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>
59 <li><a href="#ds_set">Set-Like Containers (std::set, SmallSet, SetVector, etc)</a>
61 <li><a href="#dss_sortedvectorset">A sorted 'vector'</a></li>
62 <li><a href="#dss_smallset">"llvm/ADT/SmallSet.h"</a></li>
63 <li><a href="#dss_smallptrset">"llvm/ADT/SmallPtrSet.h"</a></li>
64 <li><a href="#dss_FoldingSet">"llvm/ADT/FoldingSet.h"</a></li>
65 <li><a href="#dss_set"><set></a></li>
66 <li><a href="#dss_setvector">"llvm/ADT/SetVector.h"</a></li>
67 <li><a href="#dss_otherset">Other Options</a></li>
69 <li><a href="#ds_map">Map-Like Containers (std::map, DenseMap, etc)</a></li>
72 <li><a href="#common">Helpful Hints for Common Operations</a>
74 <li><a href="#inspection">Basic Inspection and Traversal Routines</a>
76 <li><a href="#iterate_function">Iterating over the <tt>BasicBlock</tt>s
77 in a <tt>Function</tt></a> </li>
78 <li><a href="#iterate_basicblock">Iterating over the <tt>Instruction</tt>s
79 in a <tt>BasicBlock</tt></a> </li>
80 <li><a href="#iterate_institer">Iterating over the <tt>Instruction</tt>s
81 in a <tt>Function</tt></a> </li>
82 <li><a href="#iterate_convert">Turning an iterator into a
83 class pointer</a> </li>
84 <li><a href="#iterate_complex">Finding call sites: a more
85 complex example</a> </li>
86 <li><a href="#calls_and_invokes">Treating calls and invokes
87 the same way</a> </li>
88 <li><a href="#iterate_chains">Iterating over def-use &
89 use-def chains</a> </li>
92 <li><a href="#simplechanges">Making simple changes</a>
94 <li><a href="#schanges_creating">Creating and inserting new
95 <tt>Instruction</tt>s</a> </li>
96 <li><a href="#schanges_deleting">Deleting <tt>Instruction</tt>s</a> </li>
97 <li><a href="#schanges_replacing">Replacing an <tt>Instruction</tt>
98 with another <tt>Value</tt></a> </li>
102 <li>Working with the Control Flow Graph
104 <li>Accessing predecessors and successors of a <tt>BasicBlock</tt>
112 <li><a href="#advanced">Advanced Topics</a>
114 <li><a href="#TypeResolve">LLVM Type Resolution</a>
116 <li><a href="#BuildRecType">Basic Recursive Type Construction</a></li>
117 <li><a href="#refineAbstractTypeTo">The <tt>refineAbstractTypeTo</tt> method</a></li>
118 <li><a href="#PATypeHolder">The PATypeHolder Class</a></li>
119 <li><a href="#AbstractTypeUser">The AbstractTypeUser Class</a></li>
122 <li><a href="#SymbolTable">The <tt>SymbolTable</tt> class </a></li>
125 <li><a href="#coreclasses">The Core LLVM Class Hierarchy Reference</a>
127 <li><a href="#Type">The <tt>Type</tt> class</a> </li>
128 <li><a href="#Value">The <tt>Value</tt> class</a>
130 <li><a href="#User">The <tt>User</tt> class</a>
132 <li><a href="#Instruction">The <tt>Instruction</tt> class</a>
134 <li><a href="#GetElementPtrInst">The <tt>GetElementPtrInst</tt> class</a></li>
137 <li><a href="#Module">The <tt>Module</tt> class</a></li>
138 <li><a href="#Constant">The <tt>Constant</tt> class</a>
140 <li><a href="#GlobalValue">The <tt>GlobalValue</tt> class</a>
142 <li><a href="#BasicBlock">The <tt>BasicBlock</tt>class</a></li>
143 <li><a href="#Function">The <tt>Function</tt> class</a></li>
144 <li><a href="#GlobalVariable">The <tt>GlobalVariable</tt> class</a></li>
151 <li><a href="#Argument">The <tt>Argument</tt> class</a></li>
158 <div class="doc_author">
159 <p>Written by <a href="mailto:sabre@nondot.org">Chris Lattner</a>,
160 <a href="mailto:dhurjati@cs.uiuc.edu">Dinakar Dhurjati</a>,
161 <a href="mailto:jstanley@cs.uiuc.edu">Joel Stanley</a>, and
162 <a href="mailto:rspencer@x10sys.com">Reid Spencer</a></p>
165 <!-- *********************************************************************** -->
166 <div class="doc_section">
167 <a name="introduction">Introduction </a>
169 <!-- *********************************************************************** -->
171 <div class="doc_text">
173 <p>This document is meant to highlight some of the important classes and
174 interfaces available in the LLVM source-base. This manual is not
175 intended to explain what LLVM is, how it works, and what LLVM code looks
176 like. It assumes that you know the basics of LLVM and are interested
177 in writing transformations or otherwise analyzing or manipulating the
180 <p>This document should get you oriented so that you can find your
181 way in the continuously growing source code that makes up the LLVM
182 infrastructure. Note that this manual is not intended to serve as a
183 replacement for reading the source code, so if you think there should be
184 a method in one of these classes to do something, but it's not listed,
185 check the source. Links to the <a href="/doxygen/">doxygen</a> sources
186 are provided to make this as easy as possible.</p>
188 <p>The first section of this document describes general information that is
189 useful to know when working in the LLVM infrastructure, and the second describes
190 the Core LLVM classes. In the future this manual will be extended with
191 information describing how to use extension libraries, such as dominator
192 information, CFG traversal routines, and useful utilities like the <tt><a
193 href="/doxygen/InstVisitor_8h-source.html">InstVisitor</a></tt> template.</p>
197 <!-- *********************************************************************** -->
198 <div class="doc_section">
199 <a name="general">General Information</a>
201 <!-- *********************************************************************** -->
203 <div class="doc_text">
205 <p>This section contains general information that is useful if you are working
206 in the LLVM source-base, but that isn't specific to any particular API.</p>
210 <!-- ======================================================================= -->
211 <div class="doc_subsection">
212 <a name="stl">The C++ Standard Template Library</a>
215 <div class="doc_text">
217 <p>LLVM makes heavy use of the C++ Standard Template Library (STL),
218 perhaps much more than you are used to, or have seen before. Because of
219 this, you might want to do a little background reading in the
220 techniques used and capabilities of the library. There are many good
221 pages that discuss the STL, and several books on the subject that you
222 can get, so it will not be discussed in this document.</p>
224 <p>Here are some useful links:</p>
228 <li><a href="http://www.dinkumware.com/refxcpp.html">Dinkumware C++ Library
229 reference</a> - an excellent reference for the STL and other parts of the
230 standard C++ library.</li>
232 <li><a href="http://www.tempest-sw.com/cpp/">C++ In a Nutshell</a> - This is an
233 O'Reilly book in the making. It has a decent
235 Reference that rivals Dinkumware's, and is unfortunately no longer free since the book has been
238 <li><a href="http://www.parashift.com/c++-faq-lite/">C++ Frequently Asked
241 <li><a href="http://www.sgi.com/tech/stl/">SGI's STL Programmer's Guide</a> -
243 href="http://www.sgi.com/tech/stl/stl_introduction.html">Introduction to the
246 <li><a href="http://www.research.att.com/%7Ebs/C++.html">Bjarne Stroustrup's C++
249 <li><a href="http://64.78.49.204/">
250 Bruce Eckel's Thinking in C++, 2nd ed. Volume 2 Revision 4.0 (even better, get
255 <p>You are also encouraged to take a look at the <a
256 href="CodingStandards.html">LLVM Coding Standards</a> guide which focuses on how
257 to write maintainable code more than where to put your curly braces.</p>
261 <!-- ======================================================================= -->
262 <div class="doc_subsection">
263 <a name="stl">Other useful references</a>
266 <div class="doc_text">
269 <li><a href="http://www.psc.edu/%7Esemke/cvs_branches.html">CVS
270 Branch and Tag Primer</a></li>
271 <li><a href="http://www.fortran-2000.com/ArnaudRecipes/sharedlib.html">Using
272 static and shared libraries across platforms</a></li>
277 <!-- *********************************************************************** -->
278 <div class="doc_section">
279 <a name="apis">Important and useful LLVM APIs</a>
281 <!-- *********************************************************************** -->
283 <div class="doc_text">
285 <p>Here we highlight some LLVM APIs that are generally useful and good to
286 know about when writing transformations.</p>
290 <!-- ======================================================================= -->
291 <div class="doc_subsection">
292 <a name="isa">The <tt>isa<></tt>, <tt>cast<></tt> and
293 <tt>dyn_cast<></tt> templates</a>
296 <div class="doc_text">
298 <p>The LLVM source-base makes extensive use of a custom form of RTTI.
299 These templates have many similarities to the C++ <tt>dynamic_cast<></tt>
300 operator, but they don't have some drawbacks (primarily stemming from
301 the fact that <tt>dynamic_cast<></tt> only works on classes that
302 have a v-table). Because they are used so often, you must know what they
303 do and how they work. All of these templates are defined in the <a
304 href="/doxygen/Casting_8h-source.html"><tt>llvm/Support/Casting.h</tt></a>
305 file (note that you very rarely have to include this file directly).</p>
308 <dt><tt>isa<></tt>: </dt>
310 <dd><p>The <tt>isa<></tt> operator works exactly like the Java
311 "<tt>instanceof</tt>" operator. It returns true or false depending on whether
312 a reference or pointer points to an instance of the specified class. This can
313 be very useful for constraint checking of various sorts (example below).</p>
316 <dt><tt>cast<></tt>: </dt>
318 <dd><p>The <tt>cast<></tt> operator is a "checked cast" operation. It
319 converts a pointer or reference from a base class to a derived cast, causing
320 an assertion failure if it is not really an instance of the right type. This
321 should be used in cases where you have some information that makes you believe
322 that something is of the right type. An example of the <tt>isa<></tt>
323 and <tt>cast<></tt> template is:</p>
325 <div class="doc_code">
327 static bool isLoopInvariant(const <a href="#Value">Value</a> *V, const Loop *L) {
328 if (isa<<a href="#Constant">Constant</a>>(V) || isa<<a href="#Argument">Argument</a>>(V) || isa<<a href="#GlobalValue">GlobalValue</a>>(V))
331 // <i>Otherwise, it must be an instruction...</i>
332 return !L->contains(cast<<a href="#Instruction">Instruction</a>>(V)->getParent());
337 <p>Note that you should <b>not</b> use an <tt>isa<></tt> test followed
338 by a <tt>cast<></tt>, for that use the <tt>dyn_cast<></tt>
343 <dt><tt>dyn_cast<></tt>:</dt>
345 <dd><p>The <tt>dyn_cast<></tt> operator is a "checking cast" operation.
