1 <!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.01//EN"
2 "http://www.w3.org/TR/html4/strict.dtd">
5 <title>LLVM Programmer's Manual</title>
6 <link rel="stylesheet" href="llvm.css" type="text/css">
10 <div class="doc_title">
11 LLVM Programmer's Manual
15 <li><a href="#introduction">Introduction</a></li>
16 <li><a href="#general">General Information</a>
18 <li><a href="#stl">The C++ Standard Template Library</a></li>
20 <li>The <tt>-time-passes</tt> option</li>
21 <li>How to use the LLVM Makefile system</li>
22 <li>How to write a regression test</li>
27 <li><a href="#apis">Important and useful LLVM APIs</a>
29 <li><a href="#isa">The <tt>isa<></tt>, <tt>cast<></tt>
30 and <tt>dyn_cast<></tt> templates</a> </li>
31 <li><a href="#DEBUG">The <tt>DEBUG()</tt> macro and <tt>-debug</tt>
34 <li><a href="#DEBUG_TYPE">Fine grained debug info with <tt>DEBUG_TYPE</tt>
35 and the <tt>-debug-only</tt> option</a> </li>
38 <li><a href="#Statistic">The <tt>Statistic</tt> class & <tt>-stats</tt>
41 <li>The <tt>InstVisitor</tt> template
42 <li>The general graph API
44 <li><a href="#ViewGraph">Viewing graphs while debugging code</a></li>
47 <li><a href="#datastructure">Picking the Right Data Structure for a Task</a>
49 <li><a href="#ds_sequential">Sequential Containers (std::vector, std::list, etc)</a>
51 <li><a href="#dss_fixedarrays">Fixed Size Arrays</a></li>
52 <li><a href="#dss_heaparrays">Heap Allocated Arrays</a></li>
53 <li><a href="#dss_smallvector">"llvm/ADT/SmallVector.h"</a></li>
54 <li><a href="#dss_vector"><vector></a></li>
55 <li><a href="#dss_deque"><deque></a></li>
56 <li><a href="#dss_list"><list></a></li>
57 <li><a href="#dss_ilist">llvm/ADT/ilist</a></li>
58 <li><a href="#dss_other">Other Sequential Container Options</a></li>
60 <li><a href="#ds_set">Set-Like Containers (std::set, SmallSet, SetVector, etc)</a>
62 <li><a href="#dss_sortedvectorset">A sorted 'vector'</a></li>
63 <li><a href="#dss_smallset">"llvm/ADT/SmallSet.h"</a></li>
64 <li><a href="#dss_smallptrset">"llvm/ADT/SmallPtrSet.h"</a></li>
65 <li><a href="#dss_FoldingSet">"llvm/ADT/FoldingSet.h"</a></li>
66 <li><a href="#dss_set"><set></a></li>
67 <li><a href="#dss_setvector">"llvm/ADT/SetVector.h"</a></li>
68 <li><a href="#dss_uniquevector">"llvm/ADT/UniqueVector.h"</a></li>
69 <li><a href="#dss_otherset">Other Set-Like ContainerOptions</a></li>
71 <li><a href="#ds_map">Map-Like Containers (std::map, DenseMap, etc)</a>
73 <li><a href="#dss_sortedvectormap">A sorted 'vector'</a></li>
74 <li><a href="#dss_stringmap">"llvm/ADT/StringMap.h"</a></li>
75 <li><a href="#dss_indexedmap">"llvm/ADT/IndexedMap.h"</a></li>
76 <li><a href="#dss_densemap">"llvm/ADT/DenseMap.h"</a></li>
77 <li><a href="#dss_map"><map></a></li>
78 <li><a href="#dss_othermap">Other Map-Like Container Options</a></li>
82 <li><a href="#common">Helpful Hints for Common Operations</a>
84 <li><a href="#inspection">Basic Inspection and Traversal Routines</a>
86 <li><a href="#iterate_function">Iterating over the <tt>BasicBlock</tt>s
87 in a <tt>Function</tt></a> </li>
88 <li><a href="#iterate_basicblock">Iterating over the <tt>Instruction</tt>s
89 in a <tt>BasicBlock</tt></a> </li>
90 <li><a href="#iterate_institer">Iterating over the <tt>Instruction</tt>s
91 in a <tt>Function</tt></a> </li>
92 <li><a href="#iterate_convert">Turning an iterator into a
93 class pointer</a> </li>
94 <li><a href="#iterate_complex">Finding call sites: a more
95 complex example</a> </li>
96 <li><a href="#calls_and_invokes">Treating calls and invokes
97 the same way</a> </li>
98 <li><a href="#iterate_chains">Iterating over def-use &
99 use-def chains</a> </li>
102 <li><a href="#simplechanges">Making simple changes</a>
104 <li><a href="#schanges_creating">Creating and inserting new
105 <tt>Instruction</tt>s</a> </li>
106 <li><a href="#schanges_deleting">Deleting <tt>Instruction</tt>s</a> </li>
107 <li><a href="#schanges_replacing">Replacing an <tt>Instruction</tt>
108 with another <tt>Value</tt></a> </li>
109 <li><a href="#schanges_deletingGV">Deleting <tt>GlobalVariable</tt>s</a> </li>
113 <li>Working with the Control Flow Graph
115 <li>Accessing predecessors and successors of a <tt>BasicBlock</tt>
123 <li><a href="#advanced">Advanced Topics</a>
125 <li><a href="#TypeResolve">LLVM Type Resolution</a>
127 <li><a href="#BuildRecType">Basic Recursive Type Construction</a></li>
128 <li><a href="#refineAbstractTypeTo">The <tt>refineAbstractTypeTo</tt> method</a></li>
129 <li><a href="#PATypeHolder">The PATypeHolder Class</a></li>
130 <li><a href="#AbstractTypeUser">The AbstractTypeUser Class</a></li>
133 <li><a href="#SymbolTable">The <tt>ValueSymbolTable</tt> and <tt>TypeSymbolTable</tt> classes </a></li>
136 <li><a href="#coreclasses">The Core LLVM Class Hierarchy Reference</a>
138 <li><a href="#Type">The <tt>Type</tt> class</a> </li>
139 <li><a href="#Module">The <tt>Module</tt> class</a></li>
140 <li><a href="#Value">The <tt>Value</tt> class</a>
142 <li><a href="#User">The <tt>User</tt> class</a>
144 <li><a href="#Instruction">The <tt>Instruction</tt> class</a></li>
145 <li><a href="#Constant">The <tt>Constant</tt> class</a>
147 <li><a href="#GlobalValue">The <tt>GlobalValue</tt> class</a>
149 <li><a href="#Function">The <tt>Function</tt> class</a></li>
150 <li><a href="#GlobalVariable">The <tt>GlobalVariable</tt> class</a></li>
157 <li><a href="#BasicBlock">The <tt>BasicBlock</tt> class</a></li>
158 <li><a href="#Argument">The <tt>Argument</tt> class</a></li>
165 <div class="doc_author">
166 <p>Written by <a href="mailto:sabre@nondot.org">Chris Lattner</a>,
167 <a href="mailto:dhurjati@cs.uiuc.edu">Dinakar Dhurjati</a>,
168 <a href="mailto:jstanley@cs.uiuc.edu">Joel Stanley</a>, and
169 <a href="mailto:rspencer@x10sys.com">Reid Spencer</a></p>
172 <!-- *********************************************************************** -->
173 <div class="doc_section">
174 <a name="introduction">Introduction </a>
176 <!-- *********************************************************************** -->
178 <div class="doc_text">
180 <p>This document is meant to highlight some of the important classes and
181 interfaces available in the LLVM source-base. This manual is not
182 intended to explain what LLVM is, how it works, and what LLVM code looks
183 like. It assumes that you know the basics of LLVM and are interested
184 in writing transformations or otherwise analyzing or manipulating the
187 <p>This document should get you oriented so that you can find your
188 way in the continuously growing source code that makes up the LLVM
189 infrastructure. Note that this manual is not intended to serve as a
190 replacement for reading the source code, so if you think there should be
191 a method in one of these classes to do something, but it's not listed,
192 check the source. Links to the <a href="/doxygen/">doxygen</a> sources
193 are provided to make this as easy as possible.</p>
195 <p>The first section of this document describes general information that is
196 useful to know when working in the LLVM infrastructure, and the second describes
197 the Core LLVM classes. In the future this manual will be extended with
198 information describing how to use extension libraries, such as dominator
199 information, CFG traversal routines, and useful utilities like the <tt><a
200 href="/doxygen/InstVisitor_8h-source.html">InstVisitor</a></tt> template.</p>
204 <!-- *********************************************************************** -->
205 <div class="doc_section">
206 <a name="general">General Information</a>
208 <!-- *********************************************************************** -->
210 <div class="doc_text">
212 <p>This section contains general information that is useful if you are working
213 in the LLVM source-base, but that isn't specific to any particular API.</p>
217 <!-- ======================================================================= -->
218 <div class="doc_subsection">
219 <a name="stl">The C++ Standard Template Library</a>
222 <div class="doc_text">
224 <p>LLVM makes heavy use of the C++ Standard Template Library (STL),
225 perhaps much more than you are used to, or have seen before. Because of
226 this, you might want to do a little background reading in the
227 techniques used and capabilities of the library. There are many good
228 pages that discuss the STL, and several books on the subject that you
229 can get, so it will not be discussed in this document.</p>
231 <p>Here are some useful links:</p>
235 <li><a href="http://www.dinkumware.com/refxcpp.html">Dinkumware C++ Library
236 reference</a> - an excellent reference for the STL and other parts of the
237 standard C++ library.</li>
239 <li><a href="http://www.tempest-sw.com/cpp/">C++ In a Nutshell</a> - This is an
240 O'Reilly book in the making. It has a decent
242 Reference that rivals Dinkumware's, and is unfortunately no longer free since the book has been
245 <li><a href="http://www.parashift.com/c++-faq-lite/">C++ Frequently Asked
248 <li><a href="http://www.sgi.com/tech/stl/">SGI's STL Programmer's Guide</a> -
250 href="http://www.sgi.com/tech/stl/stl_introduction.html">Introduction to the
253 <li><a href="http://www.research.att.com/%7Ebs/C++.html">Bjarne Stroustrup's C++
256 <li><a href="http://64.78.49.204/">
257 Bruce Eckel's Thinking in C++, 2nd ed. Volume 2 Revision 4.0 (even better, get
262 <p>You are also encouraged to take a look at the <a
263 href="CodingStandards.html">LLVM Coding Standards</a> guide which focuses on how
264 to write maintainable code more than where to put your curly braces.</p>
268 <!-- ======================================================================= -->
269 <div class="doc_subsection">
270 <a name="stl">Other useful references</a>
273 <div class="doc_text">
276 <li><a href="http://www.psc.edu/%7Esemke/cvs_branches.html">CVS
277 Branch and Tag Primer</a></li>
278 <li><a href="http://www.fortran-2000.com/ArnaudRecipes/sharedlib.html">Using
279 static and shared libraries across platforms</a></li>
284 <!-- *********************************************************************** -->
285 <div class="doc_section">
286 <a name="apis">Important and useful LLVM APIs</a>
288 <!-- *********************************************************************** -->
290 <div class="doc_text">
292 <p>Here we highlight some LLVM APIs that are generally useful and good to
293 know about when writing transformations.</p>
297 <!-- ======================================================================= -->
298 <div class="doc_subsection">
299 <a name="isa">The <tt>isa<></tt>, <tt>cast<></tt> and
300 <tt>dyn_cast<></tt> templates</a>
303 <div class="doc_text">
305 <p>The LLVM source-base makes extensive use of a custom form of RTTI.
306 These templates have many similarities to the C++ <tt>dynamic_cast<></tt>
307 operator, but they don't have some drawbacks (primarily stemming from
308 the fact that <tt>dynamic_cast<></tt> only works on classes that
309 have a v-table). Because they are used so often, you must know what they
310 do and how they work. All of these templates are defined in the <a
311 href="/doxygen/Casting_8h-source.html"><tt>llvm/Support/Casting.h</tt></a>
312 file (note that you very rarely have to include this file directly).</p>
315 <dt><tt>isa<></tt>: </dt>
317 <dd><p>The <tt>isa<></tt> operator works exactly like the Java
318 "<tt>instanceof</tt>" operator. It returns true or false depending on whether
319 a reference or pointer points to an instance of the specified class. This can
320 be very useful for constraint checking of various sorts (example below).</p>
323 <dt><tt>cast<></tt>: </dt>
325 <dd><p>The <tt>cast<></tt> operator is a "checked cast" operation. It
326 converts a pointer or reference from a base class to a derived cast, causing
327 an assertion failure if it is not really an instance of the right type. This
328 should be used in cases where you have some information that makes you believe
329 that something is of the right type. An example of the <tt>isa<></tt>
330 and <tt>cast<></tt> template is:</p>
332 <div class="doc_code">
334 static bool isLoopInvariant(const <a href="#Value">Value</a> *V, const Loop *L) {
335 if (isa<<a href="#Constant">Constant</a>>(V) || isa<<a href="#Argument">Argument</a>>(V) || isa<<a href="#GlobalValue">GlobalValue</a>>(V))
338 // <i>Otherwise, it must be an instruction...</i>
339 return !L->contains(cast<<a href="#Instruction">Instruction</a>>(V)->getParent());
344 <p>Note that you should <b>not</b> use an <tt>isa<></tt> test followed
345 by a <tt>cast<></tt>, for that use the <tt>dyn_cast<></tt>
350 <dt><tt>dyn_cast<></tt>:</dt>
352 <dd><p>The <tt>dyn_cast<></tt> operator is a "checking cast" operation.
