = pred_begin(BB), E = pred_end(BB); PI != E; ++PI) {
2455 BasicBlock *Pred = *PI;
2461 <p>Similarly, to iterate over successors use
2462 succ_iterator/succ_begin/succ_end.</p>
2468 <!-- ======================================================================= -->
2470 <a name="simplechanges">Making simple changes</a>
2475 <p>There are some primitive transformation operations present in the LLVM
2476 infrastructure that are worth knowing about. When performing
2477 transformations, it's fairly common to manipulate the contents of basic
2478 blocks. This section describes some of the common methods for doing so
2479 and gives example code.</p>
2481 <!--_______________________________________________________________________-->
2483 <a name="schanges_creating">Creating and inserting new
2484 <tt>Instruction</tt>s</a>
2489 <p><i>Instantiating Instructions</i></p>
2491 <p>Creation of <tt>Instruction</tt>s is straight-forward: simply call the
2492 constructor for the kind of instruction to instantiate and provide the necessary
2493 parameters. For example, an <tt>AllocaInst</tt> only <i>requires</i> a
2494 (const-ptr-to) <tt>Type</tt>. Thus:</p>
2496 <div class="doc_code">
2498 AllocaInst* ai = new AllocaInst(Type::Int32Ty);
2502 <p>will create an <tt>AllocaInst</tt> instance that represents the allocation of
2503 one integer in the current stack frame, at run time. Each <tt>Instruction</tt>
2504 subclass is likely to have varying default parameters which change the semantics
2505 of the instruction, so refer to the <a
2506 href="/doxygen/classllvm_1_1Instruction.html">doxygen documentation for the subclass of
2507 Instruction</a> that you're interested in instantiating.</p>
2509 <p><i>Naming values</i></p>
2511 <p>It is very useful to name the values of instructions when you're able to, as
2512 this facilitates the debugging of your transformations. If you end up looking
2513 at generated LLVM machine code, you definitely want to have logical names
2514 associated with the results of instructions! By supplying a value for the
2515 <tt>Name</tt> (default) parameter of the <tt>Instruction</tt> constructor, you
2516 associate a logical name with the result of the instruction's execution at
2517 run time. For example, say that I'm writing a transformation that dynamically
2518 allocates space for an integer on the stack, and that integer is going to be
2519 used as some kind of index by some other code. To accomplish this, I place an
2520 <tt>AllocaInst</tt> at the first point in the first <tt>BasicBlock</tt> of some
2521 <tt>Function</tt>, and I'm intending to use it within the same
2522 <tt>Function</tt>. I might do:</p>
2524 <div class="doc_code">
2526 AllocaInst* pa = new AllocaInst(Type::Int32Ty, 0, "indexLoc");
2530 <p>where <tt>indexLoc</tt> is now the logical name of the instruction's
2531 execution value, which is a pointer to an integer on the run time stack.</p>
2533 <p><i>Inserting instructions</i></p>
2535 <p>There are essentially two ways to insert an <tt>Instruction</tt>
2536 into an existing sequence of instructions that form a <tt>BasicBlock</tt>:</p>
2539 <li>Insertion into an explicit instruction list
2541 <p>Given a <tt>BasicBlock* pb</tt>, an <tt>Instruction* pi</tt> within that
2542 <tt>BasicBlock</tt>, and a newly-created instruction we wish to insert
2543 before <tt>*pi</tt>, we do the following: </p>
2545 <div class="doc_code">
2547 BasicBlock *pb = ...;
2548 Instruction *pi = ...;
2549 Instruction *newInst = new Instruction(...);
2551 pb->getInstList().insert(pi, newInst); // <i>Inserts newInst before pi in pb</i>
2555 <p>Appending to the end of a <tt>BasicBlock</tt> is so common that
2556 the <tt>Instruction</tt> class and <tt>Instruction</tt>-derived
2557 classes provide constructors which take a pointer to a
2558 <tt>BasicBlock</tt> to be appended to. For example code that
2561 <div class="doc_code">
2563 BasicBlock *pb = ...;
2564 Instruction *newInst = new Instruction(...);
2566 pb->getInstList().push_back(newInst); // <i>Appends newInst to pb</i>
2572 <div class="doc_code">
2574 BasicBlock *pb = ...;
2575 Instruction *newInst = new Instruction(..., pb);
2579 <p>which is much cleaner, especially if you are creating
2580 long instruction streams.</p></li>
2582 <li>Insertion into an implicit instruction list
2584 <p><tt>Instruction</tt> instances that are already in <tt>BasicBlock</tt>s
2585 are implicitly associated with an existing instruction list: the instruction
2586 list of the enclosing basic block. Thus, we could have accomplished the same
2587 thing as the above code without being given a <tt>BasicBlock</tt> by doing:
2590 <div class="doc_code">
2592 Instruction *pi = ...;
2593 Instruction *newInst = new Instruction(...);
2595 pi->getParent()->getInstList().insert(pi, newInst);
2599 <p>In fact, this sequence of steps occurs so frequently that the
2600 <tt>Instruction</tt> class and <tt>Instruction</tt>-derived classes provide
2601 constructors which take (as a default parameter) a pointer to an
2602 <tt>Instruction</tt> which the newly-created <tt>Instruction</tt> should
2603 precede. That is, <tt>Instruction</tt> constructors are capable of
2604 inserting the newly-created instance into the <tt>BasicBlock</tt> of a
2605 provided instruction, immediately before that instruction. Using an
2606 <tt>Instruction</tt> constructor with a <tt>insertBefore</tt> (default)
2607 parameter, the above code becomes:</p>
2609 <div class="doc_code">
2611 Instruction* pi = ...;
2612 Instruction* newInst = new Instruction(..., pi);
2616 <p>which is much cleaner, especially if you're creating a lot of
2617 instructions and adding them to <tt>BasicBlock</tt>s.</p></li>
2622 <!--_______________________________________________________________________-->
2624 <a name="schanges_deleting">Deleting <tt>Instruction</tt>s</a>
2629 <p>Deleting an instruction from an existing sequence of instructions that form a
2630 <a href="#BasicBlock"><tt>BasicBlock</tt></a> is very straight-forward: just
2631 call the instruction's eraseFromParent() method. For example:</p>
2633 <div class="doc_code">
2635 <a href="#Instruction">Instruction</a> *I = .. ;
2636 I->eraseFromParent();
2640 <p>This unlinks the instruction from its containing basic block and deletes
2641 it. If you'd just like to unlink the instruction from its containing basic
2642 block but not delete it, you can use the <tt>removeFromParent()</tt> method.</p>
2646 <!--_______________________________________________________________________-->
2648 <a name="schanges_replacing">Replacing an <tt>Instruction</tt> with another
2654 <h5><i>Replacing individual instructions</i></h5>
2656 <p>Including "<a href="/doxygen/BasicBlockUtils_8h-source.html">llvm/Transforms/Utils/BasicBlockUtils.h</a>"
2657 permits use of two very useful replace functions: <tt>ReplaceInstWithValue</tt>
2658 and <tt>ReplaceInstWithInst</tt>.</p>
2660 <h5><a name="schanges_deleting">Deleting <tt>Instruction</tt>s</a></h5>
2664 <li><tt>ReplaceInstWithValue</tt>
2666 <p>This function replaces all uses of a given instruction with a value,
2667 and then removes the original instruction. The following example
2668 illustrates the replacement of the result of a particular
2669 <tt>AllocaInst</tt> that allocates memory for a single integer with a null
2670 pointer to an integer.</p>
2672 <div class="doc_code">
2674 AllocaInst* instToReplace = ...;
2675 BasicBlock::iterator ii(instToReplace);
2677 ReplaceInstWithValue(instToReplace->getParent()->getInstList(), ii,
2678 Constant::getNullValue(PointerType::getUnqual(Type::Int32Ty)));
2681 <li><tt>ReplaceInstWithInst</tt>
2683 <p>This function replaces a particular instruction with another
2684 instruction, inserting the new instruction into the basic block at the
2685 location where the old instruction was, and replacing any uses of the old
2686 instruction with the new instruction. The following example illustrates
2687 the replacement of one <tt>AllocaInst</tt> with another.</p>
2689 <div class="doc_code">
2691 AllocaInst* instToReplace = ...;
2692 BasicBlock::iterator ii(instToReplace);
2694 ReplaceInstWithInst(instToReplace->getParent()->getInstList(), ii,
2695 new AllocaInst(Type::Int32Ty, 0, "ptrToReplacedInt"));
2701 <h5><i>Replacing multiple uses of <tt>User</tt>s and <tt>Value</tt>s</i></h5>
2703 <p>You can use <tt>Value::replaceAllUsesWith</tt> and
2704 <tt>User::replaceUsesOfWith</tt> to change more than one use at a time. See the
2705 doxygen documentation for the <a href="/doxygen/classllvm_1_1Value.html">Value Class</a>
2706 and <a href="/doxygen/classllvm_1_1User.html">User Class</a>, respectively, for more
2709 <!-- Value::replaceAllUsesWith User::replaceUsesOfWith Point out:
2710 include/llvm/Transforms/Utils/ especially BasicBlockUtils.h with:
2711 ReplaceInstWithValue, ReplaceInstWithInst -->
2715 <!--_______________________________________________________________________-->
2717 <a name="schanges_deletingGV">Deleting <tt>GlobalVariable</tt>s</a>
2722 <p>Deleting a global variable from a module is just as easy as deleting an
2723 Instruction. First, you must have a pointer to the global variable that you wish
2724 to delete. You use this pointer to erase it from its parent, the module.