346 It checks to see if the operand is of the specified type, and if so, returns a
347 pointer to it (this operator does not work with references). If the operand is
348 not of the correct type, a null pointer is returned. Thus, this works very
349 much like the <tt>dynamic_cast<></tt> operator in C++, and should be
350 used in the same circumstances. Typically, the <tt>dyn_cast<></tt>
351 operator is used in an <tt>if</tt> statement or some other flow control
352 statement like this:</p>
354 <div class="doc_code">
356 if (<a href="#AllocationInst">AllocationInst</a> *AI = dyn_cast<<a href="#AllocationInst">AllocationInst</a>>(Val)) {
362 <p>This form of the <tt>if</tt> statement effectively combines together a call
363 to <tt>isa<></tt> and a call to <tt>cast<></tt> into one
364 statement, which is very convenient.</p>
366 <p>Note that the <tt>dyn_cast<></tt> operator, like C++'s
367 <tt>dynamic_cast<></tt> or Java's <tt>instanceof</tt> operator, can be
368 abused. In particular, you should not use big chained <tt>if/then/else</tt>
369 blocks to check for lots of different variants of classes. If you find
370 yourself wanting to do this, it is much cleaner and more efficient to use the
371 <tt>InstVisitor</tt> class to dispatch over the instruction type directly.</p>
375 <dt><tt>cast_or_null<></tt>: </dt>
377 <dd><p>The <tt>cast_or_null<></tt> operator works just like the
378 <tt>cast<></tt> operator, except that it allows for a null pointer as an
379 argument (which it then propagates). This can sometimes be useful, allowing
380 you to combine several null checks into one.</p></dd>
382 <dt><tt>dyn_cast_or_null<></tt>: </dt>
384 <dd><p>The <tt>dyn_cast_or_null<></tt> operator works just like the
385 <tt>dyn_cast<></tt> operator, except that it allows for a null pointer
386 as an argument (which it then propagates). This can sometimes be useful,
387 allowing you to combine several null checks into one.</p></dd>
391 <p>These five templates can be used with any classes, whether they have a
392 v-table or not. To add support for these templates, you simply need to add
393 <tt>classof</tt> static methods to the class you are interested casting
394 to. Describing this is currently outside the scope of this document, but there
395 are lots of examples in the LLVM source base.</p>
399 <!-- ======================================================================= -->
400 <div class="doc_subsection">
401 <a name="DEBUG">The <tt>DEBUG()</tt> macro and <tt>-debug</tt> option</a>
404 <div class="doc_text">
406 <p>Often when working on your pass you will put a bunch of debugging printouts
407 and other code into your pass. After you get it working, you want to remove
408 it, but you may need it again in the future (to work out new bugs that you run
411 <p> Naturally, because of this, you don't want to delete the debug printouts,
412 but you don't want them to always be noisy. A standard compromise is to comment
413 them out, allowing you to enable them if you need them in the future.</p>
415 <p>The "<tt><a href="/doxygen/Debug_8h-source.html">llvm/Support/Debug.h</a></tt>"
416 file provides a macro named <tt>DEBUG()</tt> that is a much nicer solution to
417 this problem. Basically, you can put arbitrary code into the argument of the
418 <tt>DEBUG</tt> macro, and it is only executed if '<tt>opt</tt>' (or any other
419 tool) is run with the '<tt>-debug</tt>' command line argument:</p>
421 <div class="doc_code">
423 DOUT << "I am here!\n";
427 <p>Then you can run your pass like this:</p>
429 <div class="doc_code">
431 $ opt < a.bc > /dev/null -mypass
432 <i><no output></i>
433 $ opt < a.bc > /dev/null -mypass -debug
438 <p>Using the <tt>DEBUG()</tt> macro instead of a home-brewed solution allows you
439 to not have to create "yet another" command line option for the debug output for
440 your pass. Note that <tt>DEBUG()</tt> macros are disabled for optimized builds,
441 so they do not cause a performance impact at all (for the same reason, they
442 should also not contain side-effects!).</p>
444 <p>One additional nice thing about the <tt>DEBUG()</tt> macro is that you can
445 enable or disable it directly in gdb. Just use "<tt>set DebugFlag=0</tt>" or
446 "<tt>set DebugFlag=1</tt>" from the gdb if the program is running. If the
447 program hasn't been started yet, you can always just run it with
452 <!-- _______________________________________________________________________ -->
453 <div class="doc_subsubsection">
454 <a name="DEBUG_TYPE">Fine grained debug info with <tt>DEBUG_TYPE</tt> and
455 the <tt>-debug-only</tt> option</a>
458 <div class="doc_text">
460 <p>Sometimes you may find yourself in a situation where enabling <tt>-debug</tt>
461 just turns on <b>too much</b> information (such as when working on the code
462 generator). If you want to enable debug information with more fine-grained
463 control, you define the <tt>DEBUG_TYPE</tt> macro and the <tt>-debug</tt> only
464 option as follows:</p>
466 <div class="doc_code">
468 DOUT << "No debug type\n";
470 #define DEBUG_TYPE "foo"
471 DOUT << "'foo' debug type\n";
473 #define DEBUG_TYPE "bar"
474 DOUT << "'bar' debug type\n";
476 #define DEBUG_TYPE ""
477 DOUT << "No debug type (2)\n";
481 <p>Then you can run your pass like this:</p>
483 <div class="doc_code">
485 $ opt < a.bc > /dev/null -mypass
486 <i><no output></i>
487 $ opt < a.bc > /dev/null -mypass -debug
492 $ opt < a.bc > /dev/null -mypass -debug-only=foo
494 $ opt < a.bc > /dev/null -mypass -debug-only=bar
499 <p>Of course, in practice, you should only set <tt>DEBUG_TYPE</tt> at the top of
500 a file, to specify the debug type for the entire module (if you do this before
501 you <tt>#include "llvm/Support/Debug.h"</tt>, you don't have to insert the ugly
502 <tt>#undef</tt>'s). Also, you should use names more meaningful than "foo" and
503 "bar", because there is no system in place to ensure that names do not
504 conflict. If two different modules use the same string, they will all be turned
505 on when the name is specified. This allows, for example, all debug information
506 for instruction scheduling to be enabled with <tt>-debug-type=InstrSched</tt>,
507 even if the source lives in multiple files.</p>
511 <!-- ======================================================================= -->
512 <div class="doc_subsection">
513 <a name="Statistic">The <tt>Statistic</tt> class & <tt>-stats</tt>
517 <div class="doc_text">
520 href="/doxygen/Statistic_8h-source.html">llvm/ADT/Statistic.h</a></tt>" file
521 provides a class named <tt>Statistic</tt> that is used as a unified way to
522 keep track of what the LLVM compiler is doing and how effective various
523 optimizations are. It is useful to see what optimizations are contributing to
524 making a particular program run faster.</p>
526 <p>Often you may run your pass on some big program, and you're interested to see
527 how many times it makes a certain transformation. Although you can do this with
528 hand inspection, or some ad-hoc method, this is a real pain and not very useful
529 for big programs. Using the <tt>Statistic</tt> class makes it very easy to
530 keep track of this information, and the calculated information is presented in a
531 uniform manner with the rest of the passes being executed.</p>
533 <p>There are many examples of <tt>Statistic</tt> uses, but the basics of using
534 it are as follows:</p>
537 <li><p>Define your statistic like this:</p>
539 <div class="doc_code">
541 #define <a href="#DEBUG_TYPE">DEBUG_TYPE</a> "mypassname" <i>// This goes before any #includes.</i>
542 STATISTIC(NumXForms, "The # of times I did stuff");
546 <p>The <tt>STATISTIC</tt> macro defines a static variable, whose name is
547 specified by the first argument. The pass name is taken from the DEBUG_TYPE
548 macro, and the description is taken from the second argument. The variable
549 defined ("NumXForms" in this case) acts like an unsigned integer.</p></li>
551 <li><p>Whenever you make a transformation, bump the counter:</p>
553 <div class="doc_code">
555 ++NumXForms; // <i>I did stuff!</i>
562 <p>That's all you have to do. To get '<tt>opt</tt>' to print out the
563 statistics gathered, use the '<tt>-stats</tt>' option:</p>
565 <div class="doc_code">
567 $ opt -stats -mypassname < program.bc > /dev/null
568 <i>... statistics output ...</i>
572 <p> When running <tt>gccas</tt> on a C file from the SPEC benchmark
573 suite, it gives a report that looks like this:</p>
575 <div class="doc_code">
577 7646 bytecodewriter - Number of normal instructions
578 725 bytecodewriter - Number of oversized instructions
579 129996 bytecodewriter - Number of bytecode bytes written
580 2817 raise - Number of insts DCEd or constprop'd
581 3213 raise - Number of cast-of-self removed
582 5046 raise - Number of expression trees converted
583 75 raise - Number of other getelementptr's formed
584 138 raise - Number of load/store peepholes
585 42 deadtypeelim - Number of unused typenames removed from symtab
586 392 funcresolve - Number of varargs functions resolved
587 27 globaldce - Number of global variables removed
588 2 adce - Number of basic blocks removed
589 134 cee - Number of branches revectored
590 49 cee - Number of setcc instruction eliminated
591 532 gcse - Number of loads removed
592 2919 gcse - Number of instructions removed
593 86 indvars - Number of canonical indvars added
594 87 indvars - Number of aux indvars removed
595 25 instcombine - Number of dead inst eliminate
596 434 instcombine - Number of insts combined
597 248 licm - Number of load insts hoisted
598 1298 licm - Number of insts hoisted to a loop pre-header
599 3 licm - Number of insts hoisted to multiple loop preds (bad, no loop pre-header)
600 75 mem2reg - Number of alloca's promoted
601 1444 cfgsimplify - Number of blocks simplified
605 <p>Obviously, with so many optimizations, having a unified framework for this
606 stuff is very nice. Making your pass fit well into the framework makes it more
607 maintainable and useful.</p>
611 <!-- ======================================================================= -->
612 <div class="doc_subsection">
613 <a name="ViewGraph">Viewing graphs while debugging code</a>
616 <div class="doc_text">
618 <p>Several of the important data structures in LLVM are graphs: for example
619 CFGs made out of LLVM <a href="#BasicBlock">BasicBlock</a>s, CFGs made out of
620 LLVM <a href="CodeGenerator.html#machinebasicblock">MachineBasicBlock</a>s, and
621 <a href="CodeGenerator.html#selectiondag_intro">Instruction Selection
622 DAGs</a>. In many cases, while debugging various parts of the compiler, it is
623 nice to instantly visualize these graphs.</p>
625 <p>LLVM provides several callbacks that are available in a debug build to do
626 exactly that. If you call the <tt>Function::viewCFG()</tt> method, for example,
627 the current LLVM tool will pop up a window containing the CFG for the function
628 where each basic block is a node in the graph, and each node contains the
629 instructions in the block. Similarly, there also exists
630 <tt>Function::viewCFGOnly()</tt> (does not include the instructions), the
631 <tt>MachineFunction::viewCFG()</tt> and <tt>MachineFunction::viewCFGOnly()</tt>,
632 and the <tt>SelectionDAG::viewGraph()</tt> methods. Within GDB, for example,
633 you can usually use something like <tt>call DAG.viewGraph()</tt> to pop
634 up a window. Alternatively, you can sprinkle calls to these functions in your
635 code in places you want to debug.</p>
637 <p>Getting this to work requires a small amount of configuration. On Unix
638 systems with X11, install the <a href="http://www.graphviz.org">graphviz</a>
639 toolkit, and make sure 'dot' and 'gv' are in your path. If you are running on
640 Mac OS/X, download and install the Mac OS/X <a
641 href="http://www.pixelglow.com/graphviz/">Graphviz program</a>, and add
642 <tt>/Applications/Graphviz.app/Contents/MacOS/</tt> (or whereever you install
643 it) to your path. Once in your system and path are set up, rerun the LLVM
644 configure script and rebuild LLVM to enable this functionality.</p>
646 <p><tt>SelectionDAG</tt> has been extended to make it easier to locate
647 <i>interesting</i> nodes in large complex graphs. From gdb, if you
648 <tt>call DAG.setGraphColor(<i>node</i>, "<i>color</i>")</tt>, then the
649 next <tt>call DAG.viewGraph()</tt> would hilight the node in the
650 specified color (choices of colors can be found at <a
651 href="http://www.graphviz.org/doc/info/colors.html">colors</a>.) More
652 complex node attributes can be provided with <tt>call
653 DAG.setGraphAttrs(<i>node</i>, "<i>attributes</i>")</tt> (choices can be
654 found at <a href="http://www.graphviz.org/doc/info/attrs.html">Graph
655 Attributes</a>.) If you want to restart and clear all the current graph
656 attributes, then you can <tt>call DAG.clearGraphAttrs()</tt>. </p>
660 <!-- *********************************************************************** -->
661 <div class="doc_section">
662 <a name="datastructure">Picking the Right Data Structure for a Task</a>
664 <!-- *********************************************************************** -->
666 <div class="doc_text">
668 <p>LLVM has a plethora of datastructures in the <tt>llvm/ADT/</tt> directory,
669 and we commonly use STL datastructures. This section describes the tradeoffs
670 you should consider when you pick one.</p>
673 The first step is a choose your own adventure: do you want a sequential
674 container, a set-like container, or a map-like container? The most important
675 thing when choosing a container is the algorithmic properties of how you plan to
676 access the container. Based on that, you should use:</p>
679 <li>a <a href="#ds_map">map-like</a> container if you need efficient lookup
680 of an value based on another value. Map-like containers also support
681 efficient queries for containment (whether a key is in the map). Map-like
682 containers generally do not support efficient reverse mapping (values to
683 keys). If you need that, use two maps. Some map-like containers also
684 support efficient iteration through the keys in sorted order. Map-like
685 containers are the most expensive sort, only use them if you need one of
686 these capabilities.</li>
688 <li>a <a href="#ds_set">set-like</a> container if you need to put a bunch of
689 stuff into a container that automatically eliminates duplicates. Some
690 set-like containers support efficient iteration through the elements in
691 sorted order. Set-like containers are more expensive than sequential
695 <li>a <a href="#ds_sequential">sequential</a> container provides
696 the most efficient way to add elements and keeps track of the order they are
697 added to the collection. They permit duplicates and support efficient
698 iteration, but do not support efficient lookup based on a key.
704 Once the proper catagory of container is determined, you can fine tune the
705 memory use, constant factors, and cache behaviors of access by intelligently
706 picking a member of the catagory. Note that constant factors and cache behavior
707 can be a big deal. If you have a vector that usually only contains a few
708 elements (but could contain many), for example, it's much better to use
709 <a href="#dss_smallvector">SmallVector</a> than <a href="#dss_vector">vector</a>
710 . Doing so avoids (relatively) expensive malloc/free calls, which dwarf the
711 cost of adding the elements to the container. </p>
715 <!-- ======================================================================= -->
716 <div class="doc_subsection">
717 <a name="ds_sequential">Sequential Containers (std::vector, std::list, etc)</a>
720 <div class="doc_text">
721 There are a variety of sequential containers available for you, based on your
722 needs. Pick the first in this section that will do what you want.
725 <!-- _______________________________________________________________________ -->
726 <div class="doc_subsubsection">
727 <a name="dss_fixedarrays">Fixed Size Arrays</a>
730 <div class="doc_text">
731 <p>Fixed size arrays are very simple and very fast. They are good if you know
732 exactly how many elements you have, or you have a (low) upper bound on how many
736 <!-- _______________________________________________________________________ -->
737 <div class="doc_subsubsection">
738 <a name="dss_heaparrays">Heap Allocated Arrays</a>
741 <div class="doc_text">
742 <p>Heap allocated arrays (new[] + delete[]) are also simple. They are good if
743 the number of elements is variable, if you know how many elements you will need
744 before the array is allocated, and if the array is usually large (if not,
745 consider a <a href="#dss_smallvector">SmallVector</a>). The cost of a heap
746 allocated array is the cost of the new/delete (aka malloc/free). Also note that
747 if you are allocating an array of a type with a constructor, the constructor and
748 destructors will be run for every element in the array (resizable vectors only
749 construct those elements actually used).</p>
752 <!-- _______________________________________________________________________ -->
753 <div class="doc_subsubsection">
754 <a name="dss_smallvector">"llvm/ADT/SmallVector.h"</a>
757 <div class="doc_text">
758 <p><tt>SmallVector<Type, N></tt> is a simple class that looks and smells
759 just like <tt>vector<Type></tt>:
760 it supports efficient iteration, lays out elements in memory order (so you can
761 do pointer arithmetic between elements), supports efficient push_back/pop_back
762 operations, supports efficient random access to its elements, etc.</p>
764 <p>The advantage of SmallVector is that it allocates space for
765 some number of elements (N) <b>in the object itself</b>. Because of this, if
766 the SmallVector is dynamically smaller than N, no malloc is performed. This can
767 be a big win in cases where the malloc/free call is far more expensive than the
768 code that fiddles around with the elements.</p>
770 <p>This is good for vectors that are "usually small" (e.g. the number of
771 predecessors/successors of a block is usually less than 8). On the other hand,
772 this makes the size of the SmallVector itself large, so you don't want to
773 allocate lots of them (doing so will waste a lot of space). As such,
774 SmallVectors are most useful when on the stack.</p>
776 <p>SmallVector also provides a nice portable and efficient replacement for
781 <!-- _______________________________________________________________________ -->
782 <div class="doc_subsubsection">
783 <a name="dss_vector"><vector></a>
786 <div class="doc_text">
788 std::vector is well loved and respected. It is useful when SmallVector isn't:
789 when the size of the vector is often large (thus the small optimization will
790 rarely be a benefit) or if you will be allocating many instances of the vector
791 itself (which would waste space for elements that aren't in the container).
792 vector is also useful when interfacing with code that expects vectors :).
796 <!-- _______________________________________________________________________ -->
797 <div class="doc_subsubsection">
798 <a name="dss_deque"><deque></a>
801 <div class="doc_text">
802 <p>std::deque is, in some senses, a generalized version of std::vector. Like
803 std::vector, it provides constant time random access and other similar
804 properties, but it also provides efficient access to the front of the list. It
805 does not guarantee continuity of elements within memory.</p>
807 <p>In exchange for this extra flexibility, std::deque has significantly higher
808 constant factor costs than std::vector. If possible, use std::vector or
809 something cheaper.</p>
812 <!-- _______________________________________________________________________ -->
813 <div class="doc_subsubsection">
814 <a name="dss_list"><list></a>
817 <div class="doc_text">
818 <p>std::list is an extremely inefficient class that is rarely useful.