353 It checks to see if the operand is of the specified type, and if so, returns a
354 pointer to it (this operator does not work with references). If the operand is
355 not of the correct type, a null pointer is returned. Thus, this works very
356 much like the <tt>dynamic_cast<></tt> operator in C++, and should be
357 used in the same circumstances. Typically, the <tt>dyn_cast<></tt>
358 operator is used in an <tt>if</tt> statement or some other flow control
359 statement like this:</p>
361 <div class="doc_code">
363 if (<a href="#AllocationInst">AllocationInst</a> *AI = dyn_cast<<a href="#AllocationInst">AllocationInst</a>>(Val)) {
369 <p>This form of the <tt>if</tt> statement effectively combines together a call
370 to <tt>isa<></tt> and a call to <tt>cast<></tt> into one
371 statement, which is very convenient.</p>
373 <p>Note that the <tt>dyn_cast<></tt> operator, like C++'s
374 <tt>dynamic_cast<></tt> or Java's <tt>instanceof</tt> operator, can be
375 abused. In particular, you should not use big chained <tt>if/then/else</tt>
376 blocks to check for lots of different variants of classes. If you find
377 yourself wanting to do this, it is much cleaner and more efficient to use the
378 <tt>InstVisitor</tt> class to dispatch over the instruction type directly.</p>
382 <dt><tt>cast_or_null<></tt>: </dt>
384 <dd><p>The <tt>cast_or_null<></tt> operator works just like the
385 <tt>cast<></tt> operator, except that it allows for a null pointer as an
386 argument (which it then propagates). This can sometimes be useful, allowing
387 you to combine several null checks into one.</p></dd>
389 <dt><tt>dyn_cast_or_null<></tt>: </dt>
391 <dd><p>The <tt>dyn_cast_or_null<></tt> operator works just like the
392 <tt>dyn_cast<></tt> operator, except that it allows for a null pointer
393 as an argument (which it then propagates). This can sometimes be useful,
394 allowing you to combine several null checks into one.</p></dd>
398 <p>These five templates can be used with any classes, whether they have a
399 v-table or not. To add support for these templates, you simply need to add
400 <tt>classof</tt> static methods to the class you are interested casting
401 to. Describing this is currently outside the scope of this document, but there
402 are lots of examples in the LLVM source base.</p>
406 <!-- ======================================================================= -->
407 <div class="doc_subsection">
408 <a name="DEBUG">The <tt>DEBUG()</tt> macro and <tt>-debug</tt> option</a>
411 <div class="doc_text">
413 <p>Often when working on your pass you will put a bunch of debugging printouts
414 and other code into your pass. After you get it working, you want to remove
415 it, but you may need it again in the future (to work out new bugs that you run
418 <p> Naturally, because of this, you don't want to delete the debug printouts,
419 but you don't want them to always be noisy. A standard compromise is to comment
420 them out, allowing you to enable them if you need them in the future.</p>
422 <p>The "<tt><a href="/doxygen/Debug_8h-source.html">llvm/Support/Debug.h</a></tt>"
423 file provides a macro named <tt>DEBUG()</tt> that is a much nicer solution to
424 this problem. Basically, you can put arbitrary code into the argument of the
425 <tt>DEBUG</tt> macro, and it is only executed if '<tt>opt</tt>' (or any other
426 tool) is run with the '<tt>-debug</tt>' command line argument:</p>
428 <div class="doc_code">
430 DOUT << "I am here!\n";
434 <p>Then you can run your pass like this:</p>
436 <div class="doc_code">
438 $ opt < a.bc > /dev/null -mypass
439 <i><no output></i>
440 $ opt < a.bc > /dev/null -mypass -debug
445 <p>Using the <tt>DEBUG()</tt> macro instead of a home-brewed solution allows you
446 to not have to create "yet another" command line option for the debug output for
447 your pass. Note that <tt>DEBUG()</tt> macros are disabled for optimized builds,
448 so they do not cause a performance impact at all (for the same reason, they
449 should also not contain side-effects!).</p>
451 <p>One additional nice thing about the <tt>DEBUG()</tt> macro is that you can
452 enable or disable it directly in gdb. Just use "<tt>set DebugFlag=0</tt>" or
453 "<tt>set DebugFlag=1</tt>" from the gdb if the program is running. If the
454 program hasn't been started yet, you can always just run it with
459 <!-- _______________________________________________________________________ -->
460 <div class="doc_subsubsection">
461 <a name="DEBUG_TYPE">Fine grained debug info with <tt>DEBUG_TYPE</tt> and
462 the <tt>-debug-only</tt> option</a>
465 <div class="doc_text">
467 <p>Sometimes you may find yourself in a situation where enabling <tt>-debug</tt>
468 just turns on <b>too much</b> information (such as when working on the code
469 generator). If you want to enable debug information with more fine-grained
470 control, you define the <tt>DEBUG_TYPE</tt> macro and the <tt>-debug</tt> only
471 option as follows:</p>
473 <div class="doc_code">
475 DOUT << "No debug type\n";
477 #define DEBUG_TYPE "foo"
478 DOUT << "'foo' debug type\n";
480 #define DEBUG_TYPE "bar"
481 DOUT << "'bar' debug type\n";
483 #define DEBUG_TYPE ""
484 DOUT << "No debug type (2)\n";
488 <p>Then you can run your pass like this:</p>
490 <div class="doc_code">
492 $ opt < a.bc > /dev/null -mypass
493 <i><no output></i>
494 $ opt < a.bc > /dev/null -mypass -debug
499 $ opt < a.bc > /dev/null -mypass -debug-only=foo
501 $ opt < a.bc > /dev/null -mypass -debug-only=bar
506 <p>Of course, in practice, you should only set <tt>DEBUG_TYPE</tt> at the top of
507 a file, to specify the debug type for the entire module (if you do this before
508 you <tt>#include "llvm/Support/Debug.h"</tt>, you don't have to insert the ugly
509 <tt>#undef</tt>'s). Also, you should use names more meaningful than "foo" and
510 "bar", because there is no system in place to ensure that names do not
511 conflict. If two different modules use the same string, they will all be turned
512 on when the name is specified. This allows, for example, all debug information
513 for instruction scheduling to be enabled with <tt>-debug-type=InstrSched</tt>,
514 even if the source lives in multiple files.</p>
518 <!-- ======================================================================= -->
519 <div class="doc_subsection">
520 <a name="Statistic">The <tt>Statistic</tt> class & <tt>-stats</tt>
524 <div class="doc_text">
527 href="/doxygen/Statistic_8h-source.html">llvm/ADT/Statistic.h</a></tt>" file
528 provides a class named <tt>Statistic</tt> that is used as a unified way to
529 keep track of what the LLVM compiler is doing and how effective various
530 optimizations are. It is useful to see what optimizations are contributing to
531 making a particular program run faster.</p>
533 <p>Often you may run your pass on some big program, and you're interested to see
534 how many times it makes a certain transformation. Although you can do this with
535 hand inspection, or some ad-hoc method, this is a real pain and not very useful
536 for big programs. Using the <tt>Statistic</tt> class makes it very easy to
537 keep track of this information, and the calculated information is presented in a
538 uniform manner with the rest of the passes being executed.</p>
540 <p>There are many examples of <tt>Statistic</tt> uses, but the basics of using
541 it are as follows:</p>
544 <li><p>Define your statistic like this:</p>
546 <div class="doc_code">
548 #define <a href="#DEBUG_TYPE">DEBUG_TYPE</a> "mypassname" <i>// This goes before any #includes.</i>
549 STATISTIC(NumXForms, "The # of times I did stuff");
553 <p>The <tt>STATISTIC</tt> macro defines a static variable, whose name is
554 specified by the first argument. The pass name is taken from the DEBUG_TYPE
555 macro, and the description is taken from the second argument. The variable
556 defined ("NumXForms" in this case) acts like an unsigned integer.</p></li>
558 <li><p>Whenever you make a transformation, bump the counter:</p>
560 <div class="doc_code">
562 ++NumXForms; // <i>I did stuff!</i>
569 <p>That's all you have to do. To get '<tt>opt</tt>' to print out the
570 statistics gathered, use the '<tt>-stats</tt>' option:</p>
572 <div class="doc_code">
574 $ opt -stats -mypassname < program.bc > /dev/null
575 <i>... statistics output ...</i>
579 <p> When running <tt>opt</tt> on a C file from the SPEC benchmark
580 suite, it gives a report that looks like this:</p>
582 <div class="doc_code">
584 7646 bitcodewriter - Number of normal instructions
585 725 bitcodewriter - Number of oversized instructions
586 129996 bitcodewriter - Number of bitcode bytes written
587 2817 raise - Number of insts DCEd or constprop'd
588 3213 raise - Number of cast-of-self removed
589 5046 raise - Number of expression trees converted
590 75 raise - Number of other getelementptr's formed
591 138 raise - Number of load/store peepholes
592 42 deadtypeelim - Number of unused typenames removed from symtab
593 392 funcresolve - Number of varargs functions resolved
594 27 globaldce - Number of global variables removed
595 2 adce - Number of basic blocks removed
596 134 cee - Number of branches revectored
597 49 cee - Number of setcc instruction eliminated
598 532 gcse - Number of loads removed
599 2919 gcse - Number of instructions removed
600 86 indvars - Number of canonical indvars added
601 87 indvars - Number of aux indvars removed
602 25 instcombine - Number of dead inst eliminate
603 434 instcombine - Number of insts combined
604 248 licm - Number of load insts hoisted
605 1298 licm - Number of insts hoisted to a loop pre-header
606 3 licm - Number of insts hoisted to multiple loop preds (bad, no loop pre-header)
607 75 mem2reg - Number of alloca's promoted
608 1444 cfgsimplify - Number of blocks simplified
612 <p>Obviously, with so many optimizations, having a unified framework for this
613 stuff is very nice. Making your pass fit well into the framework makes it more
614 maintainable and useful.</p>
618 <!-- ======================================================================= -->
619 <div class="doc_subsection">
620 <a name="ViewGraph">Viewing graphs while debugging code</a>
623 <div class="doc_text">
625 <p>Several of the important data structures in LLVM are graphs: for example
626 CFGs made out of LLVM <a href="#BasicBlock">BasicBlock</a>s, CFGs made out of
627 LLVM <a href="CodeGenerator.html#machinebasicblock">MachineBasicBlock</a>s, and
628 <a href="CodeGenerator.html#selectiondag_intro">Instruction Selection
629 DAGs</a>. In many cases, while debugging various parts of the compiler, it is
630 nice to instantly visualize these graphs.</p>
632 <p>LLVM provides several callbacks that are available in a debug build to do
633 exactly that. If you call the <tt>Function::viewCFG()</tt> method, for example,
634 the current LLVM tool will pop up a window containing the CFG for the function
635 where each basic block is a node in the graph, and each node contains the
636 instructions in the block. Similarly, there also exists
637 <tt>Function::viewCFGOnly()</tt> (does not include the instructions), the
638 <tt>MachineFunction::viewCFG()</tt> and <tt>MachineFunction::viewCFGOnly()</tt>,
639 and the <tt>SelectionDAG::viewGraph()</tt> methods. Within GDB, for example,
640 you can usually use something like <tt>call DAG.viewGraph()</tt> to pop
641 up a window. Alternatively, you can sprinkle calls to these functions in your
642 code in places you want to debug.</p>
644 <p>Getting this to work requires a small amount of configuration. On Unix
645 systems with X11, install the <a href="http://www.graphviz.org">graphviz</a>
646 toolkit, and make sure 'dot' and 'gv' are in your path. If you are running on
647 Mac OS/X, download and install the Mac OS/X <a
648 href="http://www.pixelglow.com/graphviz/">Graphviz program</a>, and add
649 <tt>/Applications/Graphviz.app/Contents/MacOS/</tt> (or wherever you install
650 it) to your path. Once in your system and path are set up, rerun the LLVM
651 configure script and rebuild LLVM to enable this functionality.</p>
653 <p><tt>SelectionDAG</tt> has been extended to make it easier to locate
654 <i>interesting</i> nodes in large complex graphs. From gdb, if you
655 <tt>call DAG.setGraphColor(<i>node</i>, "<i>color</i>")</tt>, then the
656 next <tt>call DAG.viewGraph()</tt> would highlight the node in the
657 specified color (choices of colors can be found at <a
658 href="http://www.graphviz.org/doc/info/colors.html">colors</a>.) More
659 complex node attributes can be provided with <tt>call
660 DAG.setGraphAttrs(<i>node</i>, "<i>attributes</i>")</tt> (choices can be
661 found at <a href="http://www.graphviz.org/doc/info/attrs.html">Graph
662 Attributes</a>.) If you want to restart and clear all the current graph
663 attributes, then you can <tt>call DAG.clearGraphAttrs()</tt>. </p>
667 <!-- *********************************************************************** -->
668 <div class="doc_section">
669 <a name="datastructure">Picking the Right Data Structure for a Task</a>
671 <!-- *********************************************************************** -->
673 <div class="doc_text">
675 <p>LLVM has a plethora of data structures in the <tt>llvm/ADT/</tt> directory,
676 and we commonly use STL data structures. This section describes the trade-offs
677 you should consider when you pick one.</p>
680 The first step is a choose your own adventure: do you want a sequential
681 container, a set-like container, or a map-like container? The most important
682 thing when choosing a container is the algorithmic properties of how you plan to
683 access the container. Based on that, you should use:</p>
686 <li>a <a href="#ds_map">map-like</a> container if you need efficient look-up
687 of an value based on another value. Map-like containers also support
688 efficient queries for containment (whether a key is in the map). Map-like
689 containers generally do not support efficient reverse mapping (values to
690 keys). If you need that, use two maps. Some map-like containers also
691 support efficient iteration through the keys in sorted order. Map-like
692 containers are the most expensive sort, only use them if you need one of
693 these capabilities.</li>
695 <li>a <a href="#ds_set">set-like</a> container if you need to put a bunch of
696 stuff into a container that automatically eliminates duplicates. Some
697 set-like containers support efficient iteration through the elements in
698 sorted order. Set-like containers are more expensive than sequential
702 <li>a <a href="#ds_sequential">sequential</a> container provides
703 the most efficient way to add elements and keeps track of the order they are
704 added to the collection. They permit duplicates and support efficient
705 iteration, but do not support efficient look-up based on a key.
711 Once the proper category of container is determined, you can fine tune the
712 memory use, constant factors, and cache behaviors of access by intelligently
713 picking a member of the category. Note that constant factors and cache behavior
714 can be a big deal. If you have a vector that usually only contains a few
715 elements (but could contain many), for example, it's much better to use
716 <a href="#dss_smallvector">SmallVector</a> than <a href="#dss_vector">vector</a>
717 . Doing so avoids (relatively) expensive malloc/free calls, which dwarf the
718 cost of adding the elements to the container. </p>
722 <!-- ======================================================================= -->
723 <div class="doc_subsection">
724 <a name="ds_sequential">Sequential Containers (std::vector, std::list, etc)</a>
727 <div class="doc_text">
728 There are a variety of sequential containers available for you, based on your
729 needs. Pick the first in this section that will do what you want.
732 <!-- _______________________________________________________________________ -->
733 <div class="doc_subsubsection">
734 <a name="dss_fixedarrays">Fixed Size Arrays</a>
737 <div class="doc_text">
738 <p>Fixed size arrays are very simple and very fast. They are good if you know
739 exactly how many elements you have, or you have a (low) upper bound on how many
743 <!-- _______________________________________________________________________ -->
744 <div class="doc_subsubsection">
745 <a name="dss_heaparrays">Heap Allocated Arrays</a>
748 <div class="doc_text">
749 <p>Heap allocated arrays (new[] + delete[]) are also simple. They are good if
750 the number of elements is variable, if you know how many elements you will need
751 before the array is allocated, and if the array is usually large (if not,
752 consider a <a href="#dss_smallvector">SmallVector</a>). The cost of a heap
753 allocated array is the cost of the new/delete (aka malloc/free). Also note that
754 if you are allocating an array of a type with a constructor, the constructor and
755 destructors will be run for every element in the array (re-sizable vectors only
756 construct those elements actually used).</p>
759 <!-- _______________________________________________________________________ -->
760 <div class="doc_subsubsection">
761 <a name="dss_smallvector">"llvm/ADT/SmallVector.h"</a>
764 <div class="doc_text">
765 <p><tt>SmallVector<Type, N></tt> is a simple class that looks and smells
766 just like <tt>vector<Type></tt>:
767 it supports efficient iteration, lays out elements in memory order (so you can
768 do pointer arithmetic between elements), supports efficient push_back/pop_back
769 operations, supports efficient random access to its elements, etc.</p>
771 <p>The advantage of SmallVector is that it allocates space for
772 some number of elements (N) <b>in the object itself</b>. Because of this, if
773 the SmallVector is dynamically smaller than N, no malloc is performed. This can
774 be a big win in cases where the malloc/free call is far more expensive than the
775 code that fiddles around with the elements.</p>
777 <p>This is good for vectors that are "usually small" (e.g. the number of
778 predecessors/successors of a block is usually less than 8). On the other hand,
779 this makes the size of the SmallVector itself large, so you don't want to
780 allocate lots of them (doing so will waste a lot of space). As such,
781 SmallVectors are most useful when on the stack.</p>
783 <p>SmallVector also provides a nice portable and efficient replacement for
788 <!-- _______________________________________________________________________ -->
789 <div class="doc_subsubsection">
790 <a name="dss_vector"><vector></a>
793 <div class="doc_text">
795 std::vector is well loved and respected. It is useful when SmallVector isn't:
796 when the size of the vector is often large (thus the small optimization will
797 rarely be a benefit) or if you will be allocating many instances of the vector
798 itself (which would waste space for elements that aren't in the container).