2727 <div class="doc_code">
2729 <a href="#GlobalVariable">GlobalVariable</a> *GV = .. ;
2731 GV->eraseFromParent();
2739 <!-- ======================================================================= -->
2741 <a name="create_types">How to Create Types</a>
2746 <p>In generating IR, you may need some complex types. If you know these types
2747 statically, you can use <tt>TypeBuilder<...>::get()</tt>, defined
2748 in <tt>llvm/Support/TypeBuilder.h</tt>, to retrieve them. <tt>TypeBuilder</tt>
2749 has two forms depending on whether you're building types for cross-compilation
2750 or native library use. <tt>TypeBuilder<T, true></tt> requires
2751 that <tt>T</tt> be independent of the host environment, meaning that it's built
2753 the <a href="/doxygen/namespacellvm_1_1types.html"><tt>llvm::types</tt></a>
2754 namespace and pointers, functions, arrays, etc. built of
2755 those. <tt>TypeBuilder<T, false></tt> additionally allows native C types
2756 whose size may depend on the host compiler. For example,</p>
2758 <div class="doc_code">
2760 FunctionType *ft = TypeBuilder<types::i<8>(types::i<32>*), true>::get();
2764 <p>is easier to read and write than the equivalent</p>
2766 <div class="doc_code">
2768 std::vector<const Type*> params;
2769 params.push_back(PointerType::getUnqual(Type::Int32Ty));
2770 FunctionType *ft = FunctionType::get(Type::Int8Ty, params, false);
2774 <p>See the <a href="/doxygen/TypeBuilder_8h-source.html#l00001">class
2775 comment</a> for more details.</p>
2781 <!-- *********************************************************************** -->
2783 <a name="threading">Threads and LLVM</a>
2785 <!-- *********************************************************************** -->
2789 This section describes the interaction of the LLVM APIs with multithreading,
2790 both on the part of client applications, and in the JIT, in the hosted
2795 Note that LLVM's support for multithreading is still relatively young. Up
2796 through version 2.5, the execution of threaded hosted applications was
2797 supported, but not threaded client access to the APIs. While this use case is
2798 now supported, clients <em>must</em> adhere to the guidelines specified below to
2799 ensure proper operation in multithreaded mode.
2803 Note that, on Unix-like platforms, LLVM requires the presence of GCC's atomic
2804 intrinsics in order to support threaded operation. If you need a
2805 multhreading-capable LLVM on a platform without a suitably modern system
2806 compiler, consider compiling LLVM and LLVM-GCC in single-threaded mode, and
2807 using the resultant compiler to build a copy of LLVM with multithreading
2811 <!-- ======================================================================= -->
2813 <a name="startmultithreaded">Entering and Exiting Multithreaded Mode</a>
2819 In order to properly protect its internal data structures while avoiding
2820 excessive locking overhead in the single-threaded case, the LLVM must intialize
2821 certain data structures necessary to provide guards around its internals. To do
2822 so, the client program must invoke <tt>llvm_start_multithreaded()</tt> before
2823 making any concurrent LLVM API calls. To subsequently tear down these
2824 structures, use the <tt>llvm_stop_multithreaded()</tt> call. You can also use
2825 the <tt>llvm_is_multithreaded()</tt> call to check the status of multithreaded
2830 Note that both of these calls must be made <em>in isolation</em>. That is to
2831 say that no other LLVM API calls may be executing at any time during the
2832 execution of <tt>llvm_start_multithreaded()</tt> or <tt>llvm_stop_multithreaded
2833 </tt>. It's is the client's responsibility to enforce this isolation.
2837 The return value of <tt>llvm_start_multithreaded()</tt> indicates the success or
2838 failure of the initialization. Failure typically indicates that your copy of
2839 LLVM was built without multithreading support, typically because GCC atomic
2840 intrinsics were not found in your system compiler. In this case, the LLVM API
2841 will not be safe for concurrent calls. However, it <em>will</em> be safe for
2842 hosting threaded applications in the JIT, though <a href="#jitthreading">care
2843 must be taken</a> to ensure that side exits and the like do not accidentally
2844 result in concurrent LLVM API calls.
2848 <!-- ======================================================================= -->
2850 <a name="shutdown">Ending Execution with <tt>llvm_shutdown()</tt></a>
2855 When you are done using the LLVM APIs, you should call <tt>llvm_shutdown()</tt>
2856 to deallocate memory used for internal structures. This will also invoke
2857 <tt>llvm_stop_multithreaded()</tt> if LLVM is operating in multithreaded mode.
2858 As such, <tt>llvm_shutdown()</tt> requires the same isolation guarantees as
2859 <tt>llvm_stop_multithreaded()</tt>.
2863 Note that, if you use scope-based shutdown, you can use the
2864 <tt>llvm_shutdown_obj</tt> class, which calls <tt>llvm_shutdown()</tt> in its
2868 <!-- ======================================================================= -->
2870 <a name="managedstatic">Lazy Initialization with <tt>ManagedStatic</tt></a>
2875 <tt>ManagedStatic</tt> is a utility class in LLVM used to implement static
2876 initialization of static resources, such as the global type tables. Before the
2877 invocation of <tt>llvm_shutdown()</tt>, it implements a simple lazy
2878 initialization scheme. Once <tt>llvm_start_multithreaded()</tt> returns,
2879 however, it uses double-checked locking to implement thread-safe lazy
2884 Note that, because no other threads are allowed to issue LLVM API calls before
2885 <tt>llvm_start_multithreaded()</tt> returns, it is possible to have
2886 <tt>ManagedStatic</tt>s of <tt>llvm::sys::Mutex</tt>s.
2890 The <tt>llvm_acquire_global_lock()</tt> and <tt>llvm_release_global_lock</tt>
2891 APIs provide access to the global lock used to implement the double-checked
2892 locking for lazy initialization. These should only be used internally to LLVM,
2893 and only if you know what you're doing!
2897 <!-- ======================================================================= -->
2899 <a name="llvmcontext">Achieving Isolation with <tt>LLVMContext</tt></a>
2904 <tt>LLVMContext</tt> is an opaque class in the LLVM API which clients can use
2905 to operate multiple, isolated instances of LLVM concurrently within the same
2906 address space. For instance, in a hypothetical compile-server, the compilation
2907 of an individual translation unit is conceptually independent from all the
2908 others, and it would be desirable to be able to compile incoming translation
2909 units concurrently on independent server threads. Fortunately,
2910 <tt>LLVMContext</tt> exists to enable just this kind of scenario!
2914 Conceptually, <tt>LLVMContext</tt> provides isolation. Every LLVM entity
2915 (<tt>Module</tt>s, <tt>Value</tt>s, <tt>Type</tt>s, <tt>Constant</tt>s, etc.)
2916 in LLVM's in-memory IR belongs to an <tt>LLVMContext</tt>. Entities in
2917 different contexts <em>cannot</em> interact with each other: <tt>Module</tt>s in
2918 different contexts cannot be linked together, <tt>Function</tt>s cannot be added
2919 to <tt>Module</tt>s in different contexts, etc. What this means is that is is
2920 safe to compile on multiple threads simultaneously, as long as no two threads
2921 operate on entities within the same context.
2925 In practice, very few places in the API require the explicit specification of a
2926 <tt>LLVMContext</tt>, other than the <tt>Type</tt> creation/lookup APIs.
2927 Because every <tt>Type</tt> carries a reference to its owning context, most
2928 other entities can determine what context they belong to by looking at their
2929 own <tt>Type</tt>. If you are adding new entities to LLVM IR, please try to
2930 maintain this interface design.
2934 For clients that do <em>not</em> require the benefits of isolation, LLVM
2935 provides a convenience API <tt>getGlobalContext()</tt>. This returns a global,
2936 lazily initialized <tt>LLVMContext</tt> that may be used in situations where
2937 isolation is not a concern.