819 It performs a heap allocation for every element inserted into it, thus having an
820 extremely high constant factor, particularly for small data types. std::list
821 also only supports bidirectional iteration, not random access iteration.</p>
823 <p>In exchange for this high cost, std::list supports efficient access to both
824 ends of the list (like std::deque, but unlike std::vector or SmallVector). In
825 addition, the iterator invalidation characteristics of std::list are stronger
826 than that of a vector class: inserting or removing an element into the list does
827 not invalidate iterator or pointers to other elements in the list.</p>
830 <!-- _______________________________________________________________________ -->
831 <div class="doc_subsubsection">
832 <a name="dss_ilist">llvm/ADT/ilist</a>
835 <div class="doc_text">
836 <p><tt>ilist<T></tt> implements an 'intrusive' doubly-linked list. It is
837 intrusive, because it requires the element to store and provide access to the
838 prev/next pointers for the list.</p>
840 <p>ilist has the same drawbacks as std::list, and additionally requires an
841 ilist_traits implementation for the element type, but it provides some novel
842 characteristics. In particular, it can efficiently store polymorphic objects,
843 the traits class is informed when an element is inserted or removed from the
844 list, and ilists are guaranteed to support a constant-time splice operation.
847 <p>These properties are exactly what we want for things like Instructions and
848 basic blocks, which is why these are implemented with ilists.</p>
851 <!-- _______________________________________________________________________ -->
852 <div class="doc_subsubsection">
853 <a name="dss_other">Other options</a>
856 <div class="doc_text">
857 <p>Other STL containers are available, such as std::string.</p>
859 <p>There are also various STL adapter classes such as std::queue,
860 std::priority_queue, std::stack, etc. These provide simplified access to an
861 underlying container but don't affect the cost of the container itself.</p>
866 <!-- ======================================================================= -->
867 <div class="doc_subsection">
868 <a name="ds_set">Set-Like Containers (std::set, SmallSet, SetVector, etc)</a>
871 <div class="doc_text">
873 <p>Set-like containers are useful when you need to canonicalize multiple values
874 into a single representation. There are several different choices for how to do
875 this, providing various trade-offs.</p>
880 <!-- _______________________________________________________________________ -->
881 <div class="doc_subsubsection">
882 <a name="dss_sortedvectorset">A sorted 'vector'</a>
885 <div class="doc_text">
887 <p>If you intend to insert a lot of elements, then do a lot of queries, a
888 great approach is to use a vector (or other sequential container) with
889 std::sort+std::unique to remove duplicates. This approach works really well if
890 your usage pattern has these two distinct phases (insert then query), and can be
891 coupled with a good choice of <a href="#ds_sequential">sequential container</a>.
895 This combination provides the several nice properties: the result data is
896 contiguous in memory (good for cache locality), has few allocations, is easy to
897 address (iterators in the final vector are just indices or pointers), and can be
898 efficiently queried with a standard binary or radix search.</p>
902 <!-- _______________________________________________________________________ -->
903 <div class="doc_subsubsection">
904 <a name="dss_smallset">"llvm/ADT/SmallSet.h"</a>
907 <div class="doc_text">
909 <p>If you have a set-like datastructure that is usually small and whose elements
910 are reasonably small, a <tt>SmallSet<Type, N></tt> is a good choice. This set
911 has space for N elements in place (thus, if the set is dynamically smaller than
912 N, no malloc traffic is required) and accesses them with a simple linear search.
913 When the set grows beyond 'N' elements, it allocates a more expensive representation that
914 guarantees efficient access (for most types, it falls back to std::set, but for
915 pointers it uses something far better, <a
916 href="#dss_smallptrset">SmallPtrSet</a>).</p>
918 <p>The magic of this class is that it handles small sets extremely efficiently,
919 but gracefully handles extremely large sets without loss of efficiency. The
920 drawback is that the interface is quite small: it supports insertion, queries
921 and erasing, but does not support iteration.</p>
925 <!-- _______________________________________________________________________ -->
926 <div class="doc_subsubsection">
927 <a name="dss_smallptrset">"llvm/ADT/SmallPtrSet.h"</a>
930 <div class="doc_text">
932 <p>SmallPtrSet has all the advantages of SmallSet (and a SmallSet of pointers is
933 transparently implemented with a SmallPtrSet), but also suports iterators. If
934 more than 'N' insertions are performed, a single quadratically
935 probed hash table is allocated and grows as needed, providing extremely
936 efficient access (constant time insertion/deleting/queries with low constant
937 factors) and is very stingy with malloc traffic.</p>
939 <p>Note that, unlike std::set, the iterators of SmallPtrSet are invalidated
940 whenever an insertion occurs. Also, the values visited by the iterators are not
941 visited in sorted order.</p>
945 <!-- _______________________________________________________________________ -->
946 <div class="doc_subsubsection">
947 <a name="dss_FoldingSet">"llvm/ADT/FoldingSet.h"</a>
950 <div class="doc_text">
953 FoldingSet is an aggregate class that is really good at uniquing
954 expensive-to-create or polymorphic objects. It is a combination of a chained
955 hash table with intrusive links (uniqued objects are required to inherit from
956 FoldingSetNode) that uses <a href="#dss_smallvector">SmallVector</a> as part of
959 <p>Consider a case where you want to implement a "getOrCreateFoo" method for
960 a complex object (for example, a node in the code generator). The client has a
961 description of *what* it wants to generate (it knows the opcode and all the
962 operands), but we don't want to 'new' a node, then try inserting it into a set
963 only to find out it already exists, at which point we would have to delete it
964 and return the node that already exists.
967 <p>To support this style of client, FoldingSet perform a query with a
968 FoldingSetNodeID (which wraps SmallVector) that can be used to describe the
969 element that we want to query for. The query either returns the element
970 matching the ID or it returns an opaque ID that indicates where insertion should
971 take place. Construction of the ID usually does not require heap traffic.</p>
973 <p>Because FoldingSet uses intrusive links, it can support polymorphic objects
974 in the set (for example, you can have SDNode instances mixed with LoadSDNodes).
975 Because the elements are individually allocated, pointers to the elements are
976 stable: inserting or removing elements does not invalidate any pointers to other
982 <!-- _______________________________________________________________________ -->
983 <div class="doc_subsubsection">
984 <a name="dss_set"><set></a>
987 <div class="doc_text">
989 <p><tt>std::set</t> is a reasonable all-around set class, which is good at many
990 things but great at nothing. std::set allocates memory for each element
991 inserted (thus it is very malloc intensive) and typically stores three pointers
992 per element in the set (thus adding a large amount of per-element space
993 overhead). It offers guaranteed log(n) performance, which is not particularly
994 fast, particularly if the elements of the set are expensive to compare (e.g.
997 <p>The advantages of std::set are that its iterators are stable (deleting or
998 inserting an element from the set does not affect iterators or pointers to other
999 elements) and that iteration over the set is guaranteed to be in sorted order.
1000 If the elements in the set are large, then the relative overhead of the pointers
1001 and malloc traffic is not a big deal, but if the elements of the set are small,
1002 std::set is almost never a good choice.</p>
1006 <!-- _______________________________________________________________________ -->
1007 <div class="doc_subsubsection">
1008 <a name="dss_setvector">"llvm/ADT/SetVector.h"</a>
1011 <div class="doc_text">
1012 <p>LLVM's SetVector<Type> is actually a combination of a set along with
1013 a <a href="#ds_sequential">Sequential Container</a>. The important property
1014 that this provides is efficient insertion with uniquing (duplicate elements are
1015 ignored) with iteration support. It implements this by inserting elements into
1016 both a set-like container and the sequential container, using the set-like
1017 container for uniquing and the sequential container for iteration.
1020 <p>The difference between SetVector and other sets is that the order of
1021 iteration is guaranteed to match the order of insertion into the SetVector.
1022 This property is really important for things like sets of pointers. Because
1023 pointer values are non-deterministic (e.g. vary across runs of the program on
1024 different machines), iterating over the pointers in a std::set or other set will
1025 not be in a well-defined order.</p>
1028 The drawback of SetVector is that it requires twice as much space as a normal
1029 set and has the sum of constant factors from the set-like container and the
1030 sequential container that it uses. Use it *only* if you need to iterate over
1031 the elements in a deterministic order. SetVector is also expensive to delete
1032 elements out of (linear time).
1037 <!-- _______________________________________________________________________ -->
1038 <div class="doc_subsubsection">
1039 <a name="dss_otherset">Other Options</a>
1042 <div class="doc_text">
1045 The STL provides several other options, such as std::multiset and the various
1046 "hash_set" like containers (whether from C++TR1 or from the SGI library).</p>
1048 <p>std::multiset is useful if you're not interested in elimination of
1049 duplicates, but has all the drawbacks of std::set. A sorted vector (where you
1050 don't delete duplicate entries) or some other approach is almost always
1053 <p>The various hash_set implementations (exposed portably by
1054 "llvm/ADT/hash_set") is a simple chained hashtable. This algorithm is as malloc
1055 intensive as std::set (performing an allocation for each element inserted,
1056 thus having really high constant factors) but (usually) provides O(1)
1057 insertion/deletion of elements. This can be useful if your elements are large
1058 (thus making the constant-factor cost relatively low) or if comparisons are
1059 expensive. Element iteration does not visit elements in a useful order.</p>
1063 <!-- ======================================================================= -->
1064 <div class="doc_subsection">
1065 <a name="ds_map">Map-Like Containers (std::map, DenseMap, etc)</a>
1068 <div class="doc_text">
1079 <!-- *********************************************************************** -->
1080 <div class="doc_section">
1081 <a name="common">Helpful Hints for Common Operations</a>
1083 <!-- *********************************************************************** -->
1085 <div class="doc_text">
1087 <p>This section describes how to perform some very simple transformations of
1088 LLVM code. This is meant to give examples of common idioms used, showing the
1089 practical side of LLVM transformations. <p> Because this is a "how-to" section,
1090 you should also read about the main classes that you will be working with. The
1091 <a href="#coreclasses">Core LLVM Class Hierarchy Reference</a> contains details
1092 and descriptions of the main classes that you should know about.</p>
1096 <!-- NOTE: this section should be heavy on example code -->
1097 <!-- ======================================================================= -->
1098 <div class="doc_subsection">
1099 <a name="inspection">Basic Inspection and Traversal Routines</a>
1102 <div class="doc_text">
1104 <p>The LLVM compiler infrastructure have many different data structures that may
1105 be traversed. Following the example of the C++ standard template library, the
1106 techniques used to traverse these various data structures are all basically the
1107 same. For a enumerable sequence of values, the <tt>XXXbegin()</tt> function (or
1108 method) returns an iterator to the start of the sequence, the <tt>XXXend()</tt>
1109 function returns an iterator pointing to one past the last valid element of the
1110 sequence, and there is some <tt>XXXiterator</tt> data type that is common
1111 between the two operations.</p>
1113 <p>Because the pattern for iteration is common across many different aspects of
1114 the program representation, the standard template library algorithms may be used
1115 on them, and it is easier to remember how to iterate. First we show a few common
1116 examples of the data structures that need to be traversed. Other data
1117 structures are traversed in very similar ways.</p>
1121 <!-- _______________________________________________________________________ -->
1122 <div class="doc_subsubsection">
1123 <a name="iterate_function">Iterating over the </a><a
1124 href="#BasicBlock"><tt>BasicBlock</tt></a>s in a <a
1125 href="#Function"><tt>Function</tt></a>
1128 <div class="doc_text">
1130 <p>It's quite common to have a <tt>Function</tt> instance that you'd like to
1131 transform in some way; in particular, you'd like to manipulate its
1132 <tt>BasicBlock</tt>s. To facilitate this, you'll need to iterate over all of
1133 the <tt>BasicBlock</tt>s that constitute the <tt>Function</tt>. The following is
1134 an example that prints the name of a <tt>BasicBlock</tt> and the number of
1135 <tt>Instruction</tt>s it contains:</p>
1137 <div class="doc_code">
1139 // <i>func is a pointer to a Function instance</i>
1140 for (Function::iterator i = func->begin(), e = func->end(); i != e; ++i)
1141 // <i>Print out the name of the basic block if it has one, and then the</i>
1142 // <i>number of instructions that it contains</i>
1143 llvm::cerr << "Basic block (name=" << i->getName() << ") has "
1144 << i->size() << " instructions.\n";
1148 <p>Note that i can be used as if it were a pointer for the purposes of
1149 invoking member functions of the <tt>Instruction</tt> class. This is
1150 because the indirection operator is overloaded for the iterator
1151 classes. In the above code, the expression <tt>i->size()</tt> is
1152 exactly equivalent to <tt>(*i).size()</tt> just like you'd expect.</p>
1156 <!-- _______________________________________________________________________ -->
1157 <div class="doc_subsubsection">
1158 <a name="iterate_basicblock">Iterating over the </a><a
1159 href="#Instruction"><tt>Instruction</tt></a>s in a <a
1160 href="#BasicBlock"><tt>BasicBlock</tt></a>
1163 <div class="doc_text">
1165 <p>Just like when dealing with <tt>BasicBlock</tt>s in <tt>Function</tt>s, it's
1166 easy to iterate over the individual instructions that make up
1167 <tt>BasicBlock</tt>s. Here's a code snippet that prints out each instruction in
1168 a <tt>BasicBlock</tt>:</p>
1170 <div class="doc_code">
1172 // <i>blk is a pointer to a BasicBlock instance</i>
1173 for (BasicBlock::iterator i = blk->begin(), e = blk->end(); i != e; ++i)
1174 // <i>The next statement works since operator<<(ostream&,...)</i>
1175 // <i>is overloaded for Instruction&</i>
1176 llvm::cerr << *i << "\n";
1180 <p>However, this isn't really the best way to print out the contents of a
1181 <tt>BasicBlock</tt>! Since the ostream operators are overloaded for virtually
1182 anything you'll care about, you could have just invoked the print routine on the
1183 basic block itself: <tt>llvm::cerr << *blk << "\n";</tt>.</p>
1187 <!-- _______________________________________________________________________ -->
1188 <div class="doc_subsubsection">
1189 <a name="iterate_institer">Iterating over the </a><a
1190 href="#Instruction"><tt>Instruction</tt></a>s in a <a
1191 href="#Function"><tt>Function</tt></a>
1194 <div class="doc_text">
1196 <p>If you're finding that you commonly iterate over a <tt>Function</tt>'s
1197 <tt>BasicBlock</tt>s and then that <tt>BasicBlock</tt>'s <tt>Instruction</tt>s,
1198 <tt>InstIterator</tt> should be used instead. You'll need to include <a
1199 href="/doxygen/InstIterator_8h-source.html"><tt>llvm/Support/InstIterator.h</tt></a>,
1200 and then instantiate <tt>InstIterator</tt>s explicitly in your code. Here's a
1201 small example that shows how to dump all instructions in a function to the standard error stream:<p>
1203 <div class="doc_code">
1205 #include "<a href="/doxygen/InstIterator_8h-source.html">llvm/Support/InstIterator.h</a>"
1207 // <i>F is a ptr to a Function instance</i>
1208 for (inst_iterator i = inst_begin(F), e = inst_end(F); i != e; ++i)
1209 llvm::cerr << *i << "\n";
1213 <p>Easy, isn't it? You can also use <tt>InstIterator</tt>s to fill a
1214 worklist with its initial contents. For example, if you wanted to
1215 initialize a worklist to contain all instructions in a <tt>Function</tt>
1216 F, all you would need to do is something like:</p>
1218 <div class="doc_code">
1220 std::set<Instruction*> worklist;
1221 worklist.insert(inst_begin(F), inst_end(F));
1225 <p>The STL set <tt>worklist</tt> would now contain all instructions in the
1226 <tt>Function</tt> pointed to by F.</p>
1230 <!-- _______________________________________________________________________ -->
1231 <div class="doc_subsubsection">
1232 <a name="iterate_convert">Turning an iterator into a class pointer (and
1236 <div class="doc_text">
1238 <p>Sometimes, it'll be useful to grab a reference (or pointer) to a class
1239 instance when all you've got at hand is an iterator. Well, extracting
1240 a reference or a pointer from an iterator is very straight-forward.