799 vector is also useful when interfacing with code that expects vectors :).
802 <p>One worthwhile note about std::vector: avoid code like this:</p>
804 <div class="doc_code">
807 std::vector<foo> V;
813 <p>Instead, write this as:</p>
815 <div class="doc_code">
817 std::vector<foo> V;
825 <p>Doing so will save (at least) one heap allocation and free per iteration of
830 <!-- _______________________________________________________________________ -->
831 <div class="doc_subsubsection">
832 <a name="dss_deque"><deque></a>
835 <div class="doc_text">
836 <p>std::deque is, in some senses, a generalized version of std::vector. Like
837 std::vector, it provides constant time random access and other similar
838 properties, but it also provides efficient access to the front of the list. It
839 does not guarantee continuity of elements within memory.</p>
841 <p>In exchange for this extra flexibility, std::deque has significantly higher
842 constant factor costs than std::vector. If possible, use std::vector or
843 something cheaper.</p>
846 <!-- _______________________________________________________________________ -->
847 <div class="doc_subsubsection">
848 <a name="dss_list"><list></a>
851 <div class="doc_text">
852 <p>std::list is an extremely inefficient class that is rarely useful.
853 It performs a heap allocation for every element inserted into it, thus having an
854 extremely high constant factor, particularly for small data types. std::list
855 also only supports bidirectional iteration, not random access iteration.</p>
857 <p>In exchange for this high cost, std::list supports efficient access to both
858 ends of the list (like std::deque, but unlike std::vector or SmallVector). In
859 addition, the iterator invalidation characteristics of std::list are stronger
860 than that of a vector class: inserting or removing an element into the list does
861 not invalidate iterator or pointers to other elements in the list.</p>
864 <!-- _______________________________________________________________________ -->
865 <div class="doc_subsubsection">
866 <a name="dss_ilist">llvm/ADT/ilist</a>
869 <div class="doc_text">
870 <p><tt>ilist<T></tt> implements an 'intrusive' doubly-linked list. It is
871 intrusive, because it requires the element to store and provide access to the
872 prev/next pointers for the list.</p>
874 <p>ilist has the same drawbacks as std::list, and additionally requires an
875 ilist_traits implementation for the element type, but it provides some novel
876 characteristics. In particular, it can efficiently store polymorphic objects,
877 the traits class is informed when an element is inserted or removed from the
878 list, and ilists are guaranteed to support a constant-time splice operation.
881 <p>These properties are exactly what we want for things like Instructions and
882 basic blocks, which is why these are implemented with ilists.</p>
885 <!-- _______________________________________________________________________ -->
886 <div class="doc_subsubsection">
887 <a name="dss_other">Other Sequential Container options</a>
890 <div class="doc_text">
891 <p>Other STL containers are available, such as std::string.</p>
893 <p>There are also various STL adapter classes such as std::queue,
894 std::priority_queue, std::stack, etc. These provide simplified access to an
895 underlying container but don't affect the cost of the container itself.</p>
900 <!-- ======================================================================= -->
901 <div class="doc_subsection">
902 <a name="ds_set">Set-Like Containers (std::set, SmallSet, SetVector, etc)</a>
905 <div class="doc_text">
907 <p>Set-like containers are useful when you need to canonicalize multiple values
908 into a single representation. There are several different choices for how to do
909 this, providing various trade-offs.</p>
914 <!-- _______________________________________________________________________ -->
915 <div class="doc_subsubsection">
916 <a name="dss_sortedvectorset">A sorted 'vector'</a>
919 <div class="doc_text">
921 <p>If you intend to insert a lot of elements, then do a lot of queries, a
922 great approach is to use a vector (or other sequential container) with
923 std::sort+std::unique to remove duplicates. This approach works really well if
924 your usage pattern has these two distinct phases (insert then query), and can be
925 coupled with a good choice of <a href="#ds_sequential">sequential container</a>.
929 This combination provides the several nice properties: the result data is
930 contiguous in memory (good for cache locality), has few allocations, is easy to
931 address (iterators in the final vector are just indices or pointers), and can be
932 efficiently queried with a standard binary or radix search.</p>
936 <!-- _______________________________________________________________________ -->
937 <div class="doc_subsubsection">
938 <a name="dss_smallset">"llvm/ADT/SmallSet.h"</a>
941 <div class="doc_text">
943 <p>If you have a set-like data structure that is usually small and whose elements
944 are reasonably small, a <tt>SmallSet<Type, N></tt> is a good choice. This set
945 has space for N elements in place (thus, if the set is dynamically smaller than
946 N, no malloc traffic is required) and accesses them with a simple linear search.
947 When the set grows beyond 'N' elements, it allocates a more expensive representation that
948 guarantees efficient access (for most types, it falls back to std::set, but for
949 pointers it uses something far better, <a
950 href="#dss_smallptrset">SmallPtrSet</a>).</p>
952 <p>The magic of this class is that it handles small sets extremely efficiently,
953 but gracefully handles extremely large sets without loss of efficiency. The
954 drawback is that the interface is quite small: it supports insertion, queries
955 and erasing, but does not support iteration.</p>
959 <!-- _______________________________________________________________________ -->
960 <div class="doc_subsubsection">
961 <a name="dss_smallptrset">"llvm/ADT/SmallPtrSet.h"</a>
964 <div class="doc_text">
966 <p>SmallPtrSet has all the advantages of SmallSet (and a SmallSet of pointers is
967 transparently implemented with a SmallPtrSet), but also supports iterators. If
968 more than 'N' insertions are performed, a single quadratically
969 probed hash table is allocated and grows as needed, providing extremely
970 efficient access (constant time insertion/deleting/queries with low constant
971 factors) and is very stingy with malloc traffic.</p>
973 <p>Note that, unlike std::set, the iterators of SmallPtrSet are invalidated
974 whenever an insertion occurs. Also, the values visited by the iterators are not
975 visited in sorted order.</p>
979 <!-- _______________________________________________________________________ -->
980 <div class="doc_subsubsection">
981 <a name="dss_FoldingSet">"llvm/ADT/FoldingSet.h"</a>
984 <div class="doc_text">
987 FoldingSet is an aggregate class that is really good at uniquing
988 expensive-to-create or polymorphic objects. It is a combination of a chained
989 hash table with intrusive links (uniqued objects are required to inherit from
990 FoldingSetNode) that uses <a href="#dss_smallvector">SmallVector</a> as part of
993 <p>Consider a case where you want to implement a "getOrCreateFoo" method for
994 a complex object (for example, a node in the code generator). The client has a
995 description of *what* it wants to generate (it knows the opcode and all the
996 operands), but we don't want to 'new' a node, then try inserting it into a set
997 only to find out it already exists, at which point we would have to delete it
998 and return the node that already exists.
1001 <p>To support this style of client, FoldingSet perform a query with a
1002 FoldingSetNodeID (which wraps SmallVector) that can be used to describe the
1003 element that we want to query for. The query either returns the element
1004 matching the ID or it returns an opaque ID that indicates where insertion should
1005 take place. Construction of the ID usually does not require heap traffic.</p>
1007 <p>Because FoldingSet uses intrusive links, it can support polymorphic objects
1008 in the set (for example, you can have SDNode instances mixed with LoadSDNodes).
1009 Because the elements are individually allocated, pointers to the elements are
1010 stable: inserting or removing elements does not invalidate any pointers to other
1016 <!-- _______________________________________________________________________ -->
1017 <div class="doc_subsubsection">
1018 <a name="dss_set"><set></a>
1021 <div class="doc_text">
1023 <p><tt>std::set</tt> is a reasonable all-around set class, which is decent at
1024 many things but great at nothing. std::set allocates memory for each element
1025 inserted (thus it is very malloc intensive) and typically stores three pointers
1026 per element in the set (thus adding a large amount of per-element space
1027 overhead). It offers guaranteed log(n) performance, which is not particularly
1028 fast from a complexity standpoint (particularly if the elements of the set are
1029 expensive to compare, like strings), and has extremely high constant factors for
1030 lookup, insertion and removal.</p>
1032 <p>The advantages of std::set are that its iterators are stable (deleting or
1033 inserting an element from the set does not affect iterators or pointers to other
1034 elements) and that iteration over the set is guaranteed to be in sorted order.
1035 If the elements in the set are large, then the relative overhead of the pointers
1036 and malloc traffic is not a big deal, but if the elements of the set are small,
1037 std::set is almost never a good choice.</p>
1041 <!-- _______________________________________________________________________ -->
1042 <div class="doc_subsubsection">
1043 <a name="dss_setvector">"llvm/ADT/SetVector.h"</a>
1046 <div class="doc_text">
1047 <p>LLVM's SetVector<Type> is an adapter class that combines your choice of
1048 a set-like container along with a <a href="#ds_sequential">Sequential
1049 Container</a>. The important property
1050 that this provides is efficient insertion with uniquing (duplicate elements are
1051 ignored) with iteration support. It implements this by inserting elements into
1052 both a set-like container and the sequential container, using the set-like
1053 container for uniquing and the sequential container for iteration.
1056 <p>The difference between SetVector and other sets is that the order of
1057 iteration is guaranteed to match the order of insertion into the SetVector.
1058 This property is really important for things like sets of pointers. Because
1059 pointer values are non-deterministic (e.g. vary across runs of the program on
1060 different machines), iterating over the pointers in the set will
1061 not be in a well-defined order.</p>
1064 The drawback of SetVector is that it requires twice as much space as a normal
1065 set and has the sum of constant factors from the set-like container and the
1066 sequential container that it uses. Use it *only* if you need to iterate over
1067 the elements in a deterministic order. SetVector is also expensive to delete
1068 elements out of (linear time), unless you use it's "pop_back" method, which is
1072 <p>SetVector is an adapter class that defaults to using std::vector and std::set
1073 for the underlying containers, so it is quite expensive. However,
1074 <tt>"llvm/ADT/SetVector.h"</tt> also provides a SmallSetVector class, which
1075 defaults to using a SmallVector and SmallSet of a specified size. If you use
1076 this, and if your sets are dynamically smaller than N, you will save a lot of
1081 <!-- _______________________________________________________________________ -->
1082 <div class="doc_subsubsection">
1083 <a name="dss_uniquevector">"llvm/ADT/UniqueVector.h"</a>
1086 <div class="doc_text">
1089 UniqueVector is similar to <a href="#dss_setvector">SetVector</a>, but it
1090 retains a unique ID for each element inserted into the set. It internally
1091 contains a map and a vector, and it assigns a unique ID for each value inserted
1094 <p>UniqueVector is very expensive: its cost is the sum of the cost of
1095 maintaining both the map and vector, it has high complexity, high constant
1096 factors, and produces a lot of malloc traffic. It should be avoided.</p>
1101 <!-- _______________________________________________________________________ -->
1102 <div class="doc_subsubsection">
1103 <a name="dss_otherset">Other Set-Like Container Options</a>
1106 <div class="doc_text">
1109 The STL provides several other options, such as std::multiset and the various
1110 "hash_set" like containers (whether from C++ TR1 or from the SGI library).</p>
1112 <p>std::multiset is useful if you're not interested in elimination of
1113 duplicates, but has all the drawbacks of std::set. A sorted vector (where you
1114 don't delete duplicate entries) or some other approach is almost always
1117 <p>The various hash_set implementations (exposed portably by
1118 "llvm/ADT/hash_set") is a simple chained hashtable. This algorithm is as malloc
1119 intensive as std::set (performing an allocation for each element inserted,
1120 thus having really high constant factors) but (usually) provides O(1)
1121 insertion/deletion of elements. This can be useful if your elements are large
1122 (thus making the constant-factor cost relatively low) or if comparisons are
1123 expensive. Element iteration does not visit elements in a useful order.</p>
1127 <!-- ======================================================================= -->
1128 <div class="doc_subsection">
1129 <a name="ds_map">Map-Like Containers (std::map, DenseMap, etc)</a>
1132 <div class="doc_text">
1133 Map-like containers are useful when you want to associate data to a key. As
1134 usual, there are a lot of different ways to do this. :)
1137 <!-- _______________________________________________________________________ -->
1138 <div class="doc_subsubsection">
1139 <a name="dss_sortedvectormap">A sorted 'vector'</a>
1142 <div class="doc_text">
1145 If your usage pattern follows a strict insert-then-query approach, you can
1146 trivially use the same approach as <a href="#dss_sortedvectorset">sorted vectors
1147 for set-like containers</a>. The only difference is that your query function
1148 (which uses std::lower_bound to get efficient log(n) lookup) should only compare
1149 the key, not both the key and value. This yields the same advantages as sorted
1154 <!-- _______________________________________________________________________ -->
1155 <div class="doc_subsubsection">
1156 <a name="dss_stringmap">"llvm/ADT/StringMap.h"</a>
1159 <div class="doc_text">
1162 Strings are commonly used as keys in maps, and they are difficult to support
1163 efficiently: they are variable length, inefficient to hash and compare when
1164 long, expensive to copy, etc. StringMap is a specialized container designed to
1165 cope with these issues. It supports mapping an arbitrary range of bytes to an
1166 arbitrary other object.</p>
1168 <p>The StringMap implementation uses a quadratically-probed hash table, where
1169 the buckets store a pointer to the heap allocated entries (and some other
1170 stuff). The entries in the map must be heap allocated because the strings are
1171 variable length. The string data (key) and the element object (value) are
1172 stored in the same allocation with the string data immediately after the element
1173 object. This container guarantees the "<tt>(char*)(&Value+1)</tt>" points
1174 to the key string for a value.</p>
1176 <p>The StringMap is very fast for several reasons: quadratic probing is very
1177 cache efficient for lookups, the hash value of strings in buckets is not
1178 recomputed when lookup up an element, StringMap rarely has to touch the
1179 memory for unrelated objects when looking up a value (even when hash collisions
1180 happen), hash table growth does not recompute the hash values for strings
1181 already in the table, and each pair in the map is store in a single allocation
1182 (the string data is stored in the same allocation as the Value of a pair).</p>
1184 <p>StringMap also provides query methods that take byte ranges, so it only ever
1185 copies a string if a value is inserted into the table.</p>
1188 <!-- _______________________________________________________________________ -->
1189 <div class="doc_subsubsection">
1190 <a name="dss_indexedmap">"llvm/ADT/IndexedMap.h"</a>
1193 <div class="doc_text">
1195 IndexedMap is a specialized container for mapping small dense integers (or
1196 values that can be mapped to small dense integers) to some other type. It is
1197 internally implemented as a vector with a mapping function that maps the keys to
1198 the dense integer range.
1202 This is useful for cases like virtual registers in the LLVM code generator: they
1203 have a dense mapping that is offset by a compile-time constant (the first
1204 virtual register ID).</p>
1208 <!-- _______________________________________________________________________ -->
1209 <div class="doc_subsubsection">
1210 <a name="dss_densemap">"llvm/ADT/DenseMap.h"</a>
1213 <div class="doc_text">
1216 DenseMap is a simple quadratically probed hash table. It excels at supporting
1217 small keys and values: it uses a single allocation to hold all of the pairs that
1218 are currently inserted in the map. DenseMap is a great way to map pointers to
1219 pointers, or map other small types to each other.