2941 <!-- ======================================================================= -->
2943 <a name="jitthreading">Threads and the JIT</a>
2948 LLVM's "eager" JIT compiler is safe to use in threaded programs. Multiple
2949 threads can call <tt>ExecutionEngine::getPointerToFunction()</tt> or
2950 <tt>ExecutionEngine::runFunction()</tt> concurrently, and multiple threads can
2951 run code output by the JIT concurrently. The user must still ensure that only
2952 one thread accesses IR in a given <tt>LLVMContext</tt> while another thread
2953 might be modifying it. One way to do that is to always hold the JIT lock while
2954 accessing IR outside the JIT (the JIT <em>modifies</em> the IR by adding
2955 <tt>CallbackVH</tt>s). Another way is to only
2956 call <tt>getPointerToFunction()</tt> from the <tt>LLVMContext</tt>'s thread.
2959 <p>When the JIT is configured to compile lazily (using
2960 <tt>ExecutionEngine::DisableLazyCompilation(false)</tt>), there is currently a
2961 <a href="http://llvm.org/bugs/show_bug.cgi?id=5184">race condition</a> in
2962 updating call sites after a function is lazily-jitted. It's still possible to
2963 use the lazy JIT in a threaded program if you ensure that only one thread at a
2964 time can call any particular lazy stub and that the JIT lock guards any IR
2965 access, but we suggest using only the eager JIT in threaded programs.
2971 <!-- *********************************************************************** -->
2973 <a name="advanced">Advanced Topics</a>
2975 <!-- *********************************************************************** -->
2979 This section describes some of the advanced or obscure API's that most clients
2980 do not need to be aware of. These API's tend manage the inner workings of the
2981 LLVM system, and only need to be accessed in unusual circumstances.
2985 <!-- ======================================================================= -->
2987 <a name="SymbolTable">The <tt>ValueSymbolTable</tt> class</a>
2991 <p>The <tt><a href="http://llvm.org/doxygen/classllvm_1_1ValueSymbolTable.html">
2992 ValueSymbolTable</a></tt> class provides a symbol table that the <a
2993 href="#Function"><tt>Function</tt></a> and <a href="#Module">
2994 <tt>Module</tt></a> classes use for naming value definitions. The symbol table
2995 can provide a name for any <a href="#Value"><tt>Value</tt></a>.
2998 <p>Note that the <tt>SymbolTable</tt> class should not be directly accessed
2999 by most clients. It should only be used when iteration over the symbol table
3000 names themselves are required, which is very special purpose. Note that not
3002 <tt><a href="#Value">Value</a></tt>s have names, and those without names (i.e. they have
3003 an empty name) do not exist in the symbol table.
3006 <p>Symbol tables support iteration over the values in the symbol
3007 table with <tt>begin/end/iterator</tt> and supports querying to see if a
3008 specific name is in the symbol table (with <tt>lookup</tt>). The
3009 <tt>ValueSymbolTable</tt> class exposes no public mutator methods, instead,
3010 simply call <tt>setName</tt> on a value, which will autoinsert it into the
3011 appropriate symbol table.</p>
3017 <!-- ======================================================================= -->
3019 <a name="UserLayout">The <tt>User</tt> and owned <tt>Use</tt> classes' memory layout</a>
3023 <p>The <tt><a href="http://llvm.org/doxygen/classllvm_1_1User.html">
3024 User</a></tt> class provides a basis for expressing the ownership of <tt>User</tt>
3025 towards other <tt><a href="http://llvm.org/doxygen/classllvm_1_1Value.html">
3026 Value</a></tt>s. The <tt><a href="http://llvm.org/doxygen/classllvm_1_1Use.html">
3027 Use</a></tt> helper class is employed to do the bookkeeping and to facilitate <i>O(1)</i>
3028 addition and removal.</p>
3030 <!-- ______________________________________________________________________ -->
3033 Interaction and relationship between <tt>User</tt> and <tt>Use</tt> objects
3039 A subclass of <tt>User</tt> can choose between incorporating its <tt>Use</tt> objects
3040 or refer to them out-of-line by means of a pointer. A mixed variant
3041 (some <tt>Use</tt>s inline others hung off) is impractical and breaks the invariant
3042 that the <tt>Use</tt> objects belonging to the same <tt>User</tt> form a contiguous array.
3046 We have 2 different layouts in the <tt>User</tt> (sub)classes:
3049 The <tt>Use</tt> object(s) are inside (resp. at fixed offset) of the <tt>User</tt>
3050 object and there are a fixed number of them.</p>
3053 The <tt>Use</tt> object(s) are referenced by a pointer to an
3054 array from the <tt>User</tt> object and there may be a variable
3058 As of v2.4 each layout still possesses a direct pointer to the
3059 start of the array of <tt>Use</tt>s. Though not mandatory for layout a),
3060 we stick to this redundancy for the sake of simplicity.
3061 The <tt>User</tt> object also stores the number of <tt>Use</tt> objects it
3062 has. (Theoretically this information can also be calculated
3063 given the scheme presented below.)</p>
3065 Special forms of allocation operators (<tt>operator new</tt>)
3066 enforce the following memory layouts:</p>
3069 <li><p>Layout a) is modelled by prepending the <tt>User</tt> object by the <tt>Use[]</tt> array.</p>
3072 ...---.---.---.---.-------...
3073 | P | P | P | P | User
3074 '''---'---'---'---'-------'''
3077 <li><p>Layout b) is modelled by pointing at the <tt>Use[]</tt> array.</p>
3089 <i>(In the above figures '<tt>P</tt>' stands for the <tt>Use**</tt> that
3090 is stored in each <tt>Use</tt> object in the member <tt>Use::Prev</tt>)</i>
3094 <!-- ______________________________________________________________________ -->
3096 <a name="Waymarking">The waymarking algorithm</a>
3101 Since the <tt>Use</tt> objects are deprived of the direct (back)pointer to
3102 their <tt>User</tt> objects, there must be a fast and exact method to
3103 recover it. This is accomplished by the following scheme:</p>
3105 A bit-encoding in the 2 LSBits (least significant bits) of the <tt>Use::Prev</tt> allows to find the
3106 start of the <tt>User</tt> object:
3108 <li><tt>00</tt> —> binary digit 0</li>
3109 <li><tt>01</tt> —> binary digit 1</li>
3110 <li><tt>10</tt> —> stop and calculate (<tt>s</tt>)</li>
3111 <li><tt>11</tt> —> full stop (<tt>S</tt>)</li>
3114 Given a <tt>Use*</tt>, all we have to do is to walk till we get
3115 a stop and we either have a <tt>User</tt> immediately behind or
3116 we have to walk to the next stop picking up digits
3117 and calculating the offset:</p>
3119 .---.---.---.---.---.---.---.---.---.---.---.---.---.---.---.---.----------------
3120 | 1 | s | 1 | 0 | 1 | 0 | s | 1 | 1 | 0 | s | 1 | 1 | s | 1 | S | User (or User*)
3121 '---'---'---'---'---'---'---'---'---'---'---'---'---'---'---'---'----------------
3122 |+15 |+10 |+6 |+3 |+1
3125 | | |______________________>
3126 | |______________________________________>
3127 |__________________________________________________________>
3130 Only the significant number of bits need to be stored between the
3131 stops, so that the <i>worst case is 20 memory accesses</i> when there are
3132 1000 <tt>Use</tt> objects associated with a <tt>User</tt>.</p>
3136 <!-- ______________________________________________________________________ -->
3138 <a name="ReferenceImpl">Reference implementation</a>
3143 The following literate Haskell fragment demonstrates the concept:</p>
3145 <div class="doc_code">
3147 > import Test.QuickCheck
3149 > digits :: Int -> [Char] -> [Char]
3150 > digits 0 acc = '0' : acc
3151 > digits 1 acc = '1' : acc
3152 > digits n acc = digits (n `div` 2) $ digits (n `mod` 2) acc
3154 > dist :: Int -> [Char] -> [Char]
3157 > dist 1 acc = let r = dist 0 acc in 's' : digits (length r) r
3158 > dist n acc = dist (n - 1) $ dist 1 acc
3160 > takeLast n ss = reverse $ take n $ reverse ss
3162 > test = takeLast 40 $ dist 20 []
3167 Printing <test> gives: <tt>"1s100000s11010s10100s1111s1010s110s11s1S"</tt></p>
3169 The reverse algorithm computes the length of the string just by examining
3170 a certain prefix:</p>
3172 <div class="doc_code">
3174 > pref :: [Char] -> Int
3176 > pref ('s':'1':rest) = decode 2 1 rest
3177 > pref (_:rest) = 1 + pref rest
3179 > decode walk acc ('0':rest) = decode (walk + 1) (acc * 2) rest
3180 > decode walk acc ('1':rest) = decode (walk + 1) (acc * 2 + 1) rest
3181 > decode walk acc _ = walk + acc
3186 Now, as expected, printing <pref test> gives <tt>40</tt>.</p>
3188 We can <i>quickCheck</i> this with following property:</p>
3190 <div class="doc_code">
3192 > testcase = dist 2000 []