1241 Assuming that <tt>i</tt> is a <tt>BasicBlock::iterator</tt> and <tt>j</tt>
1242 is a <tt>BasicBlock::const_iterator</tt>:</p>
1244 <div class="doc_code">
1246 Instruction& inst = *i; // <i>Grab reference to instruction reference</i>
1247 Instruction* pinst = &*i; // <i>Grab pointer to instruction reference</i>
1248 const Instruction& inst = *j;
1252 <p>However, the iterators you'll be working with in the LLVM framework are
1253 special: they will automatically convert to a ptr-to-instance type whenever they
1254 need to. Instead of dereferencing the iterator and then taking the address of
1255 the result, you can simply assign the iterator to the proper pointer type and
1256 you get the dereference and address-of operation as a result of the assignment
1257 (behind the scenes, this is a result of overloading casting mechanisms). Thus
1258 the last line of the last example,</p>
1260 <div class="doc_code">
1262 Instruction* pinst = &*i;
1266 <p>is semantically equivalent to</p>
1268 <div class="doc_code">
1270 Instruction* pinst = i;
1274 <p>It's also possible to turn a class pointer into the corresponding iterator,
1275 and this is a constant time operation (very efficient). The following code
1276 snippet illustrates use of the conversion constructors provided by LLVM
1277 iterators. By using these, you can explicitly grab the iterator of something
1278 without actually obtaining it via iteration over some structure:</p>
1280 <div class="doc_code">
1282 void printNextInstruction(Instruction* inst) {
1283 BasicBlock::iterator it(inst);
1284 ++it; // <i>After this line, it refers to the instruction after *inst</i>
1285 if (it != inst->getParent()->end()) llvm::cerr << *it << "\n";
1292 <!--_______________________________________________________________________-->
1293 <div class="doc_subsubsection">
1294 <a name="iterate_complex">Finding call sites: a slightly more complex
1298 <div class="doc_text">
1300 <p>Say that you're writing a FunctionPass and would like to count all the
1301 locations in the entire module (that is, across every <tt>Function</tt>) where a
1302 certain function (i.e., some <tt>Function</tt>*) is already in scope. As you'll
1303 learn later, you may want to use an <tt>InstVisitor</tt> to accomplish this in a
1304 much more straight-forward manner, but this example will allow us to explore how
1305 you'd do it if you didn't have <tt>InstVisitor</tt> around. In pseudocode, this
1306 is what we want to do:</p>
1308 <div class="doc_code">
1310 initialize callCounter to zero
1311 for each Function f in the Module
1312 for each BasicBlock b in f
1313 for each Instruction i in b
1314 if (i is a CallInst and calls the given function)
1315 increment callCounter
1319 <p>And the actual code is (remember, because we're writing a
1320 <tt>FunctionPass</tt>, our <tt>FunctionPass</tt>-derived class simply has to
1321 override the <tt>runOnFunction</tt> method):</p>
1323 <div class="doc_code">
1325 Function* targetFunc = ...;
1327 class OurFunctionPass : public FunctionPass {
1329 OurFunctionPass(): callCounter(0) { }
1331 virtual runOnFunction(Function& F) {
1332 for (Function::iterator b = F.begin(), be = F.end(); b != be; ++b) {
1333 for (BasicBlock::iterator i = b->begin(); ie = b->end(); i != ie; ++i) {
1334 if (<a href="#CallInst">CallInst</a>* callInst = <a href="#isa">dyn_cast</a><<a
1335 href="#CallInst">CallInst</a>>(&*i)) {
1336 // <i>We know we've encountered a call instruction, so we</i>
1337 // <i>need to determine if it's a call to the</i>
1338 // <i>function pointed to by m_func or not</i>
1340 if (callInst->getCalledFunction() == targetFunc)
1348 unsigned callCounter;
1355 <!--_______________________________________________________________________-->
1356 <div class="doc_subsubsection">
1357 <a name="calls_and_invokes">Treating calls and invokes the same way</a>
1360 <div class="doc_text">
1362 <p>You may have noticed that the previous example was a bit oversimplified in
1363 that it did not deal with call sites generated by 'invoke' instructions. In
1364 this, and in other situations, you may find that you want to treat
1365 <tt>CallInst</tt>s and <tt>InvokeInst</tt>s the same way, even though their
1366 most-specific common base class is <tt>Instruction</tt>, which includes lots of
1367 less closely-related things. For these cases, LLVM provides a handy wrapper
1369 href="http://llvm.org/doxygen/classllvm_1_1CallSite.html"><tt>CallSite</tt></a>.
1370 It is essentially a wrapper around an <tt>Instruction</tt> pointer, with some
1371 methods that provide functionality common to <tt>CallInst</tt>s and
1372 <tt>InvokeInst</tt>s.</p>
1374 <p>This class has "value semantics": it should be passed by value, not by
1375 reference and it should not be dynamically allocated or deallocated using
1376 <tt>operator new</tt> or <tt>operator delete</tt>. It is efficiently copyable,
1377 assignable and constructable, with costs equivalents to that of a bare pointer.
1378 If you look at its definition, it has only a single pointer member.</p>
1382 <!--_______________________________________________________________________-->
1383 <div class="doc_subsubsection">
1384 <a name="iterate_chains">Iterating over def-use & use-def chains</a>
1387 <div class="doc_text">
1389 <p>Frequently, we might have an instance of the <a
1390 href="/doxygen/classllvm_1_1Value.html">Value Class</a> and we want to
1391 determine which <tt>User</tt>s use the <tt>Value</tt>. The list of all
1392 <tt>User</tt>s of a particular <tt>Value</tt> is called a <i>def-use</i> chain.
1393 For example, let's say we have a <tt>Function*</tt> named <tt>F</tt> to a
1394 particular function <tt>foo</tt>. Finding all of the instructions that
1395 <i>use</i> <tt>foo</tt> is as simple as iterating over the <i>def-use</i> chain
1398 <div class="doc_code">
1402 for (Value::use_iterator i = F->use_begin(), e = F->use_end(); i != e; ++i)
1403 if (Instruction *Inst = dyn_cast<Instruction>(*i)) {
1404 llvm::cerr << "F is used in instruction:\n";
1405 llvm::cerr << *Inst << "\n";
1410 <p>Alternately, it's common to have an instance of the <a
1411 href="/doxygen/classllvm_1_1User.html">User Class</a> and need to know what
1412 <tt>Value</tt>s are used by it. The list of all <tt>Value</tt>s used by a
1413 <tt>User</tt> is known as a <i>use-def</i> chain. Instances of class
1414 <tt>Instruction</tt> are common <tt>User</tt>s, so we might want to iterate over
1415 all of the values that a particular instruction uses (that is, the operands of
1416 the particular <tt>Instruction</tt>):</p>
1418 <div class="doc_code">
1420 Instruction* pi = ...;
1422 for (User::op_iterator i = pi->op_begin(), e = pi->op_end(); i != e; ++i) {
1430 def-use chains ("finding all users of"): Value::use_begin/use_end
1431 use-def chains ("finding all values used"): User::op_begin/op_end [op=operand]
1436 <!-- ======================================================================= -->
1437 <div class="doc_subsection">
1438 <a name="simplechanges">Making simple changes</a>
1441 <div class="doc_text">
1443 <p>There are some primitive transformation operations present in the LLVM
1444 infrastructure that are worth knowing about. When performing
1445 transformations, it's fairly common to manipulate the contents of basic
1446 blocks. This section describes some of the common methods for doing so
1447 and gives example code.</p>
1451 <!--_______________________________________________________________________-->
1452 <div class="doc_subsubsection">
1453 <a name="schanges_creating">Creating and inserting new
1454 <tt>Instruction</tt>s</a>
1457 <div class="doc_text">
1459 <p><i>Instantiating Instructions</i></p>
1461 <p>Creation of <tt>Instruction</tt>s is straight-forward: simply call the
1462 constructor for the kind of instruction to instantiate and provide the necessary
1463 parameters. For example, an <tt>AllocaInst</tt> only <i>requires</i> a
1464 (const-ptr-to) <tt>Type</tt>. Thus:</p>
1466 <div class="doc_code">
1468 AllocaInst* ai = new AllocaInst(Type::IntTy);
1472 <p>will create an <tt>AllocaInst</tt> instance that represents the allocation of
1473 one integer in the current stack frame, at runtime. Each <tt>Instruction</tt>
1474 subclass is likely to have varying default parameters which change the semantics
1475 of the instruction, so refer to the <a
1476 href="/doxygen/classllvm_1_1Instruction.html">doxygen documentation for the subclass of
1477 Instruction</a> that you're interested in instantiating.</p>
1479 <p><i>Naming values</i></p>
1481 <p>It is very useful to name the values of instructions when you're able to, as
1482 this facilitates the debugging of your transformations. If you end up looking
1483 at generated LLVM machine code, you definitely want to have logical names
1484 associated with the results of instructions! By supplying a value for the
1485 <tt>Name</tt> (default) parameter of the <tt>Instruction</tt> constructor, you
1486 associate a logical name with the result of the instruction's execution at
1487 runtime. For example, say that I'm writing a transformation that dynamically
1488 allocates space for an integer on the stack, and that integer is going to be
1489 used as some kind of index by some other code. To accomplish this, I place an
1490 <tt>AllocaInst</tt> at the first point in the first <tt>BasicBlock</tt> of some
1491 <tt>Function</tt>, and I'm intending to use it within the same
1492 <tt>Function</tt>. I might do:</p>
1494 <div class="doc_code">
1496 AllocaInst* pa = new AllocaInst(Type::IntTy, 0, "indexLoc");
1500 <p>where <tt>indexLoc</tt> is now the logical name of the instruction's
1501 execution value, which is a pointer to an integer on the runtime stack.</p>
1503 <p><i>Inserting instructions</i></p>
1505 <p>There are essentially two ways to insert an <tt>Instruction</tt>
1506 into an existing sequence of instructions that form a <tt>BasicBlock</tt>:</p>
1509 <li>Insertion into an explicit instruction list
1511 <p>Given a <tt>BasicBlock* pb</tt>, an <tt>Instruction* pi</tt> within that
1512 <tt>BasicBlock</tt>, and a newly-created instruction we wish to insert
1513 before <tt>*pi</tt>, we do the following: </p>
1515 <div class="doc_code">
1517 BasicBlock *pb = ...;
1518 Instruction *pi = ...;
1519 Instruction *newInst = new Instruction(...);
1521 pb->getInstList().insert(pi, newInst); // <i>Inserts newInst before pi in pb</i>
1525 <p>Appending to the end of a <tt>BasicBlock</tt> is so common that
1526 the <tt>Instruction</tt> class and <tt>Instruction</tt>-derived
1527 classes provide constructors which take a pointer to a
1528 <tt>BasicBlock</tt> to be appended to. For example code that
1531 <div class="doc_code">
1533 BasicBlock *pb = ...;
1534 Instruction *newInst = new Instruction(...);
1536 pb->getInstList().push_back(newInst); // <i>Appends newInst to pb</i>
1542 <div class="doc_code">
1544 BasicBlock *pb = ...;
1545 Instruction *newInst = new Instruction(..., pb);
1549 <p>which is much cleaner, especially if you are creating
1550 long instruction streams.</p></li>
1552 <li>Insertion into an implicit instruction list
1554 <p><tt>Instruction</tt> instances that are already in <tt>BasicBlock</tt>s
1555 are implicitly associated with an existing instruction list: the instruction
1556 list of the enclosing basic block. Thus, we could have accomplished the same
1557 thing as the above code without being given a <tt>BasicBlock</tt> by doing:
1560 <div class="doc_code">
1562 Instruction *pi = ...;
1563 Instruction *newInst = new Instruction(...);
1565 pi->getParent()->getInstList().insert(pi, newInst);
1569 <p>In fact, this sequence of steps occurs so frequently that the
1570 <tt>Instruction</tt> class and <tt>Instruction</tt>-derived classes provide
1571 constructors which take (as a default parameter) a pointer to an
1572 <tt>Instruction</tt> which the newly-created <tt>Instruction</tt> should
1573 precede. That is, <tt>Instruction</tt> constructors are capable of
1574 inserting the newly-created instance into the <tt>BasicBlock</tt> of a
1575 provided instruction, immediately before that instruction. Using an
1576 <tt>Instruction</tt> constructor with a <tt>insertBefore</tt> (default)
1577 parameter, the above code becomes:</p>
1579 <div class="doc_code">
1581 Instruction* pi = ...;
1582 Instruction* newInst = new Instruction(..., pi);
1586 <p>which is much cleaner, especially if you're creating a lot of
1587 instructions and adding them to <tt>BasicBlock</tt>s.</p></li>
1592 <!--_______________________________________________________________________-->
1593 <div class="doc_subsubsection">
1594 <a name="schanges_deleting">Deleting <tt>Instruction</tt>s</a>
1597 <div class="doc_text">
1599 <p>Deleting an instruction from an existing sequence of instructions that form a
1600 <a href="#BasicBlock"><tt>BasicBlock</tt></a> is very straight-forward. First,
1601 you must have a pointer to the instruction that you wish to delete. Second, you
1602 need to obtain the pointer to that instruction's basic block. You use the
1603 pointer to the basic block to get its list of instructions and then use the
1604 erase function to remove your instruction. For example:</p>
1606 <div class="doc_code">
1608 <a href="#Instruction">Instruction</a> *I = .. ;
1609 <a href="#BasicBlock">BasicBlock</a> *BB = I->getParent();
1611 BB->getInstList().erase(I);
1617 <!--_______________________________________________________________________-->
1618 <div class="doc_subsubsection">
1619 <a name="schanges_replacing">Replacing an <tt>Instruction</tt> with another
1623 <div class="doc_text">
1625 <p><i>Replacing individual instructions</i></p>
1627 <p>Including "<a href="/doxygen/BasicBlockUtils_8h-source.html">llvm/Transforms/Utils/BasicBlockUtils.h</a>"
1628 permits use of two very useful replace functions: <tt>ReplaceInstWithValue</tt>
1629 and <tt>ReplaceInstWithInst</tt>.</p>
1631 <h4><a name="schanges_deleting">Deleting <tt>Instruction</tt>s</a></h4>
1634 <li><tt>ReplaceInstWithValue</tt>
1636 <p>This function replaces all uses (within a basic block) of a given
1637 instruction with a value, and then removes the original instruction. The
1638 following example illustrates the replacement of the result of a particular
1639 <tt>AllocaInst</tt> that allocates memory for a single integer with a null
1640 pointer to an integer.</p>
1642 <div class="doc_code">
1644 AllocaInst* instToReplace = ...;
1645 BasicBlock::iterator ii(instToReplace);
1647 ReplaceInstWithValue(instToReplace->getParent()->getInstList(), ii,
1648 Constant::getNullValue(PointerType::get(Type::IntTy)));
1651 <li><tt>ReplaceInstWithInst</tt>
1653 <p>This function replaces a particular instruction with another
1654 instruction. The following example illustrates the replacement of one
1655 <tt>AllocaInst</tt> with another.</p>
1657 <div class="doc_code">
1659 AllocaInst* instToReplace = ...;
1660 BasicBlock::iterator ii(instToReplace);
1662 ReplaceInstWithInst(instToReplace->getParent()->getInstList(), ii,
1663 new AllocaInst(Type::IntTy, 0, "ptrToReplacedInt"));
1667 <p><i>Replacing multiple uses of <tt>User</tt>s and <tt>Value</tt>s</i></p>
1669 <p>You can use <tt>Value::replaceAllUsesWith</tt> and
1670 <tt>User::replaceUsesOfWith</tt> to change more than one use at a time. See the
1671 doxygen documentation for the <a href="/doxygen/classllvm_1_1Value.html">Value Class</a>
1672 and <a href="/doxygen/classllvm_1_1User.html">User Class</a>, respectively, for more
1675 <!-- Value::replaceAllUsesWith User::replaceUsesOfWith Point out:
1676 include/llvm/Transforms/Utils/ especially BasicBlockUtils.h with:
1677 ReplaceInstWithValue, ReplaceInstWithInst -->
1681 <!-- *********************************************************************** -->
1682 <div class="doc_section">
1683 <a name="advanced">Advanced Topics</a>
1685 <!-- *********************************************************************** -->
1687 <div class="doc_text">
1689 This section describes some of the advanced or obscure API's that most clients
1690 do not need to be aware of. These API's tend manage the inner workings of the