1223 There are several aspects of DenseMap that you should be aware of, however. The
1224 iterators in a densemap are invalidated whenever an insertion occurs, unlike
1225 map. Also, because DenseMap allocates space for a large number of key/value
1226 pairs (it starts with 64 by default), it will waste a lot of space if your keys
1227 or values are large. Finally, you must implement a partial specialization of
1228 DenseMapKeyInfo for the key that you want, if it isn't already supported. This
1229 is required to tell DenseMap about two special marker values (which can never be
1230 inserted into the map) that it needs internally.</p>
1234 <!-- _______________________________________________________________________ -->
1235 <div class="doc_subsubsection">
1236 <a name="dss_map"><map></a>
1239 <div class="doc_text">
1242 std::map has similar characteristics to <a href="#dss_set">std::set</a>: it uses
1243 a single allocation per pair inserted into the map, it offers log(n) lookup with
1244 an extremely large constant factor, imposes a space penalty of 3 pointers per
1245 pair in the map, etc.</p>
1247 <p>std::map is most useful when your keys or values are very large, if you need
1248 to iterate over the collection in sorted order, or if you need stable iterators
1249 into the map (i.e. they don't get invalidated if an insertion or deletion of
1250 another element takes place).</p>
1254 <!-- _______________________________________________________________________ -->
1255 <div class="doc_subsubsection">
1256 <a name="dss_othermap">Other Map-Like Container Options</a>
1259 <div class="doc_text">
1262 The STL provides several other options, such as std::multimap and the various
1263 "hash_map" like containers (whether from C++ TR1 or from the SGI library).</p>
1265 <p>std::multimap is useful if you want to map a key to multiple values, but has
1266 all the drawbacks of std::map. A sorted vector or some other approach is almost
1269 <p>The various hash_map implementations (exposed portably by
1270 "llvm/ADT/hash_map") are simple chained hash tables. This algorithm is as
1271 malloc intensive as std::map (performing an allocation for each element
1272 inserted, thus having really high constant factors) but (usually) provides O(1)
1273 insertion/deletion of elements. This can be useful if your elements are large
1274 (thus making the constant-factor cost relatively low) or if comparisons are
1275 expensive. Element iteration does not visit elements in a useful order.</p>
1280 <!-- *********************************************************************** -->
1281 <div class="doc_section">
1282 <a name="common">Helpful Hints for Common Operations</a>
1284 <!-- *********************************************************************** -->
1286 <div class="doc_text">
1288 <p>This section describes how to perform some very simple transformations of
1289 LLVM code. This is meant to give examples of common idioms used, showing the
1290 practical side of LLVM transformations. <p> Because this is a "how-to" section,
1291 you should also read about the main classes that you will be working with. The
1292 <a href="#coreclasses">Core LLVM Class Hierarchy Reference</a> contains details
1293 and descriptions of the main classes that you should know about.</p>
1297 <!-- NOTE: this section should be heavy on example code -->
1298 <!-- ======================================================================= -->
1299 <div class="doc_subsection">
1300 <a name="inspection">Basic Inspection and Traversal Routines</a>
1303 <div class="doc_text">
1305 <p>The LLVM compiler infrastructure have many different data structures that may
1306 be traversed. Following the example of the C++ standard template library, the
1307 techniques used to traverse these various data structures are all basically the
1308 same. For a enumerable sequence of values, the <tt>XXXbegin()</tt> function (or
1309 method) returns an iterator to the start of the sequence, the <tt>XXXend()</tt>
1310 function returns an iterator pointing to one past the last valid element of the
1311 sequence, and there is some <tt>XXXiterator</tt> data type that is common
1312 between the two operations.</p>
1314 <p>Because the pattern for iteration is common across many different aspects of
1315 the program representation, the standard template library algorithms may be used
1316 on them, and it is easier to remember how to iterate. First we show a few common
1317 examples of the data structures that need to be traversed. Other data
1318 structures are traversed in very similar ways.</p>
1322 <!-- _______________________________________________________________________ -->
1323 <div class="doc_subsubsection">
1324 <a name="iterate_function">Iterating over the </a><a
1325 href="#BasicBlock"><tt>BasicBlock</tt></a>s in a <a
1326 href="#Function"><tt>Function</tt></a>
1329 <div class="doc_text">
1331 <p>It's quite common to have a <tt>Function</tt> instance that you'd like to
1332 transform in some way; in particular, you'd like to manipulate its
1333 <tt>BasicBlock</tt>s. To facilitate this, you'll need to iterate over all of
1334 the <tt>BasicBlock</tt>s that constitute the <tt>Function</tt>. The following is
1335 an example that prints the name of a <tt>BasicBlock</tt> and the number of
1336 <tt>Instruction</tt>s it contains:</p>
1338 <div class="doc_code">
1340 // <i>func is a pointer to a Function instance</i>
1341 for (Function::iterator i = func->begin(), e = func->end(); i != e; ++i)
1342 // <i>Print out the name of the basic block if it has one, and then the</i>
1343 // <i>number of instructions that it contains</i>
1344 llvm::cerr << "Basic block (name=" << i->getName() << ") has "
1345 << i->size() << " instructions.\n";
1349 <p>Note that i can be used as if it were a pointer for the purposes of
1350 invoking member functions of the <tt>Instruction</tt> class. This is
1351 because the indirection operator is overloaded for the iterator
1352 classes. In the above code, the expression <tt>i->size()</tt> is
1353 exactly equivalent to <tt>(*i).size()</tt> just like you'd expect.</p>
1357 <!-- _______________________________________________________________________ -->
1358 <div class="doc_subsubsection">
1359 <a name="iterate_basicblock">Iterating over the </a><a
1360 href="#Instruction"><tt>Instruction</tt></a>s in a <a
1361 href="#BasicBlock"><tt>BasicBlock</tt></a>
1364 <div class="doc_text">
1366 <p>Just like when dealing with <tt>BasicBlock</tt>s in <tt>Function</tt>s, it's
1367 easy to iterate over the individual instructions that make up
1368 <tt>BasicBlock</tt>s. Here's a code snippet that prints out each instruction in
1369 a <tt>BasicBlock</tt>:</p>
1371 <div class="doc_code">
1373 // <i>blk is a pointer to a BasicBlock instance</i>
1374 for (BasicBlock::iterator i = blk->begin(), e = blk->end(); i != e; ++i)
1375 // <i>The next statement works since operator<<(ostream&,...)</i>
1376 // <i>is overloaded for Instruction&</i>
1377 llvm::cerr << *i << "\n";
1381 <p>However, this isn't really the best way to print out the contents of a
1382 <tt>BasicBlock</tt>! Since the ostream operators are overloaded for virtually
1383 anything you'll care about, you could have just invoked the print routine on the
1384 basic block itself: <tt>llvm::cerr << *blk << "\n";</tt>.</p>
1388 <!-- _______________________________________________________________________ -->
1389 <div class="doc_subsubsection">
1390 <a name="iterate_institer">Iterating over the </a><a
1391 href="#Instruction"><tt>Instruction</tt></a>s in a <a
1392 href="#Function"><tt>Function</tt></a>
1395 <div class="doc_text">
1397 <p>If you're finding that you commonly iterate over a <tt>Function</tt>'s
1398 <tt>BasicBlock</tt>s and then that <tt>BasicBlock</tt>'s <tt>Instruction</tt>s,
1399 <tt>InstIterator</tt> should be used instead. You'll need to include <a
1400 href="/doxygen/InstIterator_8h-source.html"><tt>llvm/Support/InstIterator.h</tt></a>,
1401 and then instantiate <tt>InstIterator</tt>s explicitly in your code. Here's a
1402 small example that shows how to dump all instructions in a function to the standard error stream:<p>
1404 <div class="doc_code">
1406 #include "<a href="/doxygen/InstIterator_8h-source.html">llvm/Support/InstIterator.h</a>"
1408 // <i>F is a pointer to a Function instance</i>
1409 for (inst_iterator i = inst_begin(F), e = inst_end(F); i != e; ++i)
1410 llvm::cerr << *i << "\n";
1414 <p>Easy, isn't it? You can also use <tt>InstIterator</tt>s to fill a
1415 work list with its initial contents. For example, if you wanted to
1416 initialize a work list to contain all instructions in a <tt>Function</tt>
1417 F, all you would need to do is something like:</p>
1419 <div class="doc_code">
1421 std::set<Instruction*> worklist;
1422 worklist.insert(inst_begin(F), inst_end(F));
1426 <p>The STL set <tt>worklist</tt> would now contain all instructions in the
1427 <tt>Function</tt> pointed to by F.</p>
1431 <!-- _______________________________________________________________________ -->
1432 <div class="doc_subsubsection">
1433 <a name="iterate_convert">Turning an iterator into a class pointer (and
1437 <div class="doc_text">
1439 <p>Sometimes, it'll be useful to grab a reference (or pointer) to a class
1440 instance when all you've got at hand is an iterator. Well, extracting
1441 a reference or a pointer from an iterator is very straight-forward.
1442 Assuming that <tt>i</tt> is a <tt>BasicBlock::iterator</tt> and <tt>j</tt>
1443 is a <tt>BasicBlock::const_iterator</tt>:</p>
1445 <div class="doc_code">
1447 Instruction& inst = *i; // <i>Grab reference to instruction reference</i>
1448 Instruction* pinst = &*i; // <i>Grab pointer to instruction reference</i>
1449 const Instruction& inst = *j;
1453 <p>However, the iterators you'll be working with in the LLVM framework are
1454 special: they will automatically convert to a ptr-to-instance type whenever they
1455 need to. Instead of dereferencing the iterator and then taking the address of
1456 the result, you can simply assign the iterator to the proper pointer type and
1457 you get the dereference and address-of operation as a result of the assignment
1458 (behind the scenes, this is a result of overloading casting mechanisms). Thus
1459 the last line of the last example,</p>
1461 <div class="doc_code">
1463 Instruction* pinst = &*i;
1467 <p>is semantically equivalent to</p>
1469 <div class="doc_code">
1471 Instruction* pinst = i;
1475 <p>It's also possible to turn a class pointer into the corresponding iterator,
1476 and this is a constant time operation (very efficient). The following code
1477 snippet illustrates use of the conversion constructors provided by LLVM
1478 iterators. By using these, you can explicitly grab the iterator of something
1479 without actually obtaining it via iteration over some structure:</p>
1481 <div class="doc_code">
1483 void printNextInstruction(Instruction* inst) {
1484 BasicBlock::iterator it(inst);
1485 ++it; // <i>After this line, it refers to the instruction after *inst</i>
1486 if (it != inst->getParent()->end()) llvm::cerr << *it << "\n";
1493 <!--_______________________________________________________________________-->
1494 <div class="doc_subsubsection">
1495 <a name="iterate_complex">Finding call sites: a slightly more complex
1499 <div class="doc_text">
1501 <p>Say that you're writing a FunctionPass and would like to count all the
1502 locations in the entire module (that is, across every <tt>Function</tt>) where a
1503 certain function (i.e., some <tt>Function</tt>*) is already in scope. As you'll
1504 learn later, you may want to use an <tt>InstVisitor</tt> to accomplish this in a
1505 much more straight-forward manner, but this example will allow us to explore how
1506 you'd do it if you didn't have <tt>InstVisitor</tt> around. In pseudo-code, this
1507 is what we want to do:</p>
1509 <div class="doc_code">
1511 initialize callCounter to zero
1512 for each Function f in the Module
1513 for each BasicBlock b in f
1514 for each Instruction i in b
1515 if (i is a CallInst and calls the given function)
1516 increment callCounter
1520 <p>And the actual code is (remember, because we're writing a
1521 <tt>FunctionPass</tt>, our <tt>FunctionPass</tt>-derived class simply has to
1522 override the <tt>runOnFunction</tt> method):</p>
1524 <div class="doc_code">
1526 Function* targetFunc = ...;
1528 class OurFunctionPass : public FunctionPass {
1530 OurFunctionPass(): callCounter(0) { }
1532 virtual runOnFunction(Function& F) {
1533 for (Function::iterator b = F.begin(), be = F.end(); b != be; ++b) {
1534 for (BasicBlock::iterator i = b->begin(); ie = b->end(); i != ie; ++i) {
1535 if (<a href="#CallInst">CallInst</a>* callInst = <a href="#isa">dyn_cast</a><<a
1536 href="#CallInst">CallInst</a>>(&*i)) {
1537 // <i>We know we've encountered a call instruction, so we</i>
1538 // <i>need to determine if it's a call to the</i>
1539 // <i>function pointed to by m_func or not</i>
1541 if (callInst->getCalledFunction() == targetFunc)
1549 unsigned callCounter;
1556 <!--_______________________________________________________________________-->
1557 <div class="doc_subsubsection">
1558 <a name="calls_and_invokes">Treating calls and invokes the same way</a>
1561 <div class="doc_text">
1563 <p>You may have noticed that the previous example was a bit oversimplified in
1564 that it did not deal with call sites generated by 'invoke' instructions. In
1565 this, and in other situations, you may find that you want to treat
1566 <tt>CallInst</tt>s and <tt>InvokeInst</tt>s the same way, even though their
1567 most-specific common base class is <tt>Instruction</tt>, which includes lots of
1568 less closely-related things. For these cases, LLVM provides a handy wrapper
1570 href="http://llvm.org/doxygen/classllvm_1_1CallSite.html"><tt>CallSite</tt></a>.
1571 It is essentially a wrapper around an <tt>Instruction</tt> pointer, with some
1572 methods that provide functionality common to <tt>CallInst</tt>s and
1573 <tt>InvokeInst</tt>s.</p>
1575 <p>This class has "value semantics": it should be passed by value, not by
1576 reference and it should not be dynamically allocated or deallocated using
1577 <tt>operator new</tt> or <tt>operator delete</tt>. It is efficiently copyable,
1578 assignable and constructable, with costs equivalents to that of a bare pointer.
1579 If you look at its definition, it has only a single pointer member.</p>
1583 <!--_______________________________________________________________________-->
1584 <div class="doc_subsubsection">
1585 <a name="iterate_chains">Iterating over def-use & use-def chains</a>
1588 <div class="doc_text">
1590 <p>Frequently, we might have an instance of the <a
1591 href="/doxygen/classllvm_1_1Value.html">Value Class</a> and we want to
1592 determine which <tt>User</tt>s use the <tt>Value</tt>. The list of all
1593 <tt>User</tt>s of a particular <tt>Value</tt> is called a <i>def-use</i> chain.