3193 > testcaseLength = length testcase
3195 > identityProp n = n > 0 && n <= testcaseLength ==> length arr == pref arr
3196 > where arr = takeLast n testcase
3201 As expected <quickCheck identityProp> gives:</p>
3204 *Main> quickCheck identityProp
3205 OK, passed 100 tests.
3208 Let's be a bit more exhaustive:</p>
3210 <div class="doc_code">
3213 > deepCheck p = check (defaultConfig { configMaxTest = 500 }) p
3218 And here is the result of <deepCheck identityProp>:</p>
3221 *Main> deepCheck identityProp
3222 OK, passed 500 tests.
3227 <!-- ______________________________________________________________________ -->
3229 <a name="Tagging">Tagging considerations</a>
3235 To maintain the invariant that the 2 LSBits of each <tt>Use**</tt> in <tt>Use</tt>
3236 never change after being set up, setters of <tt>Use::Prev</tt> must re-tag the
3237 new <tt>Use**</tt> on every modification. Accordingly getters must strip the
3240 For layout b) instead of the <tt>User</tt> we find a pointer (<tt>User*</tt> with LSBit set).
3241 Following this pointer brings us to the <tt>User</tt>. A portable trick ensures
3242 that the first bytes of <tt>User</tt> (if interpreted as a pointer) never has
3243 the LSBit set. (Portability is relying on the fact that all known compilers place the
3244 <tt>vptr</tt> in the first word of the instances.)</p>
3252 <!-- *********************************************************************** -->
3254 <a name="coreclasses">The Core LLVM Class Hierarchy Reference </a>
3256 <!-- *********************************************************************** -->
3259 <p><tt>#include "<a href="/doxygen/Type_8h-source.html">llvm/Type.h</a>"</tt>
3260 <br>doxygen info: <a href="/doxygen/classllvm_1_1Type.html">Type Class</a></p>
3262 <p>The Core LLVM classes are the primary means of representing the program
3263 being inspected or transformed. The core LLVM classes are defined in
3264 header files in the <tt>include/llvm/</tt> directory, and implemented in
3265 the <tt>lib/VMCore</tt> directory.</p>
3267 <!-- ======================================================================= -->
3269 <a name="Type">The <tt>Type</tt> class and Derived Types</a>
3274 <p><tt>Type</tt> is a superclass of all type classes. Every <tt>Value</tt> has
3275 a <tt>Type</tt>. <tt>Type</tt> cannot be instantiated directly but only
3276 through its subclasses. Certain primitive types (<tt>VoidType</tt>,
3277 <tt>LabelType</tt>, <tt>FloatType</tt> and <tt>DoubleType</tt>) have hidden
3278 subclasses. They are hidden because they offer no useful functionality beyond
3279 what the <tt>Type</tt> class offers except to distinguish themselves from
3280 other subclasses of <tt>Type</tt>.</p>
3281 <p>All other types are subclasses of <tt>DerivedType</tt>. Types can be
3282 named, but this is not a requirement. There exists exactly
3283 one instance of a given shape at any one time. This allows type equality to
3284 be performed with address equality of the Type Instance. That is, given two
3285 <tt>Type*</tt> values, the types are identical if the pointers are identical.
3288 <!-- _______________________________________________________________________ -->
3290 <a name="m_Type">Important Public Methods</a>
3296 <li><tt>bool isIntegerTy() const</tt>: Returns true for any integer type.</li>
3298 <li><tt>bool isFloatingPointTy()</tt>: Return true if this is one of the five
3299 floating point types.</li>
3301 <li><tt>bool isSized()</tt>: Return true if the type has known size. Things
3302 that don't have a size are abstract types, labels and void.</li>
3307 <!-- _______________________________________________________________________ -->
3309 <a name="derivedtypes">Important Derived Types</a>
3313 <dt><tt>IntegerType</tt></dt>
3314 <dd>Subclass of DerivedType that represents integer types of any bit width.
3315 Any bit width between <tt>IntegerType::MIN_INT_BITS</tt> (1) and
3316 <tt>IntegerType::MAX_INT_BITS</tt> (~8 million) can be represented.
3318 <li><tt>static const IntegerType* get(unsigned NumBits)</tt>: get an integer
3319 type of a specific bit width.</li>
3320 <li><tt>unsigned getBitWidth() const</tt>: Get the bit width of an integer
3324 <dt><tt>SequentialType</tt></dt>
3325 <dd>This is subclassed by ArrayType, PointerType and VectorType.
3327 <li><tt>const Type * getElementType() const</tt>: Returns the type of each
3328 of the elements in the sequential type. </li>
3331 <dt><tt>ArrayType</tt></dt>
3332 <dd>This is a subclass of SequentialType and defines the interface for array
3335 <li><tt>unsigned getNumElements() const</tt>: Returns the number of
3336 elements in the array. </li>
3339 <dt><tt>PointerType</tt></dt>
3340 <dd>Subclass of SequentialType for pointer types.</dd>
3341 <dt><tt>VectorType</tt></dt>
3342 <dd>Subclass of SequentialType for vector types. A
3343 vector type is similar to an ArrayType but is distinguished because it is
3344 a first class type whereas ArrayType is not. Vector types are used for
3345 vector operations and are usually small vectors of of an integer or floating
3347 <dt><tt>StructType</tt></dt>
3348 <dd>Subclass of DerivedTypes for struct types.</dd>
3349 <dt><tt><a name="FunctionType">FunctionType</a></tt></dt>
3350 <dd>Subclass of DerivedTypes for function types.
3352 <li><tt>bool isVarArg() const</tt>: Returns true if it's a vararg
3354 <li><tt> const Type * getReturnType() const</tt>: Returns the
3355 return type of the function.</li>
3356 <li><tt>const Type * getParamType (unsigned i)</tt>: Returns
3357 the type of the ith parameter.</li>
3358 <li><tt> const unsigned getNumParams() const</tt>: Returns the
3359 number of formal parameters.</li>
3367 <!-- ======================================================================= -->
3369 <a name="Module">The <tt>Module</tt> class</a>
3375 href="/doxygen/Module_8h-source.html">llvm/Module.h</a>"</tt><br> doxygen info:
3376 <a href="/doxygen/classllvm_1_1Module.html">Module Class</a></p>
3378 <p>The <tt>Module</tt> class represents the top level structure present in LLVM
3379 programs. An LLVM module is effectively either a translation unit of the
3380 original program or a combination of several translation units merged by the
3381 linker. The <tt>Module</tt> class keeps track of a list of <a
3382 href="#Function"><tt>Function</tt></a>s, a list of <a
3383 href="#GlobalVariable"><tt>GlobalVariable</tt></a>s, and a <a
3384 href="#SymbolTable"><tt>SymbolTable</tt></a>. Additionally, it contains a few
3385 helpful member functions that try to make common operations easy.</p>
3387 <!-- _______________________________________________________________________ -->
3389 <a name="m_Module">Important Public Members of the <tt>Module</tt> class</a>
3395 <li><tt>Module::Module(std::string name = "")</tt>
3397 <p>Constructing a <a href="#Module">Module</a> is easy. You can optionally
3398 provide a name for it (probably based on the name of the translation unit).</p>
3401 <li><tt>Module::iterator</tt> - Typedef for function list iterator<br>
3402 <tt>Module::const_iterator</tt> - Typedef for const_iterator.<br>
3404 <tt>begin()</tt>, <tt>end()</tt>
3405 <tt>size()</tt>, <tt>empty()</tt>
3407 <p>These are forwarding methods that make it easy to access the contents of
3408 a <tt>Module</tt> object's <a href="#Function"><tt>Function</tt></a>
3411 <li><tt>Module::FunctionListType &getFunctionList()</tt>
3413 <p> Returns the list of <a href="#Function"><tt>Function</tt></a>s. This is
3414 necessary to use when you need to update the list or perform a complex
3415 action that doesn't have a forwarding method.</p>
3417 <p><!-- Global Variable --></p></li>
3423 <li><tt>Module::global_iterator</tt> - Typedef for global variable list iterator<br>
3425 <tt>Module::const_global_iterator</tt> - Typedef for const_iterator.<br>
3427 <tt>global_begin()</tt>, <tt>global_end()</tt>
3428 <tt>global_size()</tt>, <tt>global_empty()</tt>
3430 <p> These are forwarding methods that make it easy to access the contents of