1691 LLVM system, and only need to be accessed in unusual circumstances.
1695 <!-- ======================================================================= -->
1696 <div class="doc_subsection">
1697 <a name="TypeResolve">LLVM Type Resolution</a>
1700 <div class="doc_text">
1703 The LLVM type system has a very simple goal: allow clients to compare types for
1704 structural equality with a simple pointer comparison (aka a shallow compare).
1705 This goal makes clients much simpler and faster, and is used throughout the LLVM
1710 Unfortunately achieving this goal is not a simple matter. In particular,
1711 recursive types and late resolution of opaque types makes the situation very
1712 difficult to handle. Fortunately, for the most part, our implementation makes
1713 most clients able to be completely unaware of the nasty internal details. The
1714 primary case where clients are exposed to the inner workings of it are when
1715 building a recursive type. In addition to this case, the LLVM bytecode reader,
1716 assembly parser, and linker also have to be aware of the inner workings of this
1721 For our purposes below, we need three concepts. First, an "Opaque Type" is
1722 exactly as defined in the <a href="LangRef.html#t_opaque">language
1723 reference</a>. Second an "Abstract Type" is any type which includes an
1724 opaque type as part of its type graph (for example "<tt>{ opaque, i32 }</tt>").
1725 Third, a concrete type is a type that is not an abstract type (e.g. "<tt>{ i32,
1731 <!-- ______________________________________________________________________ -->
1732 <div class="doc_subsubsection">
1733 <a name="BuildRecType">Basic Recursive Type Construction</a>
1736 <div class="doc_text">
1739 Because the most common question is "how do I build a recursive type with LLVM",
1740 we answer it now and explain it as we go. Here we include enough to cause this
1741 to be emitted to an output .ll file:
1744 <div class="doc_code">
1746 %mylist = type { %mylist*, i32 }
1751 To build this, use the following LLVM APIs:
1754 <div class="doc_code">
1756 // <i>Create the initial outer struct</i>
1757 <a href="#PATypeHolder">PATypeHolder</a> StructTy = OpaqueType::get();
1758 std::vector<const Type*> Elts;
1759 Elts.push_back(PointerType::get(StructTy));
1760 Elts.push_back(Type::IntTy);
1761 StructType *NewSTy = StructType::get(Elts);
1763 // <i>At this point, NewSTy = "{ opaque*, i32 }". Tell VMCore that</i>
1764 // <i>the struct and the opaque type are actually the same.</i>
1765 cast<OpaqueType>(StructTy.get())-><a href="#refineAbstractTypeTo">refineAbstractTypeTo</a>(NewSTy);
1767 // <i>NewSTy is potentially invalidated, but StructTy (a <a href="#PATypeHolder">PATypeHolder</a>) is</i>
1768 // <i>kept up-to-date</i>
1769 NewSTy = cast<StructType>(StructTy.get());
1771 // <i>Add a name for the type to the module symbol table (optional)</i>
1772 MyModule->addTypeName("mylist", NewSTy);
1777 This code shows the basic approach used to build recursive types: build a
1778 non-recursive type using 'opaque', then use type unification to close the cycle.
1779 The type unification step is performed by the <tt><a
1780 href="#refineAbstractTypeTo">refineAbstractTypeTo</a></tt> method, which is
1781 described next. After that, we describe the <a
1782 href="#PATypeHolder">PATypeHolder class</a>.
1787 <!-- ______________________________________________________________________ -->
1788 <div class="doc_subsubsection">
1789 <a name="refineAbstractTypeTo">The <tt>refineAbstractTypeTo</tt> method</a>
1792 <div class="doc_text">
1794 The <tt>refineAbstractTypeTo</tt> method starts the type unification process.
1795 While this method is actually a member of the DerivedType class, it is most
1796 often used on OpaqueType instances. Type unification is actually a recursive
1797 process. After unification, types can become structurally isomorphic to
1798 existing types, and all duplicates are deleted (to preserve pointer equality).
1802 In the example above, the OpaqueType object is definitely deleted.
1803 Additionally, if there is an "{ \2*, i32}" type already created in the system,
1804 the pointer and struct type created are <b>also</b> deleted. Obviously whenever
1805 a type is deleted, any "Type*" pointers in the program are invalidated. As
1806 such, it is safest to avoid having <i>any</i> "Type*" pointers to abstract types
1807 live across a call to <tt>refineAbstractTypeTo</tt> (note that non-abstract
1808 types can never move or be deleted). To deal with this, the <a
1809 href="#PATypeHolder">PATypeHolder</a> class is used to maintain a stable
1810 reference to a possibly refined type, and the <a
1811 href="#AbstractTypeUser">AbstractTypeUser</a> class is used to update more
1812 complex datastructures.
1817 <!-- ______________________________________________________________________ -->
1818 <div class="doc_subsubsection">
1819 <a name="PATypeHolder">The PATypeHolder Class</a>
1822 <div class="doc_text">
1824 PATypeHolder is a form of a "smart pointer" for Type objects. When VMCore
1825 happily goes about nuking types that become isomorphic to existing types, it
1826 automatically updates all PATypeHolder objects to point to the new type. In the
1827 example above, this allows the code to maintain a pointer to the resultant
1828 resolved recursive type, even though the Type*'s are potentially invalidated.
1832 PATypeHolder is an extremely light-weight object that uses a lazy union-find
1833 implementation to update pointers. For example the pointer from a Value to its
1834 Type is maintained by PATypeHolder objects.
1839 <!-- ______________________________________________________________________ -->
1840 <div class="doc_subsubsection">
1841 <a name="AbstractTypeUser">The AbstractTypeUser Class</a>
1844 <div class="doc_text">
1847 Some data structures need more to perform more complex updates when types get
1848 resolved. The <a href="#SymbolTable">SymbolTable</a> class, for example, needs
1849 move and potentially merge type planes in its representation when a pointer
1853 To support this, a class can derive from the AbstractTypeUser class. This class
1854 allows it to get callbacks when certain types are resolved. To register to get
1855 callbacks for a particular type, the DerivedType::{add/remove}AbstractTypeUser
1856 methods can be called on a type. Note that these methods only work for <i>
1857 abstract</i> types. Concrete types (those that do not include any opaque
1858 objects) can never be refined.
1863 <!-- ======================================================================= -->
1864 <div class="doc_subsection">
1865 <a name="SymbolTable">The <tt>SymbolTable</tt> class</a>
1868 <div class="doc_text">
1869 <p>This class provides a symbol table that the <a
1870 href="#Function"><tt>Function</tt></a> and <a href="#Module">
1871 <tt>Module</tt></a> classes use for naming definitions. The symbol table can
1872 provide a name for any <a href="#Value"><tt>Value</tt></a>.
1873 <tt>SymbolTable</tt> is an abstract data type. It hides the data it contains
1874 and provides access to it through a controlled interface.</p>
1876 <p>Note that the <tt>SymbolTable</tt> class should not be directly accessed
1877 by most clients. It should only be used when iteration over the symbol table
1878 names themselves are required, which is very special purpose. Note that not
1880 <a href="#Value">Value</a>s have names, and those without names (i.e. they have
1881 an empty name) do not exist in the symbol table.
1884 <p>To use the <tt>SymbolTable</tt> well, you need to understand the
1885 structure of the information it holds. The class contains two
1886 <tt>std::map</tt> objects. The first, <tt>pmap</tt>, is a map of
1887 <tt>Type*</tt> to maps of name (<tt>std::string</tt>) to <tt>Value*</tt>.
1888 Thus, Values are stored in two-dimensions and accessed by <tt>Type</tt> and
1891 <p>The interface of this class provides three basic types of operations:
1893 <li><em>Accessors</em>. Accessors provide read-only access to information
1894 such as finding a value for a name with the
1895 <a href="#SymbolTable_lookup">lookup</a> method.</li>
1896 <li><em>Mutators</em>. Mutators allow the user to add information to the
1897 <tt>SymbolTable</tt> with methods like
1898 <a href="#SymbolTable_insert"><tt>insert</tt></a>.</li>
1899 <li><em>Iterators</em>. Iterators allow the user to traverse the content
1900 of the symbol table in well defined ways, such as the method
1901 <a href="#SymbolTable_plane_begin"><tt>plane_begin</tt></a>.</li>
1906 <dt><tt>Value* lookup(const Type* Ty, const std::string& name) const</tt>:
1908 <dd>The <tt>lookup</tt> method searches the type plane given by the
1909 <tt>Ty</tt> parameter for a <tt>Value</tt> with the provided <tt>name</tt>.
1910 If a suitable <tt>Value</tt> is not found, null is returned.</dd>
1912 <dt><tt>bool isEmpty() const</tt>:</dt>
1913 <dd>This function returns true if both the value and types maps are
1919 <dt><tt>void insert(Value *Val)</tt>:</dt>
1920 <dd>This method adds the provided value to the symbol table. The Value must
1921 have both a name and a type which are extracted and used to place the value
1922 in the correct type plane under the value's name.</dd>
1924 <dt><tt>void insert(const std::string& Name, Value *Val)</tt>:</dt>
1925 <dd> Inserts a constant or type into the symbol table with the specified
1926 name. There can be a many to one mapping between names and constants
1929 <dt><tt>void remove(Value* Val)</tt>:</dt>
1930 <dd> This method removes a named value from the symbol table. The
1931 type and name of the Value are extracted from \p N and used to
1932 lookup the Value in the correct type plane. If the Value is
1933 not in the symbol table, this method silently ignores the
1936 <dt><tt>Value* remove(const std::string& Name, Value *Val)</tt>:</dt>
1937 <dd> Remove a constant or type with the specified name from the
1940 <dt><tt>Value *remove(const value_iterator& It)</tt>:</dt>
1941 <dd> Removes a specific value from the symbol table.
1942 Returns the removed value.</dd>
1944 <dt><tt>bool strip()</tt>:</dt>
1945 <dd> This method will strip the symbol table of its names leaving
1946 the type and values. </dd>
1948 <dt><tt>void clear()</tt>:</dt>
1949 <dd>Empty the symbol table completely.</dd>
1953 <p>The following functions describe three types of iterators you can obtain
1954 the beginning or end of the sequence for both const and non-const. It is
1955 important to keep track of the different kinds of iterators. There are
1956 three idioms worth pointing out:</p>
1959 <tr><th>Units</th><th>Iterator</th><th>Idiom</th></tr>
1961 <td align="left">Planes Of name/Value maps</td><td>PI</td>
1962 <td align="left"><pre><tt>
1963 for (SymbolTable::plane_const_iterator PI = ST.plane_begin(),
1964 PE = ST.plane_end(); PI != PE; ++PI ) {
1965 PI->first // <i>This is the Type* of the plane</i>
1966 PI->second // <i>This is the SymbolTable::ValueMap of name/Value pairs</i>
1971 <td align="left">name/Value pairs in a plane</td><td>VI</td>
1972 <td align="left"><pre><tt>
1973 for (SymbolTable::value_const_iterator VI = ST.value_begin(SomeType),
1974 VE = ST.value_end(SomeType); VI != VE; ++VI ) {
1975 VI->first // <i>This is the name of the Value</i>
1976 VI->second // <i>This is the Value* value associated with the name</i>
1982 <p>Using the recommended iterator names and idioms will help you avoid
1983 making mistakes. Of particular note, make sure that whenever you use
1984 value_begin(SomeType) that you always compare the resulting iterator
1985 with value_end(SomeType) not value_end(SomeOtherType) or else you
1986 will loop infinitely.</p>
1990 <dt><tt>plane_iterator plane_begin()</tt>:</dt>
1991 <dd>Get an iterator that starts at the beginning of the type planes.
1992 The iterator will iterate over the Type/ValueMap pairs in the
1995 <dt><tt>plane_const_iterator plane_begin() const</tt>:</dt>
1996 <dd>Get a const_iterator that starts at the beginning of the type
1997 planes. The iterator will iterate over the Type/ValueMap pairs
1998 in the type planes. </dd>
2000 <dt><tt>plane_iterator plane_end()</tt>:</dt>
2001 <dd>Get an iterator at the end of the type planes. This serves as
2002 the marker for end of iteration over the type planes.</dd>
2004 <dt><tt>plane_const_iterator plane_end() const</tt>:</dt>
2005 <dd>Get a const_iterator at the end of the type planes. This serves as
2006 the marker for end of iteration over the type planes.</dd>
2008 <dt><tt>value_iterator value_begin(const Type *Typ)</tt>:</dt>
2009 <dd>Get an iterator that starts at the beginning of a type plane.