1594 For example, let's say we have a <tt>Function*</tt> named <tt>F</tt> to a
1595 particular function <tt>foo</tt>. Finding all of the instructions that
1596 <i>use</i> <tt>foo</tt> is as simple as iterating over the <i>def-use</i> chain
1599 <div class="doc_code">
1603 for (Value::use_iterator i = F->use_begin(), e = F->use_end(); i != e; ++i)
1604 if (Instruction *Inst = dyn_cast<Instruction>(*i)) {
1605 llvm::cerr << "F is used in instruction:\n";
1606 llvm::cerr << *Inst << "\n";
1611 <p>Alternately, it's common to have an instance of the <a
1612 href="/doxygen/classllvm_1_1User.html">User Class</a> and need to know what
1613 <tt>Value</tt>s are used by it. The list of all <tt>Value</tt>s used by a
1614 <tt>User</tt> is known as a <i>use-def</i> chain. Instances of class
1615 <tt>Instruction</tt> are common <tt>User</tt>s, so we might want to iterate over
1616 all of the values that a particular instruction uses (that is, the operands of
1617 the particular <tt>Instruction</tt>):</p>
1619 <div class="doc_code">
1621 Instruction* pi = ...;
1623 for (User::op_iterator i = pi->op_begin(), e = pi->op_end(); i != e; ++i) {
1631 def-use chains ("finding all users of"): Value::use_begin/use_end
1632 use-def chains ("finding all values used"): User::op_begin/op_end [op=operand]
1637 <!-- ======================================================================= -->
1638 <div class="doc_subsection">
1639 <a name="simplechanges">Making simple changes</a>
1642 <div class="doc_text">
1644 <p>There are some primitive transformation operations present in the LLVM
1645 infrastructure that are worth knowing about. When performing
1646 transformations, it's fairly common to manipulate the contents of basic
1647 blocks. This section describes some of the common methods for doing so
1648 and gives example code.</p>
1652 <!--_______________________________________________________________________-->
1653 <div class="doc_subsubsection">
1654 <a name="schanges_creating">Creating and inserting new
1655 <tt>Instruction</tt>s</a>
1658 <div class="doc_text">
1660 <p><i>Instantiating Instructions</i></p>
1662 <p>Creation of <tt>Instruction</tt>s is straight-forward: simply call the
1663 constructor for the kind of instruction to instantiate and provide the necessary
1664 parameters. For example, an <tt>AllocaInst</tt> only <i>requires</i> a
1665 (const-ptr-to) <tt>Type</tt>. Thus:</p>
1667 <div class="doc_code">
1669 AllocaInst* ai = new AllocaInst(Type::IntTy);
1673 <p>will create an <tt>AllocaInst</tt> instance that represents the allocation of
1674 one integer in the current stack frame, at run time. Each <tt>Instruction</tt>
1675 subclass is likely to have varying default parameters which change the semantics
1676 of the instruction, so refer to the <a
1677 href="/doxygen/classllvm_1_1Instruction.html">doxygen documentation for the subclass of
1678 Instruction</a> that you're interested in instantiating.</p>
1680 <p><i>Naming values</i></p>
1682 <p>It is very useful to name the values of instructions when you're able to, as
1683 this facilitates the debugging of your transformations. If you end up looking
1684 at generated LLVM machine code, you definitely want to have logical names
1685 associated with the results of instructions! By supplying a value for the
1686 <tt>Name</tt> (default) parameter of the <tt>Instruction</tt> constructor, you
1687 associate a logical name with the result of the instruction's execution at
1688 run time. For example, say that I'm writing a transformation that dynamically
1689 allocates space for an integer on the stack, and that integer is going to be
1690 used as some kind of index by some other code. To accomplish this, I place an
1691 <tt>AllocaInst</tt> at the first point in the first <tt>BasicBlock</tt> of some
1692 <tt>Function</tt>, and I'm intending to use it within the same
1693 <tt>Function</tt>. I might do:</p>
1695 <div class="doc_code">
1697 AllocaInst* pa = new AllocaInst(Type::IntTy, 0, "indexLoc");
1701 <p>where <tt>indexLoc</tt> is now the logical name of the instruction's
1702 execution value, which is a pointer to an integer on the run time stack.</p>
1704 <p><i>Inserting instructions</i></p>
1706 <p>There are essentially two ways to insert an <tt>Instruction</tt>
1707 into an existing sequence of instructions that form a <tt>BasicBlock</tt>:</p>
1710 <li>Insertion into an explicit instruction list
1712 <p>Given a <tt>BasicBlock* pb</tt>, an <tt>Instruction* pi</tt> within that
1713 <tt>BasicBlock</tt>, and a newly-created instruction we wish to insert
1714 before <tt>*pi</tt>, we do the following: </p>
1716 <div class="doc_code">
1718 BasicBlock *pb = ...;
1719 Instruction *pi = ...;
1720 Instruction *newInst = new Instruction(...);
1722 pb->getInstList().insert(pi, newInst); // <i>Inserts newInst before pi in pb</i>
1726 <p>Appending to the end of a <tt>BasicBlock</tt> is so common that
1727 the <tt>Instruction</tt> class and <tt>Instruction</tt>-derived
1728 classes provide constructors which take a pointer to a
1729 <tt>BasicBlock</tt> to be appended to. For example code that
1732 <div class="doc_code">
1734 BasicBlock *pb = ...;
1735 Instruction *newInst = new Instruction(...);
1737 pb->getInstList().push_back(newInst); // <i>Appends newInst to pb</i>
1743 <div class="doc_code">
1745 BasicBlock *pb = ...;
1746 Instruction *newInst = new Instruction(..., pb);
1750 <p>which is much cleaner, especially if you are creating
1751 long instruction streams.</p></li>
1753 <li>Insertion into an implicit instruction list
1755 <p><tt>Instruction</tt> instances that are already in <tt>BasicBlock</tt>s
1756 are implicitly associated with an existing instruction list: the instruction
1757 list of the enclosing basic block. Thus, we could have accomplished the same
1758 thing as the above code without being given a <tt>BasicBlock</tt> by doing:
1761 <div class="doc_code">
1763 Instruction *pi = ...;
1764 Instruction *newInst = new Instruction(...);
1766 pi->getParent()->getInstList().insert(pi, newInst);
1770 <p>In fact, this sequence of steps occurs so frequently that the
1771 <tt>Instruction</tt> class and <tt>Instruction</tt>-derived classes provide
1772 constructors which take (as a default parameter) a pointer to an
1773 <tt>Instruction</tt> which the newly-created <tt>Instruction</tt> should
1774 precede. That is, <tt>Instruction</tt> constructors are capable of
1775 inserting the newly-created instance into the <tt>BasicBlock</tt> of a
1776 provided instruction, immediately before that instruction. Using an
1777 <tt>Instruction</tt> constructor with a <tt>insertBefore</tt> (default)
1778 parameter, the above code becomes:</p>
1780 <div class="doc_code">
1782 Instruction* pi = ...;
1783 Instruction* newInst = new Instruction(..., pi);
1787 <p>which is much cleaner, especially if you're creating a lot of
1788 instructions and adding them to <tt>BasicBlock</tt>s.</p></li>
1793 <!--_______________________________________________________________________-->
1794 <div class="doc_subsubsection">
1795 <a name="schanges_deleting">Deleting <tt>Instruction</tt>s</a>
1798 <div class="doc_text">
1800 <p>Deleting an instruction from an existing sequence of instructions that form a
1801 <a href="#BasicBlock"><tt>BasicBlock</tt></a> is very straight-forward. First,
1802 you must have a pointer to the instruction that you wish to delete. Second, you
1803 need to obtain the pointer to that instruction's basic block. You use the
1804 pointer to the basic block to get its list of instructions and then use the
1805 erase function to remove your instruction. For example:</p>
1807 <div class="doc_code">
1809 <a href="#Instruction">Instruction</a> *I = .. ;
1810 <a href="#BasicBlock">BasicBlock</a> *BB = I->getParent();
1812 BB->getInstList().erase(I);
1818 <!--_______________________________________________________________________-->
1819 <div class="doc_subsubsection">
1820 <a name="schanges_replacing">Replacing an <tt>Instruction</tt> with another
1824 <div class="doc_text">
1826 <p><i>Replacing individual instructions</i></p>
1828 <p>Including "<a href="/doxygen/BasicBlockUtils_8h-source.html">llvm/Transforms/Utils/BasicBlockUtils.h</a>"
1829 permits use of two very useful replace functions: <tt>ReplaceInstWithValue</tt>
1830 and <tt>ReplaceInstWithInst</tt>.</p>
1832 <h4><a name="schanges_deleting">Deleting <tt>Instruction</tt>s</a></h4>
1835 <li><tt>ReplaceInstWithValue</tt>
1837 <p>This function replaces all uses (within a basic block) of a given
1838 instruction with a value, and then removes the original instruction. The
1839 following example illustrates the replacement of the result of a particular
1840 <tt>AllocaInst</tt> that allocates memory for a single integer with a null
1841 pointer to an integer.</p>
1843 <div class="doc_code">
1845 AllocaInst* instToReplace = ...;
1846 BasicBlock::iterator ii(instToReplace);
1848 ReplaceInstWithValue(instToReplace->getParent()->getInstList(), ii,
1849 Constant::getNullValue(PointerType::get(Type::IntTy)));
1852 <li><tt>ReplaceInstWithInst</tt>
1854 <p>This function replaces a particular instruction with another
1855 instruction. The following example illustrates the replacement of one
1856 <tt>AllocaInst</tt> with another.</p>
1858 <div class="doc_code">
1860 AllocaInst* instToReplace = ...;
1861 BasicBlock::iterator ii(instToReplace);
1863 ReplaceInstWithInst(instToReplace->getParent()->getInstList(), ii,
1864 new AllocaInst(Type::IntTy, 0, "ptrToReplacedInt"));
1868 <p><i>Replacing multiple uses of <tt>User</tt>s and <tt>Value</tt>s</i></p>
1870 <p>You can use <tt>Value::replaceAllUsesWith</tt> and
1871 <tt>User::replaceUsesOfWith</tt> to change more than one use at a time. See the
1872 doxygen documentation for the <a href="/doxygen/classllvm_1_1Value.html">Value Class</a>
1873 and <a href="/doxygen/classllvm_1_1User.html">User Class</a>, respectively, for more
1876 <!-- Value::replaceAllUsesWith User::replaceUsesOfWith Point out:
1877 include/llvm/Transforms/Utils/ especially BasicBlockUtils.h with:
1878 ReplaceInstWithValue, ReplaceInstWithInst -->
1882 <!--_______________________________________________________________________-->
1883 <div class="doc_subsubsection">
1884 <a name="schanges_deletingGV">Deleting <tt>GlobalVariable</tt>s</a>
1887 <div class="doc_text">
1889 <p>Deleting a global variable from a module is just as easy as deleting an
1890 Instruction. First, you must have a pointer to the global variable that you wish
1891 to delete. You use this pointer to erase it from its parent, the module.
1894 <div class="doc_code">
1896 <a href="#GlobalVariable">GlobalVariable</a> *GV = .. ;
1898 GV->eraseFromParent();
1904 <!-- *********************************************************************** -->
1905 <div class="doc_section">
1906 <a name="advanced">Advanced Topics</a>
1908 <!-- *********************************************************************** -->
1910 <div class="doc_text">
1912 This section describes some of the advanced or obscure API's that most clients
1913 do not need to be aware of. These API's tend manage the inner workings of the
1914 LLVM system, and only need to be accessed in unusual circumstances.
1918 <!-- ======================================================================= -->
1919 <div class="doc_subsection">
1920 <a name="TypeResolve">LLVM Type Resolution</a>
1923 <div class="doc_text">
1926 The LLVM type system has a very simple goal: allow clients to compare types for
1927 structural equality with a simple pointer comparison (aka a shallow compare).
1928 This goal makes clients much simpler and faster, and is used throughout the LLVM
1933 Unfortunately achieving this goal is not a simple matter. In particular,
1934 recursive types and late resolution of opaque types makes the situation very
1935 difficult to handle. Fortunately, for the most part, our implementation makes
1936 most clients able to be completely unaware of the nasty internal details. The
1937 primary case where clients are exposed to the inner workings of it are when
1938 building a recursive type. In addition to this case, the LLVM bitcode reader,
1939 assembly parser, and linker also have to be aware of the inner workings of this
1944 For our purposes below, we need three concepts. First, an "Opaque Type" is
1945 exactly as defined in the <a href="LangRef.html#t_opaque">language
1946 reference</a>. Second an "Abstract Type" is any type which includes an
1947 opaque type as part of its type graph (for example "<tt>{ opaque, i32 }</tt>").
1948 Third, a concrete type is a type that is not an abstract type (e.g. "<tt>{ i32,
1954 <!-- ______________________________________________________________________ -->
1955 <div class="doc_subsubsection">
1956 <a name="BuildRecType">Basic Recursive Type Construction</a>
1959 <div class="doc_text">
1962 Because the most common question is "how do I build a recursive type with LLVM",
1963 we answer it now and explain it as we go. Here we include enough to cause this
1964 to be emitted to an output .ll file:
1967 <div class="doc_code">
1969 %mylist = type { %mylist*, i32 }
1974 To build this, use the following LLVM APIs:
1977 <div class="doc_code">
1979 // <i>Create the initial outer struct</i>
1980 <a href="#PATypeHolder">PATypeHolder</a> StructTy = OpaqueType::get();
1981 std::vector<const Type*> Elts;
1982 Elts.push_back(PointerType::get(StructTy));
1983 Elts.push_back(Type::IntTy);
1984 StructType *NewSTy = StructType::get(Elts);
1986 // <i>At this point, NewSTy = "{ opaque*, i32 }". Tell VMCore that</i>
1987 // <i>the struct and the opaque type are actually the same.</i>
1988 cast<OpaqueType>(StructTy.get())-><a href="#refineAbstractTypeTo">refineAbstractTypeTo</a>(NewSTy);
1990 // <i>NewSTy is potentially invalidated, but StructTy (a <a href="#PATypeHolder">PATypeHolder</a>) is</i>
1991 // <i>kept up-to-date</i>
1992 NewSTy = cast<StructType>(StructTy.get());
1994 // <i>Add a name for the type to the module symbol table (optional)</i>
1995 MyModule->addTypeName("mylist", NewSTy);
2000 This code shows the basic approach used to build recursive types: build a
2001 non-recursive type using 'opaque', then use type unification to close the cycle.
2002 The type unification step is performed by the <tt><a
2003 href="#refineAbstractTypeTo">refineAbstractTypeTo</a></tt> method, which is
2004 described next. After that, we describe the <a
2005 href="#PATypeHolder">PATypeHolder class</a>.
2010 <!-- ______________________________________________________________________ -->
2011 <div class="doc_subsubsection">
2012 <a name="refineAbstractTypeTo">The <tt>refineAbstractTypeTo</tt> method</a>
2015 <div class="doc_text">
2017 The <tt>refineAbstractTypeTo</tt> method starts the type unification process.
2018 While this method is actually a member of the DerivedType class, it is most
2019 often used on OpaqueType instances. Type unification is actually a recursive
2020 process. After unification, types can become structurally isomorphic to
2021 existing types, and all duplicates are deleted (to preserve pointer equality).
2025 In the example above, the OpaqueType object is definitely deleted.
2026 Additionally, if there is an "{ \2*, i32}" type already created in the system,
2027 the pointer and struct type created are <b>also</b> deleted. Obviously whenever
2028 a type is deleted, any "Type*" pointers in the program are invalidated. As
2029 such, it is safest to avoid having <i>any</i> "Type*" pointers to abstract types
2030 live across a call to <tt>refineAbstractTypeTo</tt> (note that non-abstract
2031 types can never move or be deleted). To deal with this, the <a
2032 href="#PATypeHolder">PATypeHolder</a> class is used to maintain a stable
2033 reference to a possibly refined type, and the <a
2034 href="#AbstractTypeUser">AbstractTypeUser</a> class is used to update more
2035 complex datastructures.