3431 a <tt>Module</tt> object's <a
3432 href="#GlobalVariable"><tt>GlobalVariable</tt></a> list.</p></li>
3434 <li><tt>Module::GlobalListType &getGlobalList()</tt>
3436 <p>Returns the list of <a
3437 href="#GlobalVariable"><tt>GlobalVariable</tt></a>s. This is necessary to
3438 use when you need to update the list or perform a complex action that
3439 doesn't have a forwarding method.</p>
3441 <p><!-- Symbol table stuff --> </p></li>
3447 <li><tt><a href="#SymbolTable">SymbolTable</a> *getSymbolTable()</tt>
3449 <p>Return a reference to the <a href="#SymbolTable"><tt>SymbolTable</tt></a>
3450 for this <tt>Module</tt>.</p>
3452 <p><!-- Convenience methods --></p></li>
3459 <li><tt><a href="#Function">Function</a> *getFunction(StringRef Name) const
3462 <p>Look up the specified function in the <tt>Module</tt> <a
3463 href="#SymbolTable"><tt>SymbolTable</tt></a>. If it does not exist, return
3464 <tt>null</tt>.</p></li>
3466 <li><tt><a href="#Function">Function</a> *getOrInsertFunction(const
3467 std::string &Name, const <a href="#FunctionType">FunctionType</a> *T)</tt>
3469 <p>Look up the specified function in the <tt>Module</tt> <a
3470 href="#SymbolTable"><tt>SymbolTable</tt></a>. If it does not exist, add an
3471 external declaration for the function and return it.</p></li>
3473 <li><tt>std::string getTypeName(const <a href="#Type">Type</a> *Ty)</tt>
3475 <p>If there is at least one entry in the <a
3476 href="#SymbolTable"><tt>SymbolTable</tt></a> for the specified <a
3477 href="#Type"><tt>Type</tt></a>, return it. Otherwise return the empty
3480 <li><tt>bool addTypeName(const std::string &Name, const <a
3481 href="#Type">Type</a> *Ty)</tt>
3483 <p>Insert an entry in the <a href="#SymbolTable"><tt>SymbolTable</tt></a>
3484 mapping <tt>Name</tt> to <tt>Ty</tt>. If there is already an entry for this
3485 name, true is returned and the <a
3486 href="#SymbolTable"><tt>SymbolTable</tt></a> is not modified.</p></li>
3493 <!-- ======================================================================= -->
3495 <a name="Value">The <tt>Value</tt> class</a>
3500 <p><tt>#include "<a href="/doxygen/Value_8h-source.html">llvm/Value.h</a>"</tt>
3502 doxygen info: <a href="/doxygen/classllvm_1_1Value.html">Value Class</a></p>
3504 <p>The <tt>Value</tt> class is the most important class in the LLVM Source
3505 base. It represents a typed value that may be used (among other things) as an
3506 operand to an instruction. There are many different types of <tt>Value</tt>s,
3507 such as <a href="#Constant"><tt>Constant</tt></a>s,<a
3508 href="#Argument"><tt>Argument</tt></a>s. Even <a
3509 href="#Instruction"><tt>Instruction</tt></a>s and <a
3510 href="#Function"><tt>Function</tt></a>s are <tt>Value</tt>s.</p>
3512 <p>A particular <tt>Value</tt> may be used many times in the LLVM representation
3513 for a program. For example, an incoming argument to a function (represented
3514 with an instance of the <a href="#Argument">Argument</a> class) is "used" by
3515 every instruction in the function that references the argument. To keep track
3516 of this relationship, the <tt>Value</tt> class keeps a list of all of the <a
3517 href="#User"><tt>User</tt></a>s that is using it (the <a
3518 href="#User"><tt>User</tt></a> class is a base class for all nodes in the LLVM
3519 graph that can refer to <tt>Value</tt>s). This use list is how LLVM represents
3520 def-use information in the program, and is accessible through the <tt>use_</tt>*
3521 methods, shown below.</p>
3523 <p>Because LLVM is a typed representation, every LLVM <tt>Value</tt> is typed,
3524 and this <a href="#Type">Type</a> is available through the <tt>getType()</tt>
3525 method. In addition, all LLVM values can be named. The "name" of the
3526 <tt>Value</tt> is a symbolic string printed in the LLVM code:</p>
3528 <div class="doc_code">
3530 %<b>foo</b> = add i32 1, 2
3534 <p><a name="nameWarning">The name of this instruction is "foo".</a> <b>NOTE</b>
3535 that the name of any value may be missing (an empty string), so names should
3536 <b>ONLY</b> be used for debugging (making the source code easier to read,
3537 debugging printouts), they should not be used to keep track of values or map
3538 between them. For this purpose, use a <tt>std::map</tt> of pointers to the
3539 <tt>Value</tt> itself instead.</p>
3541 <p>One important aspect of LLVM is that there is no distinction between an SSA
3542 variable and the operation that produces it. Because of this, any reference to
3543 the value produced by an instruction (or the value available as an incoming
3544 argument, for example) is represented as a direct pointer to the instance of
3546 represents this value. Although this may take some getting used to, it
3547 simplifies the representation and makes it easier to manipulate.</p>
3549 <!-- _______________________________________________________________________ -->
3551 <a name="m_Value">Important Public Members of the <tt>Value</tt> class</a>
3557 <li><tt>Value::use_iterator</tt> - Typedef for iterator over the
3559 <tt>Value::const_use_iterator</tt> - Typedef for const_iterator over
3561 <tt>unsigned use_size()</tt> - Returns the number of users of the
3563 <tt>bool use_empty()</tt> - Returns true if there are no users.<br>
3564 <tt>use_iterator use_begin()</tt> - Get an iterator to the start of
3566 <tt>use_iterator use_end()</tt> - Get an iterator to the end of the
3568 <tt><a href="#User">User</a> *use_back()</tt> - Returns the last
3569 element in the list.
3570 <p> These methods are the interface to access the def-use
3571 information in LLVM. As with all other iterators in LLVM, the naming
3572 conventions follow the conventions defined by the <a href="#stl">STL</a>.</p>
3574 <li><tt><a href="#Type">Type</a> *getType() const</tt>
3575 <p>This method returns the Type of the Value.</p>
3577 <li><tt>bool hasName() const</tt><br>
3578 <tt>std::string getName() const</tt><br>
3579 <tt>void setName(const std::string &Name)</tt>
3580 <p> This family of methods is used to access and assign a name to a <tt>Value</tt>,
3581 be aware of the <a href="#nameWarning">precaution above</a>.</p>
3583 <li><tt>void replaceAllUsesWith(Value *V)</tt>
3585 <p>This method traverses the use list of a <tt>Value</tt> changing all <a
3586 href="#User"><tt>User</tt>s</a> of the current value to refer to
3587 "<tt>V</tt>" instead. For example, if you detect that an instruction always
3588 produces a constant value (for example through constant folding), you can
3589 replace all uses of the instruction with the constant like this:</p>
3591 <div class="doc_code">
3593 Inst->replaceAllUsesWith(ConstVal);
3603 <!-- ======================================================================= -->
3605 <a name="User">The <tt>User</tt> class</a>
3611 <tt>#include "<a href="/doxygen/User_8h-source.html">llvm/User.h</a>"</tt><br>
3612 doxygen info: <a href="/doxygen/classllvm_1_1User.html">User Class</a><br>
3613 Superclass: <a href="#Value"><tt>Value</tt></a></p>
3615 <p>The <tt>User</tt> class is the common base class of all LLVM nodes that may
3616 refer to <a href="#Value"><tt>Value</tt></a>s. It exposes a list of "Operands"
3617 that are all of the <a href="#Value"><tt>Value</tt></a>s that the User is
3618 referring to. The <tt>User</tt> class itself is a subclass of
3621 <p>The operands of a <tt>User</tt> point directly to the LLVM <a
3622 href="#Value"><tt>Value</tt></a> that it refers to. Because LLVM uses Static
3623 Single Assignment (SSA) form, there can only be one definition referred to,
3624 allowing this direct connection. This connection provides the use-def
3625 information in LLVM.</p>
3627 <!-- _______________________________________________________________________ -->
3629 <a name="m_User">Important Public Members of the <tt>User</tt> class</a>
3634 <p>The <tt>User</tt> class exposes the operand list in two ways: through