2010 The iterator will iterate over the name/value pairs in the type plane.
2011 Note: The type plane must already exist before using this.</dd>
2013 <dt><tt>value_const_iterator value_begin(const Type *Typ) const</tt>:</dt>
2014 <dd>Get a const_iterator that starts at the beginning of a type plane.
2015 The iterator will iterate over the name/value pairs in the type plane.
2016 Note: The type plane must already exist before using this.</dd>
2018 <dt><tt>value_iterator value_end(const Type *Typ)</tt>:</dt>
2019 <dd>Get an iterator to the end of a type plane. This serves as the marker
2020 for end of iteration of the type plane.
2021 Note: The type plane must already exist before using this.</dd>
2023 <dt><tt>value_const_iterator value_end(const Type *Typ) const</tt>:</dt>
2024 <dd>Get a const_iterator to the end of a type plane. This serves as the
2025 marker for end of iteration of the type plane.
2026 Note: the type plane must already exist before using this.</dd>
2028 <dt><tt>plane_const_iterator find(const Type* Typ ) const</tt>:</dt>
2029 <dd>This method returns a plane_const_iterator for iteration over
2030 the type planes starting at a specific plane, given by \p Ty.</dd>
2032 <dt><tt>plane_iterator find( const Type* Typ </tt>:</dt>
2033 <dd>This method returns a plane_iterator for iteration over the
2034 type planes starting at a specific plane, given by \p Ty.</dd>
2041 <!-- *********************************************************************** -->
2042 <div class="doc_section">
2043 <a name="coreclasses">The Core LLVM Class Hierarchy Reference </a>
2045 <!-- *********************************************************************** -->
2047 <div class="doc_text">
2048 <p><tt>#include "<a href="/doxygen/Type_8h-source.html">llvm/Type.h</a>"</tt>
2049 <br>doxygen info: <a href="/doxygen/classllvm_1_1Type.html">Type Class</a></p>
2051 <p>The Core LLVM classes are the primary means of representing the program
2052 being inspected or transformed. The core LLVM classes are defined in
2053 header files in the <tt>include/llvm/</tt> directory, and implemented in
2054 the <tt>lib/VMCore</tt> directory.</p>
2058 <!-- ======================================================================= -->
2059 <div class="doc_subsection">
2060 <a name="Type">The <tt>Type</tt> class and Derived Types</a>
2063 <div class="doc_text">
2065 <p><tt>Type</tt> is a superclass of all type classes. Every <tt>Value</tt> has
2066 a <tt>Type</tt>. <tt>Type</tt> cannot be instantiated directly but only
2067 through its subclasses. Certain primitive types (<tt>VoidType</tt>,
2068 <tt>LabelType</tt>, <tt>FloatType</tt> and <tt>DoubleType</tt>) have hidden
2069 subclasses. They are hidden because they offer no useful functionality beyond
2070 what the <tt>Type</tt> class offers except to distinguish themselves from
2071 other subclasses of <tt>Type</tt>.</p>
2072 <p>All other types are subclasses of <tt>DerivedType</tt>. Types can be
2073 named, but this is not a requirement. There exists exactly
2074 one instance of a given shape at any one time. This allows type equality to
2075 be performed with address equality of the Type Instance. That is, given two
2076 <tt>Type*</tt> values, the types are identical if the pointers are identical.
2080 <!-- _______________________________________________________________________ -->
2081 <div class="doc_subsubsection">
2082 <a name="m_Value">Important Public Methods</a>
2085 <div class="doc_text">
2088 <li><tt>bool isInteger() const</tt>: Returns true for any integer type.</li>
2090 <li><tt>bool isFloatingPoint()</tt>: Return true if this is one of the two
2091 floating point types.</li>
2093 <li><tt>bool isAbstract()</tt>: Return true if the type is abstract (contains
2094 an OpaqueType anywhere in its definition).</li>
2096 <li><tt>bool isSized()</tt>: Return true if the type has known size. Things
2097 that don't have a size are abstract types, labels and void.</li>
2102 <!-- _______________________________________________________________________ -->
2103 <div class="doc_subsubsection">
2104 <a name="m_Value">Important Derived Types</a>
2106 <div class="doc_text">
2108 <dt><tt>IntegerType</tt></dt>
2109 <dd>Subclass of DerivedType that represents integer types of any bit width.
2110 Any bit width between <tt>IntegerType::MIN_INT_BITS</tt> (1) and
2111 <tt>IntegerType::MAX_INT_BITS</tt> (~8 million) can be represented.
2113 <li><tt>static const IntegerType* get(unsigned NumBits)</tt>: get an integer
2114 type of a specific bit width.</li>
2115 <li><tt>unsigned getBitWidth() const</tt>: Get the bit width of an integer
2119 <dt><tt>SequentialType</tt></dt>
2120 <dd>This is subclassed by ArrayType and PointerType
2122 <li><tt>const Type * getElementType() const</tt>: Returns the type of each
2123 of the elements in the sequential type. </li>
2126 <dt><tt>ArrayType</tt></dt>
2127 <dd>This is a subclass of SequentialType and defines the interface for array
2130 <li><tt>unsigned getNumElements() const</tt>: Returns the number of
2131 elements in the array. </li>
2134 <dt><tt>PointerType</tt></dt>
2135 <dd>Subclass of SequentialType for pointer types.</dd>
2136 <dt><tt>PackedType</tt></dt>
2137 <dd>Subclass of SequentialType for packed (vector) types. A
2138 packed type is similar to an ArrayType but is distinguished because it is
2139 a first class type wherease ArrayType is not. Packed types are used for
2140 vector operations and are usually small vectors of of an integer or floating
2142 <dt><tt>StructType</tt></dt>
2143 <dd>Subclass of DerivedTypes for struct types.</dd>
2144 <dt><tt>FunctionType</tt></dt>
2145 <dd>Subclass of DerivedTypes for function types.
2147 <li><tt>bool isVarArg() const</tt>: Returns true if its a vararg
2149 <li><tt> const Type * getReturnType() const</tt>: Returns the
2150 return type of the function.</li>
2151 <li><tt>const Type * getParamType (unsigned i)</tt>: Returns
2152 the type of the ith parameter.</li>
2153 <li><tt> const unsigned getNumParams() const</tt>: Returns the
2154 number of formal parameters.</li>
2157 <dt><tt>OpaqueType</tt></dt>
2158 <dd>Sublcass of DerivedType for abstract types. This class
2159 defines no content and is used as a placeholder for some other type. Note
2160 that OpaqueType is used (temporarily) during type resolution for forward
2161 references of types. Once the referenced type is resolved, the OpaqueType
2162 is replaced with the actual type. OpaqueType can also be used for data
2163 abstraction. At link time opaque types can be resolved to actual types
2164 of the same name.</dd>
2168 <!-- ======================================================================= -->
2169 <div class="doc_subsection">
2170 <a name="Value">The <tt>Value</tt> class</a>
2175 <p><tt>#include "<a href="/doxygen/Value_8h-source.html">llvm/Value.h</a>"</tt>
2177 doxygen info: <a href="/doxygen/classllvm_1_1Value.html">Value Class</a></p>
2179 <p>The <tt>Value</tt> class is the most important class in the LLVM Source
2180 base. It represents a typed value that may be used (among other things) as an
2181 operand to an instruction. There are many different types of <tt>Value</tt>s,
2182 such as <a href="#Constant"><tt>Constant</tt></a>s,<a
2183 href="#Argument"><tt>Argument</tt></a>s. Even <a
2184 href="#Instruction"><tt>Instruction</tt></a>s and <a
2185 href="#Function"><tt>Function</tt></a>s are <tt>Value</tt>s.</p>
2187 <p>A particular <tt>Value</tt> may be used many times in the LLVM representation
2188 for a program. For example, an incoming argument to a function (represented
2189 with an instance of the <a href="#Argument">Argument</a> class) is "used" by
2190 every instruction in the function that references the argument. To keep track
2191 of this relationship, the <tt>Value</tt> class keeps a list of all of the <a
2192 href="#User"><tt>User</tt></a>s that is using it (the <a
2193 href="#User"><tt>User</tt></a> class is a base class for all nodes in the LLVM
2194 graph that can refer to <tt>Value</tt>s). This use list is how LLVM represents
2195 def-use information in the program, and is accessible through the <tt>use_</tt>*
2196 methods, shown below.</p>
2198 <p>Because LLVM is a typed representation, every LLVM <tt>Value</tt> is typed,
2199 and this <a href="#Type">Type</a> is available through the <tt>getType()</tt>
2200 method. In addition, all LLVM values can be named. The "name" of the
2201 <tt>Value</tt> is a symbolic string printed in the LLVM code:</p>
2203 <div class="doc_code">
2205 %<b>foo</b> = add i32 1, 2
2209 <p><a name="#nameWarning">The name of this instruction is "foo".</a> <b>NOTE</b>
2210 that the name of any value may be missing (an empty string), so names should
2211 <b>ONLY</b> be used for debugging (making the source code easier to read,
2212 debugging printouts), they should not be used to keep track of values or map
2213 between them. For this purpose, use a <tt>std::map</tt> of pointers to the
2214 <tt>Value</tt> itself instead.</p>
2216 <p>One important aspect of LLVM is that there is no distinction between an SSA
2217 variable and the operation that produces it. Because of this, any reference to
2218 the value produced by an instruction (or the value available as an incoming
2219 argument, for example) is represented as a direct pointer to the instance of
2221 represents this value. Although this may take some getting used to, it
2222 simplifies the representation and makes it easier to manipulate.</p>
2226 <!-- _______________________________________________________________________ -->
2227 <div class="doc_subsubsection">
2228 <a name="m_Value">Important Public Members of the <tt>Value</tt> class</a>
2231 <div class="doc_text">
2234 <li><tt>Value::use_iterator</tt> - Typedef for iterator over the
2236 <tt>Value::use_const_iterator</tt> - Typedef for const_iterator over
2238 <tt>unsigned use_size()</tt> - Returns the number of users of the
2240 <tt>bool use_empty()</tt> - Returns true if there are no users.<br>
2241 <tt>use_iterator use_begin()</tt> - Get an iterator to the start of
2243 <tt>use_iterator use_end()</tt> - Get an iterator to the end of the
2245 <tt><a href="#User">User</a> *use_back()</tt> - Returns the last
2246 element in the list.
