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.h</a></li>
58 <li><a href="#dss_other">Other Sequential Container Options</a></li>
60 <li><a href="#ds_set">Set-Like Containers (std::set, SmallSet, SetVector, etc)</a>
62 <li><a href="#dss_sortedvectorset">A sorted 'vector'</a></li>
63 <li><a href="#dss_smallset">"llvm/ADT/SmallSet.h"</a></li>
64 <li><a href="#dss_smallptrset">"llvm/ADT/SmallPtrSet.h"</a></li>
65 <li><a href="#dss_denseset">"llvm/ADT/DenseSet.h"</a></li>
66 <li><a href="#dss_FoldingSet">"llvm/ADT/FoldingSet.h"</a></li>
67 <li><a href="#dss_set"><set></a></li>
68 <li><a href="#dss_setvector">"llvm/ADT/SetVector.h"</a></li>
69 <li><a href="#dss_uniquevector">"llvm/ADT/UniqueVector.h"</a></li>
70 <li><a href="#dss_otherset">Other Set-Like ContainerOptions</a></li>
72 <li><a href="#ds_map">Map-Like Containers (std::map, DenseMap, etc)</a>
74 <li><a href="#dss_sortedvectormap">A sorted 'vector'</a></li>
75 <li><a href="#dss_stringmap">"llvm/ADT/StringMap.h"</a></li>
76 <li><a href="#dss_indexedmap">"llvm/ADT/IndexedMap.h"</a></li>
77 <li><a href="#dss_densemap">"llvm/ADT/DenseMap.h"</a></li>
78 <li><a href="#dss_map"><map></a></li>
79 <li><a href="#dss_othermap">Other Map-Like Container Options</a></li>
81 <li><a href="#ds_bit">BitVector-like containers</a>
83 <li><a href="#dss_bitvector">A dense bitvector</a></li>
84 <li><a href="#dss_sparsebitvector">A sparse bitvector</a></li>
88 <li><a href="#common">Helpful Hints for Common Operations</a>
90 <li><a href="#inspection">Basic Inspection and Traversal Routines</a>
92 <li><a href="#iterate_function">Iterating over the <tt>BasicBlock</tt>s
93 in a <tt>Function</tt></a> </li>
94 <li><a href="#iterate_basicblock">Iterating over the <tt>Instruction</tt>s
95 in a <tt>BasicBlock</tt></a> </li>
96 <li><a href="#iterate_institer">Iterating over the <tt>Instruction</tt>s
97 in a <tt>Function</tt></a> </li>
98 <li><a href="#iterate_convert">Turning an iterator into a
99 class pointer</a> </li>
100 <li><a href="#iterate_complex">Finding call sites: a more
101 complex example</a> </li>
102 <li><a href="#calls_and_invokes">Treating calls and invokes
103 the same way</a> </li>
104 <li><a href="#iterate_chains">Iterating over def-use &
105 use-def chains</a> </li>
106 <li><a href="#iterate_preds">Iterating over predecessors &
107 successors of blocks</a></li>
110 <li><a href="#simplechanges">Making simple changes</a>
112 <li><a href="#schanges_creating">Creating and inserting new
113 <tt>Instruction</tt>s</a> </li>
114 <li><a href="#schanges_deleting">Deleting <tt>Instruction</tt>s</a> </li>
115 <li><a href="#schanges_replacing">Replacing an <tt>Instruction</tt>
116 with another <tt>Value</tt></a> </li>
117 <li><a href="#schanges_deletingGV">Deleting <tt>GlobalVariable</tt>s</a> </li>
120 <li><a href="#create_types">How to Create Types</a></li>
122 <li>Working with the Control Flow Graph
124 <li>Accessing predecessors and successors of a <tt>BasicBlock</tt>
132 <li><a href="#threading">Threads and LLVM</a>
134 <li><a href="#startmultithreaded">Entering threaded mode with <tt>llvm_start_multithreaded()</tt><a/></li>
135 <li><a href="#shutdown">Ending execution with <tt>llvm_shutdown()</tt></a></li>
136 <li><a href="#managedstatic">Lazy initialization with <tt>ManagedStatic</tt></a></li>
140 <li><a href="#advanced">Advanced Topics</a>
142 <li><a href="#TypeResolve">LLVM Type Resolution</a>
144 <li><a href="#BuildRecType">Basic Recursive Type Construction</a></li>
145 <li><a href="#refineAbstractTypeTo">The <tt>refineAbstractTypeTo</tt> method</a></li>
146 <li><a href="#PATypeHolder">The PATypeHolder Class</a></li>
147 <li><a href="#AbstractTypeUser">The AbstractTypeUser Class</a></li>
150 <li><a href="#SymbolTable">The <tt>ValueSymbolTable</tt> and <tt>TypeSymbolTable</tt> classes</a></li>
151 <li><a href="#UserLayout">The <tt>User</tt> and owned <tt>Use</tt> classes' memory layout</a></li>
154 <li><a href="#coreclasses">The Core LLVM Class Hierarchy Reference</a>