3635 an index access interface and through an iterator based interface.</p>
3638 <li><tt>Value *getOperand(unsigned i)</tt><br>
3639 <tt>unsigned getNumOperands()</tt>
3640 <p> These two methods expose the operands of the <tt>User</tt> in a
3641 convenient form for direct access.</p></li>
3643 <li><tt>User::op_iterator</tt> - Typedef for iterator over the operand
3645 <tt>op_iterator op_begin()</tt> - Get an iterator to the start of
3646 the operand list.<br>
3647 <tt>op_iterator op_end()</tt> - Get an iterator to the end of the
3649 <p> Together, these methods make up the iterator based interface to
3650 the operands of a <tt>User</tt>.</p></li>
3657 <!-- ======================================================================= -->
3659 <a name="Instruction">The <tt>Instruction</tt> class</a>
3664 <p><tt>#include "</tt><tt><a
3665 href="/doxygen/Instruction_8h-source.html">llvm/Instruction.h</a>"</tt><br>
3666 doxygen info: <a href="/doxygen/classllvm_1_1Instruction.html">Instruction Class</a><br>
3667 Superclasses: <a href="#User"><tt>User</tt></a>, <a
3668 href="#Value"><tt>Value</tt></a></p>
3670 <p>The <tt>Instruction</tt> class is the common base class for all LLVM
3671 instructions. It provides only a few methods, but is a very commonly used
3672 class. The primary data tracked by the <tt>Instruction</tt> class itself is the
3673 opcode (instruction type) and the parent <a
3674 href="#BasicBlock"><tt>BasicBlock</tt></a> the <tt>Instruction</tt> is embedded
3675 into. To represent a specific type of instruction, one of many subclasses of
3676 <tt>Instruction</tt> are used.</p>
3678 <p> Because the <tt>Instruction</tt> class subclasses the <a
3679 href="#User"><tt>User</tt></a> class, its operands can be accessed in the same
3680 way as for other <a href="#User"><tt>User</tt></a>s (with the
3681 <tt>getOperand()</tt>/<tt>getNumOperands()</tt> and
3682 <tt>op_begin()</tt>/<tt>op_end()</tt> methods).</p> <p> An important file for
3683 the <tt>Instruction</tt> class is the <tt>llvm/Instruction.def</tt> file. This
3684 file contains some meta-data about the various different types of instructions
3685 in LLVM. It describes the enum values that are used as opcodes (for example
3686 <tt>Instruction::Add</tt> and <tt>Instruction::ICmp</tt>), as well as the
3687 concrete sub-classes of <tt>Instruction</tt> that implement the instruction (for
3688 example <tt><a href="#BinaryOperator">BinaryOperator</a></tt> and <tt><a
3689 href="#CmpInst">CmpInst</a></tt>). Unfortunately, the use of macros in
3690 this file confuses doxygen, so these enum values don't show up correctly in the
3691 <a href="/doxygen/classllvm_1_1Instruction.html">doxygen output</a>.</p>
3693 <!-- _______________________________________________________________________ -->
3695 <a name="s_Instruction">
3696 Important Subclasses of the <tt>Instruction</tt> class
3701 <li><tt><a name="BinaryOperator">BinaryOperator</a></tt>
3702 <p>This subclasses represents all two operand instructions whose operands
3703 must be the same type, except for the comparison instructions.</p></li>
3704 <li><tt><a name="CastInst">CastInst</a></tt>
3705 <p>This subclass is the parent of the 12 casting instructions. It provides
3706 common operations on cast instructions.</p>
3707 <li><tt><a name="CmpInst">CmpInst</a></tt>
3708 <p>This subclass respresents the two comparison instructions,
3709 <a href="LangRef.html#i_icmp">ICmpInst</a> (integer opreands), and
3710 <a href="LangRef.html#i_fcmp">FCmpInst</a> (floating point operands).</p>
3711 <li><tt><a name="TerminatorInst">TerminatorInst</a></tt>
3712 <p>This subclass is the parent of all terminator instructions (those which
3713 can terminate a block).</p>
3717 <!-- _______________________________________________________________________ -->
3719 <a name="m_Instruction">
3720 Important Public Members of the <tt>Instruction</tt> class
3727 <li><tt><a href="#BasicBlock">BasicBlock</a> *getParent()</tt>
3728 <p>Returns the <a href="#BasicBlock"><tt>BasicBlock</tt></a> that
3729 this <tt>Instruction</tt> is embedded into.</p></li>
3730 <li><tt>bool mayWriteToMemory()</tt>
3731 <p>Returns true if the instruction writes to memory, i.e. it is a
3732 <tt>call</tt>,<tt>free</tt>,<tt>invoke</tt>, or <tt>store</tt>.</p></li>
3733 <li><tt>unsigned getOpcode()</tt>
3734 <p>Returns the opcode for the <tt>Instruction</tt>.</p></li>
3735 <li><tt><a href="#Instruction">Instruction</a> *clone() const</tt>
3736 <p>Returns another instance of the specified instruction, identical
3737 in all ways to the original except that the instruction has no parent
3738 (ie it's not embedded into a <a href="#BasicBlock"><tt>BasicBlock</tt></a>),
3739 and it has no name</p></li>
3746 <!-- ======================================================================= -->
3748 <a name="Constant">The <tt>Constant</tt> class and subclasses</a>
3753 <p>Constant represents a base class for different types of constants. It
3754 is subclassed by ConstantInt, ConstantArray, etc. for representing
3755 the various types of Constants. <a href="#GlobalValue">GlobalValue</a> is also
3756 a subclass, which represents the address of a global variable or function.
3759 <!-- _______________________________________________________________________ -->
3760 <h4>Important Subclasses of Constant</h4>
3763 <li>ConstantInt : This subclass of Constant represents an integer constant of
3766 <li><tt>const APInt& getValue() const</tt>: Returns the underlying
3767 value of this constant, an APInt value.</li>
3768 <li><tt>int64_t getSExtValue() const</tt>: Converts the underlying APInt
3769 value to an int64_t via sign extension. If the value (not the bit width)
3770 of the APInt is too large to fit in an int64_t, an assertion will result.
3771 For this reason, use of this method is discouraged.</li>
3772 <li><tt>uint64_t getZExtValue() const</tt>: Converts the underlying APInt
3773 value to a uint64_t via zero extension. IF the value (not the bit width)
3774 of the APInt is too large to fit in a uint64_t, an assertion will result.
3775 For this reason, use of this method is discouraged.</li>
3776 <li><tt>static ConstantInt* get(const APInt& Val)</tt>: Returns the
3777 ConstantInt object that represents the value provided by <tt>Val</tt>.
3778 The type is implied as the IntegerType that corresponds to the bit width
3779 of <tt>Val</tt>.</li>
3780 <li><tt>static ConstantInt* get(const Type *Ty, uint64_t Val)</tt>:
3781 Returns the ConstantInt object that represents the value provided by
3782 <tt>Val</tt> for integer type <tt>Ty</tt>.</li>
3785 <li>ConstantFP : This class represents a floating point constant.
3787 <li><tt>double getValue() const</tt>: Returns the underlying value of
3788 this constant. </li>
3791 <li>ConstantArray : This represents a constant array.
3793 <li><tt>const std::vector<Use> &getValues() const</tt>: Returns
3794 a vector of component constants that makeup this array. </li>
3797 <li>ConstantStruct : This represents a constant struct.
3799 <li><tt>const std::vector<Use> &getValues() const</tt>: Returns
3800 a vector of component constants that makeup this array. </li>
3803 <li>GlobalValue : This represents either a global variable or a function. In
3804 either case, the value is a constant fixed address (after linking).
3811 <!-- ======================================================================= -->
3813 <a name="GlobalValue">The <tt>GlobalValue</tt> class</a>
3819 href="/doxygen/GlobalValue_8h-source.html">llvm/GlobalValue.h</a>"</tt><br>
3820 doxygen info: <a href="/doxygen/classllvm_1_1GlobalValue.html">GlobalValue
3822 Superclasses: <a href="#Constant"><tt>Constant</tt></a>,
3823 <a href="#User"><tt>User</tt></a>, <a href="#Value"><tt>Value</tt></a></p>
3825 <p>Global values (<a href="#GlobalVariable"><tt>GlobalVariable</tt></a>s or <a
3826 href="#Function"><tt>Function</tt></a>s) are the only LLVM values that are
3827 visible in the bodies of all <a href="#Function"><tt>Function</tt></a>s.