2247 <p> These methods are the interface to access the def-use
2248 information in LLVM. As with all other iterators in LLVM, the naming
2249 conventions follow the conventions defined by the <a href="#stl">STL</a>.</p>
2251 <li><tt><a href="#Type">Type</a> *getType() const</tt>
2252 <p>This method returns the Type of the Value.</p>
2254 <li><tt>bool hasName() const</tt><br>
2255 <tt>std::string getName() const</tt><br>
2256 <tt>void setName(const std::string &Name)</tt>
2257 <p> This family of methods is used to access and assign a name to a <tt>Value</tt>,
2258 be aware of the <a href="#nameWarning">precaution above</a>.</p>
2260 <li><tt>void replaceAllUsesWith(Value *V)</tt>
2262 <p>This method traverses the use list of a <tt>Value</tt> changing all <a
2263 href="#User"><tt>User</tt>s</a> of the current value to refer to
2264 "<tt>V</tt>" instead. For example, if you detect that an instruction always
2265 produces a constant value (for example through constant folding), you can
2266 replace all uses of the instruction with the constant like this:</p>
2268 <div class="doc_code">
2270 Inst->replaceAllUsesWith(ConstVal);
2278 <!-- ======================================================================= -->
2279 <div class="doc_subsection">
2280 <a name="User">The <tt>User</tt> class</a>
2283 <div class="doc_text">
2286 <tt>#include "<a href="/doxygen/User_8h-source.html">llvm/User.h</a>"</tt><br>
2287 doxygen info: <a href="/doxygen/classllvm_1_1User.html">User Class</a><br>
2288 Superclass: <a href="#Value"><tt>Value</tt></a></p>
2290 <p>The <tt>User</tt> class is the common base class of all LLVM nodes that may
2291 refer to <a href="#Value"><tt>Value</tt></a>s. It exposes a list of "Operands"
2292 that are all of the <a href="#Value"><tt>Value</tt></a>s that the User is
2293 referring to. The <tt>User</tt> class itself is a subclass of
2296 <p>The operands of a <tt>User</tt> point directly to the LLVM <a
2297 href="#Value"><tt>Value</tt></a> that it refers to. Because LLVM uses Static
2298 Single Assignment (SSA) form, there can only be one definition referred to,
2299 allowing this direct connection. This connection provides the use-def
2300 information in LLVM.</p>
2304 <!-- _______________________________________________________________________ -->
2305 <div class="doc_subsubsection">
2306 <a name="m_User">Important Public Members of the <tt>User</tt> class</a>
2309 <div class="doc_text">
2311 <p>The <tt>User</tt> class exposes the operand list in two ways: through
2312 an index access interface and through an iterator based interface.</p>
2315 <li><tt>Value *getOperand(unsigned i)</tt><br>
2316 <tt>unsigned getNumOperands()</tt>
2317 <p> These two methods expose the operands of the <tt>User</tt> in a
2318 convenient form for direct access.</p></li>
2320 <li><tt>User::op_iterator</tt> - Typedef for iterator over the operand
2322 <tt>op_iterator op_begin()</tt> - Get an iterator to the start of
2323 the operand list.<br>
2324 <tt>op_iterator op_end()</tt> - Get an iterator to the end of the
2326 <p> Together, these methods make up the iterator based interface to
2327 the operands of a <tt>User</tt>.</p></li>
2332 <!-- ======================================================================= -->
2333 <div class="doc_subsection">
2334 <a name="Instruction">The <tt>Instruction</tt> class</a>
2337 <div class="doc_text">
2339 <p><tt>#include "</tt><tt><a
2340 href="/doxygen/Instruction_8h-source.html">llvm/Instruction.h</a>"</tt><br>
2341 doxygen info: <a href="/doxygen/classllvm_1_1Instruction.html">Instruction Class</a><br>
2342 Superclasses: <a href="#User"><tt>User</tt></a>, <a
2343 href="#Value"><tt>Value</tt></a></p>
2345 <p>The <tt>Instruction</tt> class is the common base class for all LLVM
2346 instructions. It provides only a few methods, but is a very commonly used
2347 class. The primary data tracked by the <tt>Instruction</tt> class itself is the
2348 opcode (instruction type) and the parent <a
2349 href="#BasicBlock"><tt>BasicBlock</tt></a> the <tt>Instruction</tt> is embedded
2350 into. To represent a specific type of instruction, one of many subclasses of
2351 <tt>Instruction</tt> are used.</p>
2353 <p> Because the <tt>Instruction</tt> class subclasses the <a
2354 href="#User"><tt>User</tt></a> class, its operands can be accessed in the same
2355 way as for other <a href="#User"><tt>User</tt></a>s (with the
2356 <tt>getOperand()</tt>/<tt>getNumOperands()</tt> and
2357 <tt>op_begin()</tt>/<tt>op_end()</tt> methods).</p> <p> An important file for
2358 the <tt>Instruction</tt> class is the <tt>llvm/Instruction.def</tt> file. This
2359 file contains some meta-data about the various different types of instructions
2360 in LLVM. It describes the enum values that are used as opcodes (for example
2361 <tt>Instruction::Add</tt> and <tt>Instruction::ICmp</tt>), as well as the
2362 concrete sub-classes of <tt>Instruction</tt> that implement the instruction (for
2363 example <tt><a href="#BinaryOperator">BinaryOperator</a></tt> and <tt><a
2364 href="#CmpInst">CmpInst</a></tt>). Unfortunately, the use of macros in
2365 this file confuses doxygen, so these enum values don't show up correctly in the
2366 <a href="/doxygen/classllvm_1_1Instruction.html">doxygen output</a>.</p>
2370 <!-- _______________________________________________________________________ -->
2371 <div class="doc_subsubsection">
2372 <a name="s_Instruction">Important Subclasses of the <tt>Instruction</tt>
2375 <div class="doc_text">
2377 <li><tt><a name="BinaryOperator">BinaryOperator</a></tt>
2378 <p>This subclasses represents all two operand instructions whose operands
2379 must be the same type, except for the comparison instructions.</p></li>
2380 <li><tt><a name="CastInst">CastInst</a></tt>
2381 <p>This subclass is the parent of the 12 casting instructions. It provides
2382 common operations on cast instructions.</p>
2383 <li><tt><a name="CmpInst">CmpInst</a></tt>
2384 <p>This subclass respresents the two comparison instructions,
2385 <a href="LangRef.html#i_icmp">ICmpInst</a> (integer opreands), and
2386 <a href="LangRef.html#i_fcmp">FCmpInst</a> (floating point operands).</p>
2387 <li><tt><a name="TerminatorInst">TerminatorInst</a></tt>
2388 <p>This subclass is the parent of all terminator instructions (those which
2389 can terminate a block).</p>
2393 <!-- _______________________________________________________________________ -->
2394 <div class="doc_subsubsection">
2395 <a name="m_Instruction">Important Public Members of the <tt>Instruction</tt>
2399 <div class="doc_text">
2402 <li><tt><a href="#BasicBlock">BasicBlock</a> *getParent()</tt>
2403 <p>Returns the <a href="#BasicBlock"><tt>BasicBlock</tt></a> that
2404 this <tt>Instruction</tt> is embedded into.</p></li>
2405 <li><tt>bool mayWriteToMemory()</tt>
2406 <p>Returns true if the instruction writes to memory, i.e. it is a
2407 <tt>call</tt>,<tt>free</tt>,<tt>invoke</tt>, or <tt>store</tt>.</p></li>
2408 <li><tt>unsigned getOpcode()</tt>
2409 <p>Returns the opcode for the <tt>Instruction</tt>.</p></li>
2410 <li><tt><a href="#Instruction">Instruction</a> *clone() const</tt>
2411 <p>Returns another instance of the specified instruction, identical
2412 in all ways to the original except that the instruction has no parent
2413 (ie it's not embedded into a <a href="#BasicBlock"><tt>BasicBlock</tt></a>),
2414 and it has no name</p></li>
2419 <!-- ======================================================================= -->
2420 <div class="doc_subsection">
2421 <a name="BasicBlock">The <tt>BasicBlock</tt> class</a>
2424 <div class="doc_text">
2427 href="/doxygen/BasicBlock_8h-source.html">llvm/BasicBlock.h</a>"</tt><br>
2428 doxygen info: <a href="/doxygen/structllvm_1_1BasicBlock.html">BasicBlock
2430 Superclass: <a href="#Value"><tt>Value</tt></a></p>
2432 <p>This class represents a single entry multiple exit section of the code,
2433 commonly known as a basic block by the compiler community. The
2434 <tt>BasicBlock</tt> class maintains a list of <a
2435 href="#Instruction"><tt>Instruction</tt></a>s, which form the body of the block.
2436 Matching the language definition, the last element of this list of instructions
2437 is always a terminator instruction (a subclass of the <a
2438 href="#TerminatorInst"><tt>TerminatorInst</tt></a> class).</p>
2440 <p>In addition to tracking the list of instructions that make up the block, the
2441 <tt>BasicBlock</tt> class also keeps track of the <a
2442 href="#Function"><tt>Function</tt></a> that it is embedded into.</p>
2444 <p>Note that <tt>BasicBlock</tt>s themselves are <a
2445 href="#Value"><tt>Value</tt></a>s, because they are referenced by instructions
2446 like branches and can go in the switch tables. <tt>BasicBlock</tt>s have type
2451 <!-- _______________________________________________________________________ -->
2452 <div class="doc_subsubsection">
2453 <a name="m_BasicBlock">Important Public Members of the <tt>BasicBlock</tt>
2457 <div class="doc_text">
2461 <li><tt>BasicBlock(const std::string &Name = "", </tt><tt><a
2462 href="#Function">Function</a> *Parent = 0)</tt>
2464 <p>The <tt>BasicBlock</tt> constructor is used to create new basic blocks for
2465 insertion into a function. The constructor optionally takes a name for the new
2466 block, and a <a href="#Function"><tt>Function</tt></a> to insert it into. If
2467 the <tt>Parent</tt> parameter is specified, the new <tt>BasicBlock</tt> is
2468 automatically inserted at the end of the specified <a
2469 href="#Function"><tt>Function</tt></a>, if not specified, the BasicBlock must be
2470 manually inserted into the <a href="#Function"><tt>Function</tt></a>.</p></li>
2472 <li><tt>BasicBlock::iterator</tt> - Typedef for instruction list iterator<br>
2473 <tt>BasicBlock::const_iterator</tt> - Typedef for const_iterator.<br>
2474 <tt>begin()</tt>, <tt>end()</tt>, <tt>front()</tt>, <tt>back()</tt>,
2475 <tt>size()</tt>, <tt>empty()</tt>
2476 STL-style functions for accessing the instruction list.
2478 <p>These methods and typedefs are forwarding functions that have the same
2479 semantics as the standard library methods of the same names. These methods
2480 expose the underlying instruction list of a basic block in a way that is easy to
2481 manipulate. To get the full complement of container operations (including
2482 operations to update the list), you must use the <tt>getInstList()</tt>
2485 <li><tt>BasicBlock::InstListType &getInstList()</tt>
2487 <p>This method is used to get access to the underlying container that actually
2488 holds the Instructions. This method must be used when there isn't a forwarding
2489 function in the <tt>BasicBlock</tt> class for the operation that you would like
2490 to perform. Because there are no forwarding functions for "updating"
2491 operations, you need to use this if you want to update the contents of a
2492 <tt>BasicBlock</tt>.</p></li>
2494 <li><tt><a href="#Function">Function</a> *getParent()</tt>
2496 <p> Returns a pointer to <a href="#Function"><tt>Function</tt></a> the block is
2497 embedded into, or a null pointer if it is homeless.</p></li>
2499 <li><tt><a href="#TerminatorInst">TerminatorInst</a> *getTerminator()</tt>
2501 <p> Returns a pointer to the terminator instruction that appears at the end of
2502 the <tt>BasicBlock</tt>. If there is no terminator instruction, or if the last
2503 instruction in the block is not a terminator, then a null pointer is
2510 <!-- ======================================================================= -->
2511 <div class="doc_subsection">
2512 <a name="GlobalValue">The <tt>GlobalValue</tt> class</a>
2515 <div class="doc_text">
2518 href="/doxygen/GlobalValue_8h-source.html">llvm/GlobalValue.h</a>"</tt><br>
2519 doxygen info: <a href="/doxygen/classllvm_1_1GlobalValue.html">GlobalValue
2521 Superclasses: <a href="#Constant"><tt>Constant</tt></a>,
2522 <a href="#User"><tt>User</tt></a>, <a href="#Value"><tt>Value</tt></a></p>
2524 <p>Global values (<a href="#GlobalVariable"><tt>GlobalVariable</tt></a>s or <a
2525 href="#Function"><tt>Function</tt></a>s) are the only LLVM values that are
2526 visible in the bodies of all <a href="#Function"><tt>Function</tt></a>s.
2527 Because they are visible at global scope, they are also subject to linking with
2528 other globals defined in different translation units. To control the linking
2529 process, <tt>GlobalValue</tt>s know their linkage rules. Specifically,
2530 <tt>GlobalValue</tt>s know whether they have internal or external linkage, as
2531 defined by the <tt>LinkageTypes</tt> enumeration.</p>
2533 <p>If a <tt>GlobalValue</tt> has internal linkage (equivalent to being
2534 <tt>static</tt> in C), it is not visible to code outside the current translation
2535 unit, and does not participate in linking. If it has external linkage, it is
2536 visible to external code, and does participate in linking. In addition to
2537 linkage information, <tt>GlobalValue</tt>s keep track of which <a
2538 href="#Module"><tt>Module</tt></a> they are currently part of.</p>
2540 <p>Because <tt>GlobalValue</tt>s are memory objects, they are always referred to
2541 by their <b>address</b>. As such, the <a href="#Type"><tt>Type</tt></a> of a
2542 global is always a pointer to its contents. It is important to remember this
2543 when using the <tt>GetElementPtrInst</tt> instruction because this pointer must
2544 be dereferenced first. For example, if you have a <tt>GlobalVariable</tt> (a
2545 subclass of <tt>GlobalValue)</tt> that is an array of 24 ints, type <tt>[24 x
2546 i32]</tt>, then the <tt>GlobalVariable</tt> is a pointer to that array. Although
2547 the address of the first element of this array and the value of the
2548 <tt>GlobalVariable</tt> are the same, they have different types. The
2549 <tt>GlobalVariable</tt>'s type is <tt>[24 x i32]</tt>. The first element's type
2550 is <tt>i32.</tt> Because of this, accessing a global value requires you to
2551 dereference the pointer with <tt>GetElementPtrInst</tt> first, then its elements
2552 can be accessed. This is explained in the <a href="LangRef.html#globalvars">LLVM
2553 Language Reference Manual</a>.</p>
2557 <!-- _______________________________________________________________________ -->
2558 <div class="doc_subsubsection">
2559 <a name="m_GlobalValue">Important Public Members of the <tt>GlobalValue</tt>
2563 <div class="doc_text">
2566 <li><tt>bool hasInternalLinkage() const</tt><br>
2567 <tt>bool hasExternalLinkage() const</tt><br>
2568 <tt>void setInternalLinkage(bool HasInternalLinkage)</tt>
2569 <p> These methods manipulate the linkage characteristics of the <tt>GlobalValue</tt>.</p>
2572 <li><tt><a href="#Module">Module</a> *getParent()</tt>
2573 <p> This returns the <a href="#Module"><tt>Module</tt></a> that the
2574 GlobalValue is currently embedded into.</p></li>
2579 <!-- ======================================================================= -->
2580 <div class="doc_subsection">
2581 <a name="Function">The <tt>Function</tt> class</a>
2584 <div class="doc_text">
2587 href="/doxygen/Function_8h-source.html">llvm/Function.h</a>"</tt><br> doxygen
2588 info: <a href="/doxygen/classllvm_1_1Function.html">Function Class</a><br>
2589 Superclasses: <a href="#GlobalValue"><tt>GlobalValue</tt></a>,
2590 <a href="#Constant"><tt>Constant</tt></a>,
2591 <a href="#User"><tt>User</tt></a>,
2592 <a href="#Value"><tt>Value</tt></a></p>
2594 <p>The <tt>Function</tt> class represents a single procedure in LLVM. It is
2595 actually one of the more complex classes in the LLVM heirarchy because it must
2596 keep track of a large amount of data. The <tt>Function</tt> class keeps track
2597 of a list of <a href="#BasicBlock"><tt>BasicBlock</tt></a>s, a list of formal
2598 <a href="#Argument"><tt>Argument</tt></a>s, and a
2599 <a href="#SymbolTable"><tt>SymbolTable</tt></a>.</p>
2601 <p>The list of <a href="#BasicBlock"><tt>BasicBlock</tt></a>s is the most
2602 commonly used part of <tt>Function</tt> objects. The list imposes an implicit
2603 ordering of the blocks in the function, which indicate how the code will be