2040 <!-- ______________________________________________________________________ -->
2041 <div class="doc_subsubsection">
2042 <a name="PATypeHolder">The PATypeHolder Class</a>
2045 <div class="doc_text">
2047 PATypeHolder is a form of a "smart pointer" for Type objects. When VMCore
2048 happily goes about nuking types that become isomorphic to existing types, it
2049 automatically updates all PATypeHolder objects to point to the new type. In the
2050 example above, this allows the code to maintain a pointer to the resultant
2051 resolved recursive type, even though the Type*'s are potentially invalidated.
2055 PATypeHolder is an extremely light-weight object that uses a lazy union-find
2056 implementation to update pointers. For example the pointer from a Value to its
2057 Type is maintained by PATypeHolder objects.
2062 <!-- ______________________________________________________________________ -->
2063 <div class="doc_subsubsection">
2064 <a name="AbstractTypeUser">The AbstractTypeUser Class</a>
2067 <div class="doc_text">
2070 Some data structures need more to perform more complex updates when types get
2071 resolved. To support this, a class can derive from the AbstractTypeUser class.
2073 allows it to get callbacks when certain types are resolved. To register to get
2074 callbacks for a particular type, the DerivedType::{add/remove}AbstractTypeUser
2075 methods can be called on a type. Note that these methods only work for <i>
2076 abstract</i> types. Concrete types (those that do not include any opaque
2077 objects) can never be refined.
2082 <!-- ======================================================================= -->
2083 <div class="doc_subsection">
2084 <a name="SymbolTable">The <tt>ValueSymbolTable</tt> and
2085 <tt>TypeSymbolTable</tt> classes</a>
2088 <div class="doc_text">
2089 <p>The <tt><a href="http://llvm.org/doxygen/classllvm_1_1ValueSymbolTable.html">
2090 ValueSymbolTable</a></tt> class provides a symbol table that the <a
2091 href="#Function"><tt>Function</tt></a> and <a href="#Module">
2092 <tt>Module</tt></a> classes use for naming value definitions. The symbol table
2093 can provide a name for any <a href="#Value"><tt>Value</tt></a>.
2094 The <tt><a href="http://llvm.org/doxygen/classllvm_1_1TypeSymbolTable.html">
2095 TypeSymbolTable</a></tt> class is used by the <tt>Module</tt> class to store
2096 names for types.</p>
2098 <p>Note that the <tt>SymbolTable</tt> class should not be directly accessed
2099 by most clients. It should only be used when iteration over the symbol table
2100 names themselves are required, which is very special purpose. Note that not
2102 <a href="#Value">Value</a>s have names, and those without names (i.e. they have
2103 an empty name) do not exist in the symbol table.
2106 <p>These symbol tables support iteration over the values/types in the symbol
2107 table with <tt>begin/end/iterator</tt> and supports querying to see if a
2108 specific name is in the symbol table (with <tt>lookup</tt>). The
2109 <tt>ValueSymbolTable</tt> class exposes no public mutator methods, instead,
2110 simply call <tt>setName</tt> on a value, which will autoinsert it into the
2111 appropriate symbol table. For types, use the Module::addTypeName method to
2112 insert entries into the symbol table.</p>
2118 <!-- *********************************************************************** -->
2119 <div class="doc_section">
2120 <a name="coreclasses">The Core LLVM Class Hierarchy Reference </a>
2122 <!-- *********************************************************************** -->
2124 <div class="doc_text">
2125 <p><tt>#include "<a href="/doxygen/Type_8h-source.html">llvm/Type.h</a>"</tt>
2126 <br>doxygen info: <a href="/doxygen/classllvm_1_1Type.html">Type Class</a></p>
2128 <p>The Core LLVM classes are the primary means of representing the program
2129 being inspected or transformed. The core LLVM classes are defined in
2130 header files in the <tt>include/llvm/</tt> directory, and implemented in
2131 the <tt>lib/VMCore</tt> directory.</p>
2135 <!-- ======================================================================= -->
2136 <div class="doc_subsection">
2137 <a name="Type">The <tt>Type</tt> class and Derived Types</a>
2140 <div class="doc_text">
2142 <p><tt>Type</tt> is a superclass of all type classes. Every <tt>Value</tt> has
2143 a <tt>Type</tt>. <tt>Type</tt> cannot be instantiated directly but only
2144 through its subclasses. Certain primitive types (<tt>VoidType</tt>,
2145 <tt>LabelType</tt>, <tt>FloatType</tt> and <tt>DoubleType</tt>) have hidden
2146 subclasses. They are hidden because they offer no useful functionality beyond
2147 what the <tt>Type</tt> class offers except to distinguish themselves from
2148 other subclasses of <tt>Type</tt>.</p>
2149 <p>All other types are subclasses of <tt>DerivedType</tt>. Types can be
2150 named, but this is not a requirement. There exists exactly
2151 one instance of a given shape at any one time. This allows type equality to
2152 be performed with address equality of the Type Instance. That is, given two
2153 <tt>Type*</tt> values, the types are identical if the pointers are identical.
2157 <!-- _______________________________________________________________________ -->
2158 <div class="doc_subsubsection">
2159 <a name="m_Value">Important Public Methods</a>
2162 <div class="doc_text">
2165 <li><tt>bool isInteger() const</tt>: Returns true for any integer type.</li>
2167 <li><tt>bool isFloatingPoint()</tt>: Return true if this is one of the two
2168 floating point types.</li>
2170 <li><tt>bool isAbstract()</tt>: Return true if the type is abstract (contains
2171 an OpaqueType anywhere in its definition).</li>
2173 <li><tt>bool isSized()</tt>: Return true if the type has known size. Things
2174 that don't have a size are abstract types, labels and void.</li>
2179 <!-- _______________________________________________________________________ -->
2180 <div class="doc_subsubsection">
2181 <a name="m_Value">Important Derived Types</a>
2183 <div class="doc_text">
2185 <dt><tt>IntegerType</tt></dt>
2186 <dd>Subclass of DerivedType that represents integer types of any bit width.
2187 Any bit width between <tt>IntegerType::MIN_INT_BITS</tt> (1) and
2188 <tt>IntegerType::MAX_INT_BITS</tt> (~8 million) can be represented.
2190 <li><tt>static const IntegerType* get(unsigned NumBits)</tt>: get an integer
2191 type of a specific bit width.</li>
2192 <li><tt>unsigned getBitWidth() const</tt>: Get the bit width of an integer
2196 <dt><tt>SequentialType</tt></dt>
2197 <dd>This is subclassed by ArrayType and PointerType
2199 <li><tt>const Type * getElementType() const</tt>: Returns the type of each
2200 of the elements in the sequential type. </li>
2203 <dt><tt>ArrayType</tt></dt>
2204 <dd>This is a subclass of SequentialType and defines the interface for array
2207 <li><tt>unsigned getNumElements() const</tt>: Returns the number of
2208 elements in the array. </li>
2211 <dt><tt>PointerType</tt></dt>
2212 <dd>Subclass of SequentialType for pointer types.</dd>
2213 <dt><tt>VectorType</tt></dt>
2214 <dd>Subclass of SequentialType for vector types. A
2215 vector type is similar to an ArrayType but is distinguished because it is
2216 a first class type wherease ArrayType is not. Vector types are used for
2217 vector operations and are usually small vectors of of an integer or floating
2219 <dt><tt>StructType</tt></dt>
2220 <dd>Subclass of DerivedTypes for struct types.</dd>
2221 <dt><tt><a name="FunctionType">FunctionType</a></tt></dt>
2222 <dd>Subclass of DerivedTypes for function types.
2224 <li><tt>bool isVarArg() const</tt>: Returns true if its a vararg
2226 <li><tt> const Type * getReturnType() const</tt>: Returns the
2227 return type of the function.</li>
2228 <li><tt>const Type * getParamType (unsigned i)</tt>: Returns
2229 the type of the ith parameter.</li>
2230 <li><tt> const unsigned getNumParams() const</tt>: Returns the
2231 number of formal parameters.</li>
2234 <dt><tt>OpaqueType</tt></dt>
2235 <dd>Sublcass of DerivedType for abstract types. This class
2236 defines no content and is used as a placeholder for some other type. Note
2237 that OpaqueType is used (temporarily) during type resolution for forward
2238 references of types. Once the referenced type is resolved, the OpaqueType
2239 is replaced with the actual type. OpaqueType can also be used for data
2240 abstraction. At link time opaque types can be resolved to actual types
2241 of the same name.</dd>
2247 <!-- ======================================================================= -->
2248 <div class="doc_subsection">
2249 <a name="Module">The <tt>Module</tt> class</a>
2252 <div class="doc_text">
2255 href="/doxygen/Module_8h-source.html">llvm/Module.h</a>"</tt><br> doxygen info:
2256 <a href="/doxygen/classllvm_1_1Module.html">Module Class</a></p>
2258 <p>The <tt>Module</tt> class represents the top level structure present in LLVM
2259 programs. An LLVM module is effectively either a translation unit of the
2260 original program or a combination of several translation units merged by the
2261 linker. The <tt>Module</tt> class keeps track of a list of <a
2262 href="#Function"><tt>Function</tt></a>s, a list of <a
2263 href="#GlobalVariable"><tt>GlobalVariable</tt></a>s, and a <a
2264 href="#SymbolTable"><tt>SymbolTable</tt></a>. Additionally, it contains a few
2265 helpful member functions that try to make common operations easy.</p>
2269 <!-- _______________________________________________________________________ -->
2270 <div class="doc_subsubsection">
2271 <a name="m_Module">Important Public Members of the <tt>Module</tt> class</a>
2274 <div class="doc_text">
2277 <li><tt>Module::Module(std::string name = "")</tt></li>
2280 <p>Constructing a <a href="#Module">Module</a> is easy. You can optionally
2281 provide a name for it (probably based on the name of the translation unit).</p>
2284 <li><tt>Module::iterator</tt> - Typedef for function list iterator<br>
2285 <tt>Module::const_iterator</tt> - Typedef for const_iterator.<br>
2287 <tt>begin()</tt>, <tt>end()</tt>
2288 <tt>size()</tt>, <tt>empty()</tt>
2290 <p>These are forwarding methods that make it easy to access the contents of
2291 a <tt>Module</tt> object's <a href="#Function"><tt>Function</tt></a>
2294 <li><tt>Module::FunctionListType &getFunctionList()</tt>
2296 <p> Returns the list of <a href="#Function"><tt>Function</tt></a>s. This is
2297 necessary to use when you need to update the list or perform a complex
2298 action that doesn't have a forwarding method.</p>
2300 <p><!-- Global Variable --></p></li>
2306 <li><tt>Module::global_iterator</tt> - Typedef for global variable list iterator<br>
2308 <tt>Module::const_global_iterator</tt> - Typedef for const_iterator.<br>
2310 <tt>global_begin()</tt>, <tt>global_end()</tt>
2311 <tt>global_size()</tt>, <tt>global_empty()</tt>
2313 <p> These are forwarding methods that make it easy to access the contents of
2314 a <tt>Module</tt> object's <a
2315 href="#GlobalVariable"><tt>GlobalVariable</tt></a> list.</p></li>
2317 <li><tt>Module::GlobalListType &getGlobalList()</tt>
2319 <p>Returns the list of <a
2320 href="#GlobalVariable"><tt>GlobalVariable</tt></a>s. This is necessary to
2321 use when you need to update the list or perform a complex action that
2322 doesn't have a forwarding method.</p>
2324 <p><!-- Symbol table stuff --> </p></li>
2330 <li><tt><a href="#SymbolTable">SymbolTable</a> *getSymbolTable()</tt>
2332 <p>Return a reference to the <a href="#SymbolTable"><tt>SymbolTable</tt></a>
2333 for this <tt>Module</tt>.</p>
2335 <p><!-- Convenience methods --></p></li>
2341 <li><tt><a href="#Function">Function</a> *getFunction(const std::string
2342 &Name, const <a href="#FunctionType">FunctionType</a> *Ty)</tt>
2344 <p>Look up the specified function in the <tt>Module</tt> <a
2345 href="#SymbolTable"><tt>SymbolTable</tt></a>. If it does not exist, return
2346 <tt>null</tt>.</p></li>
2348 <li><tt><a href="#Function">Function</a> *getOrInsertFunction(const
2349 std::string &Name, const <a href="#FunctionType">FunctionType</a> *T)</tt>
2351 <p>Look up the specified function in the <tt>Module</tt> <a
2352 href="#SymbolTable"><tt>SymbolTable</tt></a>. If it does not exist, add an
2353 external declaration for the function and return it.</p></li>
2355 <li><tt>std::string getTypeName(const <a href="#Type">Type</a> *Ty)</tt>
2357 <p>If there is at least one entry in the <a
2358 href="#SymbolTable"><tt>SymbolTable</tt></a> for the specified <a
2359 href="#Type"><tt>Type</tt></a>, return it. Otherwise return the empty
2362 <li><tt>bool addTypeName(const std::string &Name, const <a
2363 href="#Type">Type</a> *Ty)</tt>
2365 <p>Insert an entry in the <a href="#SymbolTable"><tt>SymbolTable</tt></a>
2366 mapping <tt>Name</tt> to <tt>Ty</tt>. If there is already an entry for this
2367 name, true is returned and the <a
2368 href="#SymbolTable"><tt>SymbolTable</tt></a> is not modified.</p></li>
2374 <!-- ======================================================================= -->
2375 <div class="doc_subsection">
2376 <a name="Value">The <tt>Value</tt> class</a>
2379 <div class="doc_text">
2381 <p><tt>#include "<a href="/doxygen/Value_8h-source.html">llvm/Value.h</a>"</tt>
2383 doxygen info: <a href="/doxygen/classllvm_1_1Value.html">Value Class</a></p>
2385 <p>The <tt>Value</tt> class is the most important class in the LLVM Source
2386 base. It represents a typed value that may be used (among other things) as an
2387 operand to an instruction. There are many different types of <tt>Value</tt>s,
2388 such as <a href="#Constant"><tt>Constant</tt></a>s,<a
2389 href="#Argument"><tt>Argument</tt></a>s. Even <a
2390 href="#Instruction"><tt>Instruction</tt></a>s and <a
2391 href="#Function"><tt>Function</tt></a>s are <tt>Value</tt>s.</p>
2393 <p>A particular <tt>Value</tt> may be used many times in the LLVM representation
2394 for a program. For example, an incoming argument to a function (represented
2395 with an instance of the <a href="#Argument">Argument</a> class) is "used" by
2396 every instruction in the function that references the argument. To keep track
2397 of this relationship, the <tt>Value</tt> class keeps a list of all of the <a
2398 href="#User"><tt>User</tt></a>s that is using it (the <a
2399 href="#User"><tt>User</tt></a> class is a base class for all nodes in the LLVM
2400 graph that can refer to <tt>Value</tt>s). This use list is how LLVM represents
2401 def-use information in the program, and is accessible through the <tt>use_</tt>*
2402 methods, shown below.</p>
2404 <p>Because LLVM is a typed representation, every LLVM <tt>Value</tt> is typed,
2405 and this <a href="#Type">Type</a> is available through the <tt>getType()</tt>
2406 method. In addition, all LLVM values can be named. The "name" of the
2407 <tt>Value</tt> is a symbolic string printed in the LLVM code:</p>
2409 <div class="doc_code">
2411 %<b>foo</b> = add i32 1, 2
2415 <p><a name="nameWarning">The name of this instruction is "foo".</a> <b>NOTE</b>
2416 that the name of any value may be missing (an empty string), so names should
2417 <b>ONLY</b> be used for debugging (making the source code easier to read,
2418 debugging printouts), they should not be used to keep track of values or map
2419 between them. For this purpose, use a <tt>std::map</tt> of pointers to the
2420 <tt>Value</tt> itself instead.</p>
2422 <p>One important aspect of LLVM is that there is no distinction between an SSA
2423 variable and the operation that produces it. Because of this, any reference to
2424 the value produced by an instruction (or the value available as an incoming
2425 argument, for example) is represented as a direct pointer to the instance of
2427 represents this value. Although this may take some getting used to, it
2428 simplifies the representation and makes it easier to manipulate.</p>
2432 <!-- _______________________________________________________________________ -->
2433 <div class="doc_subsubsection">
2434 <a name="m_Value">Important Public Members of the <tt>Value</tt> class</a>
2437 <div class="doc_text">
2440 <li><tt>Value::use_iterator</tt> - Typedef for iterator over the
2442 <tt>Value::use_const_iterator</tt> - Typedef for const_iterator over
2444 <tt>unsigned use_size()</tt> - Returns the number of users of the
2446 <tt>bool use_empty()</tt> - Returns true if there are no users.<br>
2447 <tt>use_iterator use_begin()</tt> - Get an iterator to the start of
2449 <tt>use_iterator use_end()</tt> - Get an iterator to the end of the
2451 <tt><a href="#User">User</a> *use_back()</tt> - Returns the last
2452 element in the list.