3828 Because they are visible at global scope, they are also subject to linking with
3829 other globals defined in different translation units. To control the linking
3830 process, <tt>GlobalValue</tt>s know their linkage rules. Specifically,
3831 <tt>GlobalValue</tt>s know whether they have internal or external linkage, as
3832 defined by the <tt>LinkageTypes</tt> enumeration.</p>
3834 <p>If a <tt>GlobalValue</tt> has internal linkage (equivalent to being
3835 <tt>static</tt> in C), it is not visible to code outside the current translation
3836 unit, and does not participate in linking. If it has external linkage, it is
3837 visible to external code, and does participate in linking. In addition to
3838 linkage information, <tt>GlobalValue</tt>s keep track of which <a
3839 href="#Module"><tt>Module</tt></a> they are currently part of.</p>
3841 <p>Because <tt>GlobalValue</tt>s are memory objects, they are always referred to
3842 by their <b>address</b>. As such, the <a href="#Type"><tt>Type</tt></a> of a
3843 global is always a pointer to its contents. It is important to remember this
3844 when using the <tt>GetElementPtrInst</tt> instruction because this pointer must
3845 be dereferenced first. For example, if you have a <tt>GlobalVariable</tt> (a
3846 subclass of <tt>GlobalValue)</tt> that is an array of 24 ints, type <tt>[24 x
3847 i32]</tt>, then the <tt>GlobalVariable</tt> is a pointer to that array. Although
3848 the address of the first element of this array and the value of the
3849 <tt>GlobalVariable</tt> are the same, they have different types. The
3850 <tt>GlobalVariable</tt>'s type is <tt>[24 x i32]</tt>. The first element's type
3851 is <tt>i32.</tt> Because of this, accessing a global value requires you to
3852 dereference the pointer with <tt>GetElementPtrInst</tt> first, then its elements
3853 can be accessed. This is explained in the <a href="LangRef.html#globalvars">LLVM
3854 Language Reference Manual</a>.</p>
3856 <!-- _______________________________________________________________________ -->
3858 <a name="m_GlobalValue">
3859 Important Public Members of the <tt>GlobalValue</tt> class
3866 <li><tt>bool hasInternalLinkage() const</tt><br>
3867 <tt>bool hasExternalLinkage() const</tt><br>
3868 <tt>void setInternalLinkage(bool HasInternalLinkage)</tt>
3869 <p> These methods manipulate the linkage characteristics of the <tt>GlobalValue</tt>.</p>
3872 <li><tt><a href="#Module">Module</a> *getParent()</tt>
3873 <p> This returns the <a href="#Module"><tt>Module</tt></a> that the
3874 GlobalValue is currently embedded into.</p></li>
3881 <!-- ======================================================================= -->
3883 <a name="Function">The <tt>Function</tt> class</a>
3889 href="/doxygen/Function_8h-source.html">llvm/Function.h</a>"</tt><br> doxygen
3890 info: <a href="/doxygen/classllvm_1_1Function.html">Function Class</a><br>
3891 Superclasses: <a href="#GlobalValue"><tt>GlobalValue</tt></a>,
3892 <a href="#Constant"><tt>Constant</tt></a>,
3893 <a href="#User"><tt>User</tt></a>,
3894 <a href="#Value"><tt>Value</tt></a></p>
3896 <p>The <tt>Function</tt> class represents a single procedure in LLVM. It is
3897 actually one of the more complex classes in the LLVM hierarchy because it must
3898 keep track of a large amount of data. The <tt>Function</tt> class keeps track
3899 of a list of <a href="#BasicBlock"><tt>BasicBlock</tt></a>s, a list of formal
3900 <a href="#Argument"><tt>Argument</tt></a>s, and a
3901 <a href="#SymbolTable"><tt>SymbolTable</tt></a>.</p>
3903 <p>The list of <a href="#BasicBlock"><tt>BasicBlock</tt></a>s is the most
3904 commonly used part of <tt>Function</tt> objects. The list imposes an implicit
3905 ordering of the blocks in the function, which indicate how the code will be
3906 laid out by the backend. Additionally, the first <a
3907 href="#BasicBlock"><tt>BasicBlock</tt></a> is the implicit entry node for the
3908 <tt>Function</tt>. It is not legal in LLVM to explicitly branch to this initial
3909 block. There are no implicit exit nodes, and in fact there may be multiple exit
3910 nodes from a single <tt>Function</tt>. If the <a
3911 href="#BasicBlock"><tt>BasicBlock</tt></a> list is empty, this indicates that
3912 the <tt>Function</tt> is actually a function declaration: the actual body of the
3913 function hasn't been linked in yet.</p>
3915 <p>In addition to a list of <a href="#BasicBlock"><tt>BasicBlock</tt></a>s, the
3916 <tt>Function</tt> class also keeps track of the list of formal <a
3917 href="#Argument"><tt>Argument</tt></a>s that the function receives. This
3918 container manages the lifetime of the <a href="#Argument"><tt>Argument</tt></a>
3919 nodes, just like the <a href="#BasicBlock"><tt>BasicBlock</tt></a> list does for
3920 the <a href="#BasicBlock"><tt>BasicBlock</tt></a>s.</p>
3922 <p>The <a href="#SymbolTable"><tt>SymbolTable</tt></a> is a very rarely used
3923 LLVM feature that is only used when you have to look up a value by name. Aside
3924 from that, the <a href="#SymbolTable"><tt>SymbolTable</tt></a> is used
3925 internally to make sure that there are not conflicts between the names of <a
3926 href="#Instruction"><tt>Instruction</tt></a>s, <a
3927 href="#BasicBlock"><tt>BasicBlock</tt></a>s, or <a
3928 href="#Argument"><tt>Argument</tt></a>s in the function body.</p>
3930 <p>Note that <tt>Function</tt> is a <a href="#GlobalValue">GlobalValue</a>
3931 and therefore also a <a href="#Constant">Constant</a>. The value of the function
3932 is its address (after linking) which is guaranteed to be constant.</p>
3934 <!-- _______________________________________________________________________ -->
3936 <a name="m_Function">
3937 Important Public Members of the <tt>Function</tt> class
3944 <li><tt>Function(const </tt><tt><a href="#FunctionType">FunctionType</a>
3945 *Ty, LinkageTypes Linkage, const std::string &N = "", Module* Parent = 0)</tt>
3947 <p>Constructor used when you need to create new <tt>Function</tt>s to add
3948 the the program. The constructor must specify the type of the function to
3949 create and what type of linkage the function should have. The <a
3950 href="#FunctionType"><tt>FunctionType</tt></a> argument
3951 specifies the formal arguments and return value for the function. The same
3952 <a href="#FunctionType"><tt>FunctionType</tt></a> value can be used to
3953 create multiple functions. The <tt>Parent</tt> argument specifies the Module
3954 in which the function is defined. If this argument is provided, the function
3955 will automatically be inserted into that module's list of
3958 <li><tt>bool isDeclaration()</tt>
3960 <p>Return whether or not the <tt>Function</tt> has a body defined. If the
3961 function is "external", it does not have a body, and thus must be resolved
3962 by linking with a function defined in a different translation unit.</p></li>
3964 <li><tt>Function::iterator</tt> - Typedef for basic block list iterator<br>
3965 <tt>Function::const_iterator</tt> - Typedef for const_iterator.<br>
3967 <tt>begin()</tt>, <tt>end()</tt>
3968 <tt>size()</tt>, <tt>empty()</tt>
3970 <p>These are forwarding methods that make it easy to access the contents of
3971 a <tt>Function</tt> object's <a href="#BasicBlock"><tt>BasicBlock</tt></a>
3974 <li><tt>Function::BasicBlockListType &getBasicBlockList()</tt>
3976 <p>Returns the list of <a href="#BasicBlock"><tt>BasicBlock</tt></a>s. This
3977 is necessary to use when you need to update the list or perform a complex
3978 action that doesn't have a forwarding method.</p></li>
3980 <li><tt>Function::arg_iterator</tt> - Typedef for the argument list
3982 <tt>Function::const_arg_iterator</tt> - Typedef for const_iterator.<br>
3984 <tt>arg_begin()</tt>, <tt>arg_end()</tt>
3985 <tt>arg_size()</tt>, <tt>arg_empty()</tt>
3987 <p>These are forwarding methods that make it easy to access the contents of
3988 a <tt>Function</tt> object's <a href="#Argument"><tt>Argument</tt></a>
3991 <li><tt>Function::ArgumentListType &getArgumentList()</tt>
3993 <p>Returns the list of <a href="#Argument"><tt>Argument</tt></a>s. This is
3994 necessary to use when you need to update the list or perform a complex
3995 action that doesn't have a forwarding method.</p></li>
3997 <li><tt><a href="#BasicBlock">BasicBlock</a> &getEntryBlock()</tt>
3999 <p>Returns the entry <a href="#BasicBlock"><tt>BasicBlock</tt></a> for the
4000 function. Because the entry block for the function is always the first
4001 block, this returns the first block of the <tt>Function</tt>.</p></li>
4003 <li><tt><a href="#Type">Type</a> *getReturnType()</tt><br>
4004 <tt><a href="#FunctionType">FunctionType</a> *getFunctionType()</tt>
4006 <p>This traverses the <a href="#Type"><tt>Type</tt></a> of the
4007 <tt>Function</tt> and returns the return type of the function, or the <a
4008 href="#FunctionType"><tt>FunctionType</tt></a> of the actual
4011 <li><tt><a href="#SymbolTable">SymbolTable</a> *getSymbolTable()</tt>
4013 <p> Return a pointer to the <a href="#SymbolTable"><tt>SymbolTable</tt></a>
4014 for this <tt>Function</tt>.</p></li>
4021 <!-- ======================================================================= -->
4023 <a name="GlobalVariable">The <tt>GlobalVariable</tt> class</a>
4029 href="/doxygen/GlobalVariable_8h-source.html">llvm/GlobalVariable.h</a>"</tt>
4031 doxygen info: <a href="/doxygen/classllvm_1_1GlobalVariable.html">GlobalVariable
4033 Superclasses: <a href="#GlobalValue"><tt>GlobalValue</tt></a>,
4034 <a href="#Constant"><tt>Constant</tt></a>,
4035 <a href="#User"><tt>User</tt></a>,
4036 <a href="#Value"><tt>Value</tt></a></p>
4038 <p>Global variables are represented with the (surprise surprise)
4039 <tt>GlobalVariable</tt> class. Like functions, <tt>GlobalVariable</tt>s are also
4040 subclasses of <a href="#GlobalValue"><tt>GlobalValue</tt></a>, and as such are
4041 always referenced by their address (global values must live in memory, so their
4042 "name" refers to their constant address). See
4043 <a href="#GlobalValue"><tt>GlobalValue</tt></a> for more on this. Global
4044 variables may have an initial value (which must be a
4045 <a href="#Constant"><tt>Constant</tt></a>), and if they have an initializer,
4046 they may be marked as "constant" themselves (indicating that their contents
4047 never change at runtime).</p>
4049 <!-- _______________________________________________________________________ -->
4051 <a name="m_GlobalVariable">
4052 Important Public Members of the <tt>GlobalVariable</tt> class
4059 <li><tt>GlobalVariable(const </tt><tt><a href="#Type">Type</a> *Ty, bool
4060 isConstant, LinkageTypes& Linkage, <a href="#Constant">Constant</a>
4061 *Initializer = 0, const std::string &Name = "", Module* Parent = 0)</tt>
4063 <p>Create a new global variable of the specified type. If
4064 <tt>isConstant</tt> is true then the global variable will be marked as
4065 unchanging for the program. The Linkage parameter specifies the type of
4066 linkage (internal, external, weak, linkonce, appending) for the variable.