2604 layed out by the backend. Additionally, the first <a
2605 href="#BasicBlock"><tt>BasicBlock</tt></a> is the implicit entry node for the
2606 <tt>Function</tt>. It is not legal in LLVM to explicitly branch to this initial
2607 block. There are no implicit exit nodes, and in fact there may be multiple exit
2608 nodes from a single <tt>Function</tt>. If the <a
2609 href="#BasicBlock"><tt>BasicBlock</tt></a> list is empty, this indicates that
2610 the <tt>Function</tt> is actually a function declaration: the actual body of the
2611 function hasn't been linked in yet.</p>
2613 <p>In addition to a list of <a href="#BasicBlock"><tt>BasicBlock</tt></a>s, the
2614 <tt>Function</tt> class also keeps track of the list of formal <a
2615 href="#Argument"><tt>Argument</tt></a>s that the function receives. This
2616 container manages the lifetime of the <a href="#Argument"><tt>Argument</tt></a>
2617 nodes, just like the <a href="#BasicBlock"><tt>BasicBlock</tt></a> list does for
2618 the <a href="#BasicBlock"><tt>BasicBlock</tt></a>s.</p>
2620 <p>The <a href="#SymbolTable"><tt>SymbolTable</tt></a> is a very rarely used
2621 LLVM feature that is only used when you have to look up a value by name. Aside
2622 from that, the <a href="#SymbolTable"><tt>SymbolTable</tt></a> is used
2623 internally to make sure that there are not conflicts between the names of <a
2624 href="#Instruction"><tt>Instruction</tt></a>s, <a
2625 href="#BasicBlock"><tt>BasicBlock</tt></a>s, or <a
2626 href="#Argument"><tt>Argument</tt></a>s in the function body.</p>
2628 <p>Note that <tt>Function</tt> is a <a href="#GlobalValue">GlobalValue</a>
2629 and therefore also a <a href="#Constant">Constant</a>. The value of the function
2630 is its address (after linking) which is guaranteed to be constant.</p>
2633 <!-- _______________________________________________________________________ -->
2634 <div class="doc_subsubsection">
2635 <a name="m_Function">Important Public Members of the <tt>Function</tt>
2639 <div class="doc_text">
2642 <li><tt>Function(const </tt><tt><a href="#FunctionType">FunctionType</a>
2643 *Ty, LinkageTypes Linkage, const std::string &N = "", Module* Parent = 0)</tt>
2645 <p>Constructor used when you need to create new <tt>Function</tt>s to add
2646 the the program. The constructor must specify the type of the function to
2647 create and what type of linkage the function should have. The <a
2648 href="#FunctionType"><tt>FunctionType</tt></a> argument
2649 specifies the formal arguments and return value for the function. The same
2650 <a href="#FunctionTypel"><tt>FunctionType</tt></a> value can be used to
2651 create multiple functions. The <tt>Parent</tt> argument specifies the Module
2652 in which the function is defined. If this argument is provided, the function
2653 will automatically be inserted into that module's list of
2656 <li><tt>bool isExternal()</tt>
2658 <p>Return whether or not the <tt>Function</tt> has a body defined. If the
2659 function is "external", it does not have a body, and thus must be resolved
2660 by linking with a function defined in a different translation unit.</p></li>
2662 <li><tt>Function::iterator</tt> - Typedef for basic block list iterator<br>
2663 <tt>Function::const_iterator</tt> - Typedef for const_iterator.<br>
2665 <tt>begin()</tt>, <tt>end()</tt>
2666 <tt>size()</tt>, <tt>empty()</tt>
2668 <p>These are forwarding methods that make it easy to access the contents of
2669 a <tt>Function</tt> object's <a href="#BasicBlock"><tt>BasicBlock</tt></a>
2672 <li><tt>Function::BasicBlockListType &getBasicBlockList()</tt>
2674 <p>Returns the list of <a href="#BasicBlock"><tt>BasicBlock</tt></a>s. This
2675 is necessary to use when you need to update the list or perform a complex
2676 action that doesn't have a forwarding method.</p></li>
2678 <li><tt>Function::arg_iterator</tt> - Typedef for the argument list
2680 <tt>Function::const_arg_iterator</tt> - Typedef for const_iterator.<br>
2682 <tt>arg_begin()</tt>, <tt>arg_end()</tt>
2683 <tt>arg_size()</tt>, <tt>arg_empty()</tt>
2685 <p>These are forwarding methods that make it easy to access the contents of
2686 a <tt>Function</tt> object's <a href="#Argument"><tt>Argument</tt></a>
2689 <li><tt>Function::ArgumentListType &getArgumentList()</tt>
2691 <p>Returns the list of <a href="#Argument"><tt>Argument</tt></a>s. This is
2692 necessary to use when you need to update the list or perform a complex
2693 action that doesn't have a forwarding method.</p></li>
2695 <li><tt><a href="#BasicBlock">BasicBlock</a> &getEntryBlock()</tt>
2697 <p>Returns the entry <a href="#BasicBlock"><tt>BasicBlock</tt></a> for the
2698 function. Because the entry block for the function is always the first
2699 block, this returns the first block of the <tt>Function</tt>.</p></li>
2701 <li><tt><a href="#Type">Type</a> *getReturnType()</tt><br>
2702 <tt><a href="#FunctionType">FunctionType</a> *getFunctionType()</tt>
2704 <p>This traverses the <a href="#Type"><tt>Type</tt></a> of the
2705 <tt>Function</tt> and returns the return type of the function, or the <a
2706 href="#FunctionType"><tt>FunctionType</tt></a> of the actual
2709 <li><tt><a href="#SymbolTable">SymbolTable</a> *getSymbolTable()</tt>
2711 <p> Return a pointer to the <a href="#SymbolTable"><tt>SymbolTable</tt></a>
2712 for this <tt>Function</tt>.</p></li>
2717 <!-- ======================================================================= -->
2718 <div class="doc_subsection">
2719 <a name="GlobalVariable">The <tt>GlobalVariable</tt> class</a>
2722 <div class="doc_text">
2725 href="/doxygen/GlobalVariable_8h-source.html">llvm/GlobalVariable.h</a>"</tt>
2727 doxygen info: <a href="/doxygen/classllvm_1_1GlobalVariable.html">GlobalVariable
2729 Superclasses: <a href="#GlobalValue"><tt>GlobalValue</tt></a>,
2730 <a href="#Constant"><tt>Constant</tt></a>,
2731 <a href="#User"><tt>User</tt></a>,
2732 <a href="#Value"><tt>Value</tt></a></p>
2734 <p>Global variables are represented with the (suprise suprise)
2735 <tt>GlobalVariable</tt> class. Like functions, <tt>GlobalVariable</tt>s are also
2736 subclasses of <a href="#GlobalValue"><tt>GlobalValue</tt></a>, and as such are
2737 always referenced by their address (global values must live in memory, so their
2738 "name" refers to their constant address). See
2739 <a href="#GlobalValue"><tt>GlobalValue</tt></a> for more on this. Global
2740 variables may have an initial value (which must be a
2741 <a href="#Constant"><tt>Constant</tt></a>), and if they have an initializer,
2742 they may be marked as "constant" themselves (indicating that their contents
2743 never change at runtime).</p>
2746 <!-- _______________________________________________________________________ -->
2747 <div class="doc_subsubsection">
2748 <a name="m_GlobalVariable">Important Public Members of the
2749 <tt>GlobalVariable</tt> class</a>
2752 <div class="doc_text">
2755 <li><tt>GlobalVariable(const </tt><tt><a href="#Type">Type</a> *Ty, bool
2756 isConstant, LinkageTypes& Linkage, <a href="#Constant">Constant</a>
2757 *Initializer = 0, const std::string &Name = "", Module* Parent = 0)</tt>
2759 <p>Create a new global variable of the specified type. If
2760 <tt>isConstant</tt> is true then the global variable will be marked as
2761 unchanging for the program. The Linkage parameter specifies the type of
2762 linkage (internal, external, weak, linkonce, appending) for the variable. If
2763 the linkage is InternalLinkage, WeakLinkage, or LinkOnceLinkage, then
2764 the resultant global variable will have internal linkage. AppendingLinkage
2765 concatenates together all instances (in different translation units) of the
2766 variable into a single variable but is only applicable to arrays. See
2767 the <a href="LangRef.html#modulestructure">LLVM Language Reference</a> for
2768 further details on linkage types. Optionally an initializer, a name, and the
2769 module to put the variable into may be specified for the global variable as
2772 <li><tt>bool isConstant() const</tt>
2774 <p>Returns true if this is a global variable that is known not to
2775 be modified at runtime.</p></li>
2777 <li><tt>bool hasInitializer()</tt>
2779 <p>Returns true if this <tt>GlobalVariable</tt> has an intializer.</p></li>
2781 <li><tt><a href="#Constant">Constant</a> *getInitializer()</tt>
2783 <p>Returns the intial value for a <tt>GlobalVariable</tt>. It is not legal
2784 to call this method if there is no initializer.</p></li>
2789 <!-- ======================================================================= -->
2790 <div class="doc_subsection">
2791 <a name="Module">The <tt>Module</tt> class</a>
2794 <div class="doc_text">
2797 href="/doxygen/Module_8h-source.html">llvm/Module.h</a>"</tt><br> doxygen info:
2798 <a href="/doxygen/classllvm_1_1Module.html">Module Class</a></p>
2800 <p>The <tt>Module</tt> class represents the top level structure present in LLVM
2801 programs. An LLVM module is effectively either a translation unit of the
2802 original program or a combination of several translation units merged by the
2803 linker. The <tt>Module</tt> class keeps track of a list of <a
2804 href="#Function"><tt>Function</tt></a>s, a list of <a
2805 href="#GlobalVariable"><tt>GlobalVariable</tt></a>s, and a <a
2806 href="#SymbolTable"><tt>SymbolTable</tt></a>. Additionally, it contains a few
2807 helpful member functions that try to make common operations easy.</p>
2811 <!-- _______________________________________________________________________ -->
2812 <div class="doc_subsubsection">
2813 <a name="m_Module">Important Public Members of the <tt>Module</tt> class</a>
2816 <div class="doc_text">
2819 <li><tt>Module::Module(std::string name = "")</tt></li>
2822 <p>Constructing a <a href="#Module">Module</a> is easy. You can optionally
2823 provide a name for it (probably based on the name of the translation unit).</p>
2826 <li><tt>Module::iterator</tt> - Typedef for function list iterator<br>
2827 <tt>Module::const_iterator</tt> - Typedef for const_iterator.<br>
2829 <tt>begin()</tt>, <tt>end()</tt>
2830 <tt>size()</tt>, <tt>empty()</tt>
2832 <p>These are forwarding methods that make it easy to access the contents of
2833 a <tt>Module</tt> object's <a href="#Function"><tt>Function</tt></a>
2836 <li><tt>Module::FunctionListType &getFunctionList()</tt>
2838 <p> Returns the list of <a href="#Function"><tt>Function</tt></a>s. This is
2839 necessary to use when you need to update the list or perform a complex
2840 action that doesn't have a forwarding method.</p>
2842 <p><!-- Global Variable --></p></li>
2848 <li><tt>Module::global_iterator</tt> - Typedef for global variable list iterator<br>
2850 <tt>Module::const_global_iterator</tt> - Typedef for const_iterator.<br>
2852 <tt>global_begin()</tt>, <tt>global_end()</tt>
2853 <tt>global_size()</tt>, <tt>global_empty()</tt>
2855 <p> These are forwarding methods that make it easy to access the contents of
2856 a <tt>Module</tt> object's <a
2857 href="#GlobalVariable"><tt>GlobalVariable</tt></a> list.</p></li>
2859 <li><tt>Module::GlobalListType &getGlobalList()</tt>
2861 <p>Returns the list of <a
2862 href="#GlobalVariable"><tt>GlobalVariable</tt></a>s. This is necessary to
2863 use when you need to update the list or perform a complex action that
2864 doesn't have a forwarding method.</p>
2866 <p><!-- Symbol table stuff --> </p></li>
2872 <li><tt><a href="#SymbolTable">SymbolTable</a> *getSymbolTable()</tt>
2874 <p>Return a reference to the <a href="#SymbolTable"><tt>SymbolTable</tt></a>
2875 for this <tt>Module</tt>.</p>
2877 <p><!-- Convenience methods --></p></li>
2883 <li><tt><a href="#Function">Function</a> *getFunction(const std::string
2884 &Name, const <a href="#FunctionType">FunctionType</a> *Ty)</tt>
2886 <p>Look up the specified function in the <tt>Module</tt> <a
2887 href="#SymbolTable"><tt>SymbolTable</tt></a>. If it does not exist, return
2888 <tt>null</tt>.</p></li>
2890 <li><tt><a href="#Function">Function</a> *getOrInsertFunction(const
2891 std::string &Name, const <a href="#FunctionType">FunctionType</a> *T)</tt>
2893 <p>Look up the specified function in the <tt>Module</tt> <a
2894 href="#SymbolTable"><tt>SymbolTable</tt></a>. If it does not exist, add an
2895 external declaration for the function and return it.</p></li>
2897 <li><tt>std::string getTypeName(const <a href="#Type">Type</a> *Ty)</tt>
2899 <p>If there is at least one entry in the <a
2900 href="#SymbolTable"><tt>SymbolTable</tt></a> for the specified <a
2901 href="#Type"><tt>Type</tt></a>, return it. Otherwise return the empty
2904 <li><tt>bool addTypeName(const std::string &Name, const <a
2905 href="#Type">Type</a> *Ty)</tt>
2907 <p>Insert an entry in the <a href="#SymbolTable"><tt>SymbolTable</tt></a>
2908 mapping <tt>Name</tt> to <tt>Ty</tt>. If there is already an entry for this
2909 name, true is returned and the <a
2910 href="#SymbolTable"><tt>SymbolTable</tt></a> is not modified.</p></li>
2915 <!-- ======================================================================= -->
2916 <div class="doc_subsection">
2917 <a name="Constant">The <tt>Constant</tt> class and subclasses</a>
2920 <div class="doc_text">
2922 <p>Constant represents a base class for different types of constants. It
2923 is subclassed by ConstantInt, ConstantArray, etc. for representing
2924 the various types of Constants.</p>
2928 <!-- _______________________________________________________________________ -->
2929 <div class="doc_subsubsection">
2930 <a name="m_Constant">Important Public Methods</a>
2932 <div class="doc_text">
2935 <!-- _______________________________________________________________________ -->
2936 <div class="doc_subsubsection">Important Subclasses of Constant </div>
2937 <div class="doc_text">
2939 <li>ConstantInt : This subclass of Constant represents an integer constant of
2940 any width, including boolean (1 bit integer).
2942 <li><tt>int64_t getSExtValue() const</tt>: Returns the underlying value of
2943 this constant as a sign extended signed integer value.</li>
2944 <li><tt>uint64_t getZExtValue() const</tt>: Returns the underlying value
2945 of this constant as a zero extended unsigned integer value.</li>
2946 <li><tt>static ConstantInt* get(const Type *Ty, uint64_t Val)</tt>:
2947 Returns the ConstantInt object that represents the value provided by
2948 <tt>Val</tt> for integer type <tt>Ty</tt>.</li>
2951 <li>ConstantFP : This class represents a floating point constant.
2953 <li><tt>double getValue() const</tt>: Returns the underlying value of
2954 this constant. </li>
2957 <li>ConstantArray : This represents a constant array.
2959 <li><tt>const std::vector<Use> &getValues() const</tt>: Returns
2960 a vector of component constants that makeup this array. </li>
2963 <li>ConstantStruct : This represents a constant struct.
2965 <li><tt>const std::vector<Use> &getValues() const</tt>: Returns
2966 a vector of component constants that makeup this array. </li>
2969 <li>GlobalValue : This represents either a global variable or a function. In
2970 either case, the value is a constant fixed address (after linking).
2974 <!-- ======================================================================= -->
2975 <div class="doc_subsection">
2976 <a name="Argument">The <tt>Argument</tt> class</a>
2979 <div class="doc_text">
2981 <p>This subclass of Value defines the interface for incoming formal
2982 arguments to a function. A Function maintains a list of its formal
2983 arguments. An argument has a pointer to the parent Function.</p>
2987 <!-- *********************************************************************** -->
2990 <a href="http://jigsaw.w3.org/css-validator/check/referer"><img
2991 src="http://jigsaw.w3.org/css-validator/images/vcss" alt="Valid CSS!"></a>
2992 <a href="http://validator.w3.org/check/referer"><img
2993 src="http://www.w3.org/Icons/valid-html401" alt="Valid HTML 4.01!" /></a>
2995 <a href="mailto:dhurjati@cs.uiuc.edu">Dinakar Dhurjati</a> and
2996 <a href="mailto:sabre@nondot.org">Chris Lattner</a><br>
2997 <a href="http://llvm.org">The LLVM Compiler Infrastructure</a><br>
2998 Last modified: $Date$