2453 <p> These methods are the interface to access the def-use
2454 information in LLVM. As with all other iterators in LLVM, the naming
2455 conventions follow the conventions defined by the <a href="#stl">STL</a>.</p>
2457 <li><tt><a href="#Type">Type</a> *getType() const</tt>
2458 <p>This method returns the Type of the Value.</p>
2460 <li><tt>bool hasName() const</tt><br>
2461 <tt>std::string getName() const</tt><br>
2462 <tt>void setName(const std::string &Name)</tt>
2463 <p> This family of methods is used to access and assign a name to a <tt>Value</tt>,
2464 be aware of the <a href="#nameWarning">precaution above</a>.</p>
2466 <li><tt>void replaceAllUsesWith(Value *V)</tt>
2468 <p>This method traverses the use list of a <tt>Value</tt> changing all <a
2469 href="#User"><tt>User</tt>s</a> of the current value to refer to
2470 "<tt>V</tt>" instead. For example, if you detect that an instruction always
2471 produces a constant value (for example through constant folding), you can
2472 replace all uses of the instruction with the constant like this:</p>
2474 <div class="doc_code">
2476 Inst->replaceAllUsesWith(ConstVal);
2484 <!-- ======================================================================= -->
2485 <div class="doc_subsection">
2486 <a name="User">The <tt>User</tt> class</a>
2489 <div class="doc_text">
2492 <tt>#include "<a href="/doxygen/User_8h-source.html">llvm/User.h</a>"</tt><br>
2493 doxygen info: <a href="/doxygen/classllvm_1_1User.html">User Class</a><br>
2494 Superclass: <a href="#Value"><tt>Value</tt></a></p>
2496 <p>The <tt>User</tt> class is the common base class of all LLVM nodes that may
2497 refer to <a href="#Value"><tt>Value</tt></a>s. It exposes a list of "Operands"
2498 that are all of the <a href="#Value"><tt>Value</tt></a>s that the User is
2499 referring to. The <tt>User</tt> class itself is a subclass of
2502 <p>The operands of a <tt>User</tt> point directly to the LLVM <a
2503 href="#Value"><tt>Value</tt></a> that it refers to. Because LLVM uses Static
2504 Single Assignment (SSA) form, there can only be one definition referred to,
2505 allowing this direct connection. This connection provides the use-def
2506 information in LLVM.</p>
2510 <!-- _______________________________________________________________________ -->
2511 <div class="doc_subsubsection">
2512 <a name="m_User">Important Public Members of the <tt>User</tt> class</a>
2515 <div class="doc_text">
2517 <p>The <tt>User</tt> class exposes the operand list in two ways: through
2518 an index access interface and through an iterator based interface.</p>
2521 <li><tt>Value *getOperand(unsigned i)</tt><br>
2522 <tt>unsigned getNumOperands()</tt>
2523 <p> These two methods expose the operands of the <tt>User</tt> in a
2524 convenient form for direct access.</p></li>
2526 <li><tt>User::op_iterator</tt> - Typedef for iterator over the operand
2528 <tt>op_iterator op_begin()</tt> - Get an iterator to the start of
2529 the operand list.<br>
2530 <tt>op_iterator op_end()</tt> - Get an iterator to the end of the
2532 <p> Together, these methods make up the iterator based interface to
2533 the operands of a <tt>User</tt>.</p></li>
2538 <!-- ======================================================================= -->
2539 <div class="doc_subsection">
2540 <a name="Instruction">The <tt>Instruction</tt> class</a>
2543 <div class="doc_text">
2545 <p><tt>#include "</tt><tt><a
2546 href="/doxygen/Instruction_8h-source.html">llvm/Instruction.h</a>"</tt><br>
2547 doxygen info: <a href="/doxygen/classllvm_1_1Instruction.html">Instruction Class</a><br>
2548 Superclasses: <a href="#User"><tt>User</tt></a>, <a
2549 href="#Value"><tt>Value</tt></a></p>
2551 <p>The <tt>Instruction</tt> class is the common base class for all LLVM
2552 instructions. It provides only a few methods, but is a very commonly used
2553 class. The primary data tracked by the <tt>Instruction</tt> class itself is the
2554 opcode (instruction type) and the parent <a
2555 href="#BasicBlock"><tt>BasicBlock</tt></a> the <tt>Instruction</tt> is embedded
2556 into. To represent a specific type of instruction, one of many subclasses of
2557 <tt>Instruction</tt> are used.</p>
2559 <p> Because the <tt>Instruction</tt> class subclasses the <a
2560 href="#User"><tt>User</tt></a> class, its operands can be accessed in the same
2561 way as for other <a href="#User"><tt>User</tt></a>s (with the
2562 <tt>getOperand()</tt>/<tt>getNumOperands()</tt> and
2563 <tt>op_begin()</tt>/<tt>op_end()</tt> methods).</p> <p> An important file for
2564 the <tt>Instruction</tt> class is the <tt>llvm/Instruction.def</tt> file. This
2565 file contains some meta-data about the various different types of instructions
2566 in LLVM. It describes the enum values that are used as opcodes (for example
2567 <tt>Instruction::Add</tt> and <tt>Instruction::ICmp</tt>), as well as the
2568 concrete sub-classes of <tt>Instruction</tt> that implement the instruction (for
2569 example <tt><a href="#BinaryOperator">BinaryOperator</a></tt> and <tt><a
2570 href="#CmpInst">CmpInst</a></tt>). Unfortunately, the use of macros in
2571 this file confuses doxygen, so these enum values don't show up correctly in the
2572 <a href="/doxygen/classllvm_1_1Instruction.html">doxygen output</a>.</p>
2576 <!-- _______________________________________________________________________ -->
2577 <div class="doc_subsubsection">
2578 <a name="s_Instruction">Important Subclasses of the <tt>Instruction</tt>
2581 <div class="doc_text">
2583 <li><tt><a name="BinaryOperator">BinaryOperator</a></tt>
2584 <p>This subclasses represents all two operand instructions whose operands
2585 must be the same type, except for the comparison instructions.</p></li>
2586 <li><tt><a name="CastInst">CastInst</a></tt>
2587 <p>This subclass is the parent of the 12 casting instructions. It provides
2588 common operations on cast instructions.</p>
2589 <li><tt><a name="CmpInst">CmpInst</a></tt>
2590 <p>This subclass respresents the two comparison instructions,
2591 <a href="LangRef.html#i_icmp">ICmpInst</a> (integer opreands), and
2592 <a href="LangRef.html#i_fcmp">FCmpInst</a> (floating point operands).</p>
2593 <li><tt><a name="TerminatorInst">TerminatorInst</a></tt>
2594 <p>This subclass is the parent of all terminator instructions (those which
2595 can terminate a block).</p>
2599 <!-- _______________________________________________________________________ -->
2600 <div class="doc_subsubsection">
2601 <a name="m_Instruction">Important Public Members of the <tt>Instruction</tt>
2605 <div class="doc_text">
2608 <li><tt><a href="#BasicBlock">BasicBlock</a> *getParent()</tt>
2609 <p>Returns the <a href="#BasicBlock"><tt>BasicBlock</tt></a> that
2610 this <tt>Instruction</tt> is embedded into.</p></li>
2611 <li><tt>bool mayWriteToMemory()</tt>
2612 <p>Returns true if the instruction writes to memory, i.e. it is a
2613 <tt>call</tt>,<tt>free</tt>,<tt>invoke</tt>, or <tt>store</tt>.</p></li>
2614 <li><tt>unsigned getOpcode()</tt>
2615 <p>Returns the opcode for the <tt>Instruction</tt>.</p></li>
2616 <li><tt><a href="#Instruction">Instruction</a> *clone() const</tt>
2617 <p>Returns another instance of the specified instruction, identical
2618 in all ways to the original except that the instruction has no parent
2619 (ie it's not embedded into a <a href="#BasicBlock"><tt>BasicBlock</tt></a>),
2620 and it has no name</p></li>
2625 <!-- ======================================================================= -->
2626 <div class="doc_subsection">
2627 <a name="Constant">The <tt>Constant</tt> class and subclasses</a>
2630 <div class="doc_text">
2632 <p>Constant represents a base class for different types of constants. It
2633 is subclassed by ConstantInt, ConstantArray, etc. for representing
2634 the various types of Constants. <a href="#GlobalValue">GlobalValue</a> is also
2635 a subclass, which represents the address of a global variable or function.
2640 <!-- _______________________________________________________________________ -->
2641 <div class="doc_subsubsection">Important Subclasses of Constant </div>
2642 <div class="doc_text">
2644 <li>ConstantInt : This subclass of Constant represents an integer constant of
2647 <li><tt>const APInt& getValue() const</tt>: Returns the underlying
2648 value of this constant, an APInt value.</li>
2649 <li><tt>int64_t getSExtValue() const</tt>: Converts the underlying APInt
2650 value to an int64_t via sign extension. If the value (not the bit width)
2651 of the APInt is too large to fit in an int64_t, an assertion will result.
2652 For this reason, use of this method is discouraged.</li>
2653 <li><tt>uint64_t getZExtValue() const</tt>: Converts the underlying APInt
2654 value to a uint64_t via zero extension. IF the value (not the bit width)
2655 of the APInt is too large to fit in a uint64_t, an assertion will result.
2656 For this reason, use of this method is discouraged.</li>
2657 <li><tt>static ConstantInt* get(const APInt& Val)</tt>: Returns the
2658 ConstantInt object that represents the value provided by <tt>Val</tt>.
2659 The type is implied as the IntegerType that corresponds to the bit width
2660 of <tt>Val</tt>.</li>
2661 <li><tt>static ConstantInt* get(const Type *Ty, uint64_t Val)</tt>:
2662 Returns the ConstantInt object that represents the value provided by
2663 <tt>Val</tt> for integer type <tt>Ty</tt>.</li>
2666 <li>ConstantFP : This class represents a floating point constant.
2668 <li><tt>double getValue() const</tt>: Returns the underlying value of
2669 this constant. </li>
2672 <li>ConstantArray : This represents a constant array.
2674 <li><tt>const std::vector<Use> &getValues() const</tt>: Returns
2675 a vector of component constants that makeup this array. </li>
2678 <li>ConstantStruct : This represents a constant struct.
2680 <li><tt>const std::vector<Use> &getValues() const</tt>: Returns
2681 a vector of component constants that makeup this array. </li>
2684 <li>GlobalValue : This represents either a global variable or a function. In
2685 either case, the value is a constant fixed address (after linking).
2691 <!-- ======================================================================= -->
2692 <div class="doc_subsection">
2693 <a name="GlobalValue">The <tt>GlobalValue</tt> class</a>
2696 <div class="doc_text">
2699 href="/doxygen/GlobalValue_8h-source.html">llvm/GlobalValue.h</a>"</tt><br>
2700 doxygen info: <a href="/doxygen/classllvm_1_1GlobalValue.html">GlobalValue
2702 Superclasses: <a href="#Constant"><tt>Constant</tt></a>,
2703 <a href="#User"><tt>User</tt></a>, <a href="#Value"><tt>Value</tt></a></p>
2705 <p>Global values (<a href="#GlobalVariable"><tt>GlobalVariable</tt></a>s or <a
2706 href="#Function"><tt>Function</tt></a>s) are the only LLVM values that are
2707 visible in the bodies of all <a href="#Function"><tt>Function</tt></a>s.