4067 If the linkage is InternalLinkage, WeakAnyLinkage, WeakODRLinkage,
4068 LinkOnceAnyLinkage or LinkOnceODRLinkage, then the resultant
4069 global variable will have internal linkage. AppendingLinkage concatenates
4070 together all instances (in different translation units) of the variable
4071 into a single variable but is only applicable to arrays. See
4072 the <a href="LangRef.html#modulestructure">LLVM Language Reference</a> for
4073 further details on linkage types. Optionally an initializer, a name, and the
4074 module to put the variable into may be specified for the global variable as
4077 <li><tt>bool isConstant() const</tt>
4079 <p>Returns true if this is a global variable that is known not to
4080 be modified at runtime.</p></li>
4082 <li><tt>bool hasInitializer()</tt>
4084 <p>Returns true if this <tt>GlobalVariable</tt> has an intializer.</p></li>
4086 <li><tt><a href="#Constant">Constant</a> *getInitializer()</tt>
4088 <p>Returns the initial value for a <tt>GlobalVariable</tt>. It is not legal
4089 to call this method if there is no initializer.</p></li>
4096 <!-- ======================================================================= -->
4098 <a name="BasicBlock">The <tt>BasicBlock</tt> class</a>
4104 href="/doxygen/BasicBlock_8h-source.html">llvm/BasicBlock.h</a>"</tt><br>
4105 doxygen info: <a href="/doxygen/classllvm_1_1BasicBlock.html">BasicBlock
4107 Superclass: <a href="#Value"><tt>Value</tt></a></p>
4109 <p>This class represents a single entry single exit section of the code,
4110 commonly known as a basic block by the compiler community. The
4111 <tt>BasicBlock</tt> class maintains a list of <a
4112 href="#Instruction"><tt>Instruction</tt></a>s, which form the body of the block.
4113 Matching the language definition, the last element of this list of instructions
4114 is always a terminator instruction (a subclass of the <a
4115 href="#TerminatorInst"><tt>TerminatorInst</tt></a> class).</p>
4117 <p>In addition to tracking the list of instructions that make up the block, the
4118 <tt>BasicBlock</tt> class also keeps track of the <a
4119 href="#Function"><tt>Function</tt></a> that it is embedded into.</p>
4121 <p>Note that <tt>BasicBlock</tt>s themselves are <a
4122 href="#Value"><tt>Value</tt></a>s, because they are referenced by instructions
4123 like branches and can go in the switch tables. <tt>BasicBlock</tt>s have type
4126 <!-- _______________________________________________________________________ -->
4128 <a name="m_BasicBlock">
4129 Important Public Members of the <tt>BasicBlock</tt> class
4136 <li><tt>BasicBlock(const std::string &Name = "", </tt><tt><a
4137 href="#Function">Function</a> *Parent = 0)</tt>
4139 <p>The <tt>BasicBlock</tt> constructor is used to create new basic blocks for
4140 insertion into a function. The constructor optionally takes a name for the new
4141 block, and a <a href="#Function"><tt>Function</tt></a> to insert it into. If
4142 the <tt>Parent</tt> parameter is specified, the new <tt>BasicBlock</tt> is
4143 automatically inserted at the end of the specified <a
4144 href="#Function"><tt>Function</tt></a>, if not specified, the BasicBlock must be
4145 manually inserted into the <a href="#Function"><tt>Function</tt></a>.</p></li>
4147 <li><tt>BasicBlock::iterator</tt> - Typedef for instruction list iterator<br>
4148 <tt>BasicBlock::const_iterator</tt> - Typedef for const_iterator.<br>
4149 <tt>begin()</tt>, <tt>end()</tt>, <tt>front()</tt>, <tt>back()</tt>,
4150 <tt>size()</tt>, <tt>empty()</tt>
4151 STL-style functions for accessing the instruction list.
4153 <p>These methods and typedefs are forwarding functions that have the same
4154 semantics as the standard library methods of the same names. These methods
4155 expose the underlying instruction list of a basic block in a way that is easy to
4156 manipulate. To get the full complement of container operations (including
4157 operations to update the list), you must use the <tt>getInstList()</tt>
4160 <li><tt>BasicBlock::InstListType &getInstList()</tt>
4162 <p>This method is used to get access to the underlying container that actually
4163 holds the Instructions. This method must be used when there isn't a forwarding
4164 function in the <tt>BasicBlock</tt> class for the operation that you would like
4165 to perform. Because there are no forwarding functions for "updating"
4166 operations, you need to use this if you want to update the contents of a
4167 <tt>BasicBlock</tt>.</p></li>
4169 <li><tt><a href="#Function">Function</a> *getParent()</tt>
4171 <p> Returns a pointer to <a href="#Function"><tt>Function</tt></a> the block is
4172 embedded into, or a null pointer if it is homeless.</p></li>
4174 <li><tt><a href="#TerminatorInst">TerminatorInst</a> *getTerminator()</tt>
4176 <p> Returns a pointer to the terminator instruction that appears at the end of
4177 the <tt>BasicBlock</tt>. If there is no terminator instruction, or if the last
4178 instruction in the block is not a terminator, then a null pointer is
4187 <!-- ======================================================================= -->
4189 <a name="Argument">The <tt>Argument</tt> class</a>
4194 <p>This subclass of Value defines the interface for incoming formal
4195 arguments to a function. A Function maintains a list of its formal
4196 arguments. An argument has a pointer to the parent Function.</p>
4202 <!-- *********************************************************************** -->
4205 <a href="http://jigsaw.w3.org/css-validator/check/referer"><img
4206 src="http://jigsaw.w3.org/css-validator/images/vcss-blue" alt="Valid CSS"></a>
4207 <a href="http://validator.w3.org/check/referer"><img
4208 src="http://www.w3.org/Icons/valid-html401" alt="Valid HTML 4.01 Strict"></a>
4210 <a href="mailto:dhurjati@cs.uiuc.edu">Dinakar Dhurjati</a> and
4211 <a href="mailto:sabre@nondot.org">Chris Lattner</a><br>
4212 <a href="http://llvm.org/">The LLVM Compiler Infrastructure</a><br>
4213 Last modified: $Date$