2708 Because they are visible at global scope, they are also subject to linking with
2709 other globals defined in different translation units. To control the linking
2710 process, <tt>GlobalValue</tt>s know their linkage rules. Specifically,
2711 <tt>GlobalValue</tt>s know whether they have internal or external linkage, as
2712 defined by the <tt>LinkageTypes</tt> enumeration.</p>
2714 <p>If a <tt>GlobalValue</tt> has internal linkage (equivalent to being
2715 <tt>static</tt> in C), it is not visible to code outside the current translation
2716 unit, and does not participate in linking. If it has external linkage, it is
2717 visible to external code, and does participate in linking. In addition to
2718 linkage information, <tt>GlobalValue</tt>s keep track of which <a
2719 href="#Module"><tt>Module</tt></a> they are currently part of.</p>
2721 <p>Because <tt>GlobalValue</tt>s are memory objects, they are always referred to
2722 by their <b>address</b>. As such, the <a href="#Type"><tt>Type</tt></a> of a
2723 global is always a pointer to its contents. It is important to remember this
2724 when using the <tt>GetElementPtrInst</tt> instruction because this pointer must
2725 be dereferenced first. For example, if you have a <tt>GlobalVariable</tt> (a
2726 subclass of <tt>GlobalValue)</tt> that is an array of 24 ints, type <tt>[24 x
2727 i32]</tt>, then the <tt>GlobalVariable</tt> is a pointer to that array. Although
2728 the address of the first element of this array and the value of the
2729 <tt>GlobalVariable</tt> are the same, they have different types. The
2730 <tt>GlobalVariable</tt>'s type is <tt>[24 x i32]</tt>. The first element's type
2731 is <tt>i32.</tt> Because of this, accessing a global value requires you to
2732 dereference the pointer with <tt>GetElementPtrInst</tt> first, then its elements
2733 can be accessed. This is explained in the <a href="LangRef.html#globalvars">LLVM
2734 Language Reference Manual</a>.</p>
2738 <!-- _______________________________________________________________________ -->
2739 <div class="doc_subsubsection">
2740 <a name="m_GlobalValue">Important Public Members of the <tt>GlobalValue</tt>
2744 <div class="doc_text">
2747 <li><tt>bool hasInternalLinkage() const</tt><br>
2748 <tt>bool hasExternalLinkage() const</tt><br>
2749 <tt>void setInternalLinkage(bool HasInternalLinkage)</tt>
2750 <p> These methods manipulate the linkage characteristics of the <tt>GlobalValue</tt>.</p>
2753 <li><tt><a href="#Module">Module</a> *getParent()</tt>
2754 <p> This returns the <a href="#Module"><tt>Module</tt></a> that the
2755 GlobalValue is currently embedded into.</p></li>
2760 <!-- ======================================================================= -->
2761 <div class="doc_subsection">
2762 <a name="Function">The <tt>Function</tt> class</a>
2765 <div class="doc_text">
2768 href="/doxygen/Function_8h-source.html">llvm/Function.h</a>"</tt><br> doxygen
2769 info: <a href="/doxygen/classllvm_1_1Function.html">Function Class</a><br>
2770 Superclasses: <a href="#GlobalValue"><tt>GlobalValue</tt></a>,
2771 <a href="#Constant"><tt>Constant</tt></a>,
2772 <a href="#User"><tt>User</tt></a>,
2773 <a href="#Value"><tt>Value</tt></a></p>
2775 <p>The <tt>Function</tt> class represents a single procedure in LLVM. It is
2776 actually one of the more complex classes in the LLVM heirarchy because it must
2777 keep track of a large amount of data. The <tt>Function</tt> class keeps track
2778 of a list of <a href="#BasicBlock"><tt>BasicBlock</tt></a>s, a list of formal
2779 <a href="#Argument"><tt>Argument</tt></a>s, and a
2780 <a href="#SymbolTable"><tt>SymbolTable</tt></a>.</p>
2782 <p>The list of <a href="#BasicBlock"><tt>BasicBlock</tt></a>s is the most
2783 commonly used part of <tt>Function</tt> objects. The list imposes an implicit
2784 ordering of the blocks in the function, which indicate how the code will be
2785 layed out by the backend. Additionally, the first <a
2786 href="#BasicBlock"><tt>BasicBlock</tt></a> is the implicit entry node for the
2787 <tt>Function</tt>. It is not legal in LLVM to explicitly branch to this initial
2788 block. There are no implicit exit nodes, and in fact there may be multiple exit
2789 nodes from a single <tt>Function</tt>. If the <a
2790 href="#BasicBlock"><tt>BasicBlock</tt></a> list is empty, this indicates that
2791 the <tt>Function</tt> is actually a function declaration: the actual body of the
2792 function hasn't been linked in yet.</p>
2794 <p>In addition to a list of <a href="#BasicBlock"><tt>BasicBlock</tt></a>s, the
2795 <tt>Function</tt> class also keeps track of the list of formal <a
2796 href="#Argument"><tt>Argument</tt></a>s that the function receives. This
2797 container manages the lifetime of the <a href="#Argument"><tt>Argument</tt></a>
2798 nodes, just like the <a href="#BasicBlock"><tt>BasicBlock</tt></a> list does for
2799 the <a href="#BasicBlock"><tt>BasicBlock</tt></a>s.</p>
2801 <p>The <a href="#SymbolTable"><tt>SymbolTable</tt></a> is a very rarely used
2802 LLVM feature that is only used when you have to look up a value by name. Aside
2803 from that, the <a href="#SymbolTable"><tt>SymbolTable</tt></a> is used
2804 internally to make sure that there are not conflicts between the names of <a
2805 href="#Instruction"><tt>Instruction</tt></a>s, <a
2806 href="#BasicBlock"><tt>BasicBlock</tt></a>s, or <a
2807 href="#Argument"><tt>Argument</tt></a>s in the function body.</p>
2809 <p>Note that <tt>Function</tt> is a <a href="#GlobalValue">GlobalValue</a>
2810 and therefore also a <a href="#Constant">Constant</a>. The value of the function
2811 is its address (after linking) which is guaranteed to be constant.</p>
2814 <!-- _______________________________________________________________________ -->
2815 <div class="doc_subsubsection">
2816 <a name="m_Function">Important Public Members of the <tt>Function</tt>
2820 <div class="doc_text">
2823 <li><tt>Function(const </tt><tt><a href="#FunctionType">FunctionType</a>
2824 *Ty, LinkageTypes Linkage, const std::string &N = "", Module* Parent = 0)</tt>
2826 <p>Constructor used when you need to create new <tt>Function</tt>s to add
2827 the the program. The constructor must specify the type of the function to
2828 create and what type of linkage the function should have. The <a
2829 href="#FunctionType"><tt>FunctionType</tt></a> argument
2830 specifies the formal arguments and return value for the function. The same
2831 <a href="#FunctionType"><tt>FunctionType</tt></a> value can be used to
2832 create multiple functions. The <tt>Parent</tt> argument specifies the Module
2833 in which the function is defined. If this argument is provided, the function
2834 will automatically be inserted into that module's list of
2837 <li><tt>bool isExternal()</tt>
2839 <p>Return whether or not the <tt>Function</tt> has a body defined. If the
2840 function is "external", it does not have a body, and thus must be resolved
2841 by linking with a function defined in a different translation unit.</p></li>
2843 <li><tt>Function::iterator</tt> - Typedef for basic block list iterator<br>
2844 <tt>Function::const_iterator</tt> - Typedef for const_iterator.<br>
2846 <tt>begin()</tt>, <tt>end()</tt>
2847 <tt>size()</tt>, <tt>empty()</tt>
2849 <p>These are forwarding methods that make it easy to access the contents of
2850 a <tt>Function</tt> object's <a href="#BasicBlock"><tt>BasicBlock</tt></a>
2853 <li><tt>Function::BasicBlockListType &getBasicBlockList()</tt>
2855 <p>Returns the list of <a href="#BasicBlock"><tt>BasicBlock</tt></a>s. This
2856 is necessary to use when you need to update the list or perform a complex
2857 action that doesn't have a forwarding method.</p></li>
2859 <li><tt>Function::arg_iterator</tt> - Typedef for the argument list
2861 <tt>Function::const_arg_iterator</tt> - Typedef for const_iterator.<br>
2863 <tt>arg_begin()</tt>, <tt>arg_end()</tt>
2864 <tt>arg_size()</tt>, <tt>arg_empty()</tt>
2866 <p>These are forwarding methods that make it easy to access the contents of
2867 a <tt>Function</tt> object's <a href="#Argument"><tt>Argument</tt></a>
2870 <li><tt>Function::ArgumentListType &getArgumentList()</tt>
2872 <p>Returns the list of <a href="#Argument"><tt>Argument</tt></a>s. This is
2873 necessary to use when you need to update the list or perform a complex
2874 action that doesn't have a forwarding method.</p></li>
2876 <li><tt><a href="#BasicBlock">BasicBlock</a> &getEntryBlock()</tt>
2878 <p>Returns the entry <a href="#BasicBlock"><tt>BasicBlock</tt></a> for the
2879 function. Because the entry block for the function is always the first
2880 block, this returns the first block of the <tt>Function</tt>.</p></li>
2882 <li><tt><a href="#Type">Type</a> *getReturnType()</tt><br>
2883 <tt><a href="#FunctionType">FunctionType</a> *getFunctionType()</tt>
2885 <p>This traverses the <a href="#Type"><tt>Type</tt></a> of the
2886 <tt>Function</tt> and returns the return type of the function, or the <a
2887 href="#FunctionType"><tt>FunctionType</tt></a> of the actual
2890 <li><tt><a href="#SymbolTable">SymbolTable</a> *getSymbolTable()</tt>
2892 <p> Return a pointer to the <a href="#SymbolTable"><tt>SymbolTable</tt></a>
2893 for this <tt>Function</tt>.</p></li>
2898 <!-- ======================================================================= -->
2899 <div class="doc_subsection">
2900 <a name="GlobalVariable">The <tt>GlobalVariable</tt> class</a>
2903 <div class="doc_text">
2906 href="/doxygen/GlobalVariable_8h-source.html">llvm/GlobalVariable.h</a>"</tt>
2908 doxygen info: <a href="/doxygen/classllvm_1_1GlobalVariable.html">GlobalVariable
2910 Superclasses: <a href="#GlobalValue"><tt>GlobalValue</tt></a>,
2911 <a href="#Constant"><tt>Constant</tt></a>,
2912 <a href="#User"><tt>User</tt></a>,
2913 <a href="#Value"><tt>Value</tt></a></p>
2915 <p>Global variables are represented with the (suprise suprise)
2916 <tt>GlobalVariable</tt> class. Like functions, <tt>GlobalVariable</tt>s are also
2917 subclasses of <a href="#GlobalValue"><tt>GlobalValue</tt></a>, and as such are
2918 always referenced by their address (global values must live in memory, so their
2919 "name" refers to their constant address). See
2920 <a href="#GlobalValue"><tt>GlobalValue</tt></a> for more on this. Global
2921 variables may have an initial value (which must be a
2922 <a href="#Constant"><tt>Constant</tt></a>), and if they have an initializer,
2923 they may be marked as "constant" themselves (indicating that their contents
2924 never change at runtime).</p>
2927 <!-- _______________________________________________________________________ -->
2928 <div class="doc_subsubsection">
2929 <a name="m_GlobalVariable">Important Public Members of the
2930 <tt>GlobalVariable</tt> class</a>
2933 <div class="doc_text">
2936 <li><tt>GlobalVariable(const </tt><tt><a href="#Type">Type</a> *Ty, bool
2937 isConstant, LinkageTypes& Linkage, <a href="#Constant">Constant</a>
2938 *Initializer = 0, const std::string &Name = "", Module* Parent = 0)</tt>
2940 <p>Create a new global variable of the specified type. If
2941 <tt>isConstant</tt> is true then the global variable will be marked as
2942 unchanging for the program. The Linkage parameter specifies the type of
2943 linkage (internal, external, weak, linkonce, appending) for the variable. If
2944 the linkage is InternalLinkage, WeakLinkage, or LinkOnceLinkage, then
2945 the resultant global variable will have internal linkage. AppendingLinkage
2946 concatenates together all instances (in different translation units) of the
2947 variable into a single variable but is only applicable to arrays. See
2948 the <a href="LangRef.html#modulestructure">LLVM Language Reference</a> for
2949 further details on linkage types. Optionally an initializer, a name, and the
2950 module to put the variable into may be specified for the global variable as
2953 <li><tt>bool isConstant() const</tt>
2955 <p>Returns true if this is a global variable that is known not to
2956 be modified at runtime.</p></li>
2958 <li><tt>bool hasInitializer()</tt>
2960 <p>Returns true if this <tt>GlobalVariable</tt> has an intializer.</p></li>
2962 <li><tt><a href="#Constant">Constant</a> *getInitializer()</tt>
2964 <p>Returns the intial value for a <tt>GlobalVariable</tt>. It is not legal
2965 to call this method if there is no initializer.</p></li>
2971 <!-- ======================================================================= -->
2972 <div class="doc_subsection">
2973 <a name="BasicBlock">The <tt>BasicBlock</tt> class</a>
2976 <div class="doc_text">
2979 href="/doxygen/BasicBlock_8h-source.html">llvm/BasicBlock.h</a>"</tt><br>
2980 doxygen info: <a href="/doxygen/structllvm_1_1BasicBlock.html">BasicBlock
2982 Superclass: <a href="#Value"><tt>Value</tt></a></p>
2984 <p>This class represents a single entry multiple exit section of the code,
2985 commonly known as a basic block by the compiler community. The
2986 <tt>BasicBlock</tt> class maintains a list of <a
2987 href="#Instruction"><tt>Instruction</tt></a>s, which form the body of the block.
2988 Matching the language definition, the last element of this list of instructions
2989 is always a terminator instruction (a subclass of the <a
2990 href="#TerminatorInst"><tt>TerminatorInst</tt></a> class).</p>
2992 <p>In addition to tracking the list of instructions that make up the block, the
2993 <tt>BasicBlock</tt> class also keeps track of the <a
2994 href="#Function"><tt>Function</tt></a> that it is embedded into.</p>
2996 <p>Note that <tt>BasicBlock</tt>s themselves are <a
2997 href="#Value"><tt>Value</tt></a>s, because they are referenced by instructions
2998 like branches and can go in the switch tables. <tt>BasicBlock</tt>s have type
3003 <!-- _______________________________________________________________________ -->
3004 <div class="doc_subsubsection">
3005 <a name="m_BasicBlock">Important Public Members of the <tt>BasicBlock</tt>
3009 <div class="doc_text">
3012 <li><tt>BasicBlock(const std::string &Name = "", </tt><tt><a
3013 href="#Function">Function</a> *Parent = 0)</tt>
3015 <p>The <tt>BasicBlock</tt> constructor is used to create new basic blocks for
3016 insertion into a function. The constructor optionally takes a name for the new
3017 block, and a <a href="#Function"><tt>Function</tt></a> to insert it into. If
3018 the <tt>Parent</tt> parameter is specified, the new <tt>BasicBlock</tt> is
3019 automatically inserted at the end of the specified <a
3020 href="#Function"><tt>Function</tt></a>, if not specified, the BasicBlock must be
3021 manually inserted into the <a href="#Function"><tt>Function</tt></a>.</p></li>
3023 <li><tt>BasicBlock::iterator</tt> - Typedef for instruction list iterator<br>
3024 <tt>BasicBlock::const_iterator</tt> - Typedef for const_iterator.<br>
3025 <tt>begin()</tt>, <tt>end()</tt>, <tt>front()</tt>, <tt>back()</tt>,
3026 <tt>size()</tt>, <tt>empty()</tt>
3027 STL-style functions for accessing the instruction list.
3029 <p>These methods and typedefs are forwarding functions that have the same
3030 semantics as the standard library methods of the same names. These methods
3031 expose the underlying instruction list of a basic block in a way that is easy to
3032 manipulate. To get the full complement of container operations (including
3033 operations to update the list), you must use the <tt>getInstList()</tt>
3036 <li><tt>BasicBlock::InstListType &getInstList()</tt>
3038 <p>This method is used to get access to the underlying container that actually
3039 holds the Instructions. This method must be used when there isn't a forwarding
3040 function in the <tt>BasicBlock</tt> class for the operation that you would like
3041 to perform. Because there are no forwarding functions for "updating"
3042 operations, you need to use this if you want to update the contents of a
3043 <tt>BasicBlock</tt>.</p></li>
3045 <li><tt><a href="#Function">Function</a> *getParent()</tt>
3047 <p> Returns a pointer to <a href="#Function"><tt>Function</tt></a> the block is
3048 embedded into, or a null pointer if it is homeless.</p></li>
3050 <li><tt><a href="#TerminatorInst">TerminatorInst</a> *getTerminator()</tt>
3052 <p> Returns a pointer to the terminator instruction that appears at the end of
3053 the <tt>BasicBlock</tt>. If there is no terminator instruction, or if the last
3054 instruction in the block is not a terminator, then a null pointer is
3062 <!-- ======================================================================= -->
3063 <div class="doc_subsection">
3064 <a name="Argument">The <tt>Argument</tt> class</a>
3067 <div class="doc_text">
3069 <p>This subclass of Value defines the interface for incoming formal
3070 arguments to a function. A Function maintains a list of its formal
3071 arguments. An argument has a pointer to the parent Function.</p>
3075 <!-- *********************************************************************** -->
3078 <a href="http://jigsaw.w3.org/css-validator/check/referer"><img
3079 src="http://jigsaw.w3.org/css-validator/images/vcss" alt="Valid CSS!"></a>
3080 <a href="http://validator.w3.org/check/referer"><img
3081 src="http://www.w3.org/Icons/valid-html401" alt="Valid HTML 4.01!" /></a>
3083 <a href="mailto:dhurjati@cs.uiuc.edu">Dinakar Dhurjati</a> and
3084 <a href="mailto:sabre@nondot.org">Chris Lattner</a><br>
3085 <a href="http://llvm.org">The LLVM Compiler Infrastructure</a><br>
3086 Last modified: $Date$