156 <li><a href="#Type">The <tt>Type</tt> class</a> </li>
157 <li><a href="#Module">The <tt>Module</tt> class</a></li>
158 <li><a href="#Value">The <tt>Value</tt> class</a>
160 <li><a href="#User">The <tt>User</tt> class</a>
162 <li><a href="#Instruction">The <tt>Instruction</tt> class</a></li>
163 <li><a href="#Constant">The <tt>Constant</tt> class</a>
165 <li><a href="#GlobalValue">The <tt>GlobalValue</tt> class</a>
167 <li><a href="#Function">The <tt>Function</tt> class</a></li>
168 <li><a href="#GlobalVariable">The <tt>GlobalVariable</tt> class</a></li>
175 <li><a href="#BasicBlock">The <tt>BasicBlock</tt> class</a></li>
176 <li><a href="#Argument">The <tt>Argument</tt> class</a></li>
183 <div class="doc_author">
184 <p>Written by <a href="mailto:sabre@nondot.org">Chris Lattner</a>,
185 <a href="mailto:dhurjati@cs.uiuc.edu">Dinakar Dhurjati</a>,
186 <a href="mailto:ggreif@gmail.com">Gabor Greif</a>,
187 <a href="mailto:jstanley@cs.uiuc.edu">Joel Stanley</a>,
188 <a href="mailto:rspencer@x10sys.com">Reid Spencer</a> and
189 <a href="mailto:owen@apple.com">Owen Anderson</a></p>
192 <!-- *********************************************************************** -->
193 <div class="doc_section">
194 <a name="introduction">Introduction </a>
196 <!-- *********************************************************************** -->
198 <div class="doc_text">
200 <p>This document is meant to highlight some of the important classes and
201 interfaces available in the LLVM source-base. This manual is not
202 intended to explain what LLVM is, how it works, and what LLVM code looks
203 like. It assumes that you know the basics of LLVM and are interested
204 in writing transformations or otherwise analyzing or manipulating the
207 <p>This document should get you oriented so that you can find your
208 way in the continuously growing source code that makes up the LLVM
209 infrastructure. Note that this manual is not intended to serve as a
210 replacement for reading the source code, so if you think there should be
211 a method in one of these classes to do something, but it's not listed,
212 check the source. Links to the <a href="/doxygen/">doxygen</a> sources
213 are provided to make this as easy as possible.</p>
215 <p>The first section of this document describes general information that is
216 useful to know when working in the LLVM infrastructure, and the second describes
217 the Core LLVM classes. In the future this manual will be extended with
218 information describing how to use extension libraries, such as dominator
219 information, CFG traversal routines, and useful utilities like the <tt><a
220 href="/doxygen/InstVisitor_8h-source.html">InstVisitor</a></tt> template.</p>
224 <!-- *********************************************************************** -->
225 <div class="doc_section">
226 <a name="general">General Information</a>
228 <!-- *********************************************************************** -->
230 <div class="doc_text">
232 <p>This section contains general information that is useful if you are working
233 in the LLVM source-base, but that isn't specific to any particular API.</p>
237 <!-- ======================================================================= -->
238 <div class="doc_subsection">
239 <a name="stl">The C++ Standard Template Library</a>
242 <div class="doc_text">
244 <p>LLVM makes heavy use of the C++ Standard Template Library (STL),
245 perhaps much more than you are used to, or have seen before. Because of
246 this, you might want to do a little background reading in the
247 techniques used and capabilities of the library. There are many good
248 pages that discuss the STL, and several books on the subject that you
249 can get, so it will not be discussed in this document.</p>
251 <p>Here are some useful links:</p>
255 <li><a href="http://www.dinkumware.com/refxcpp.html">Dinkumware C++ Library
256 reference</a> - an excellent reference for the STL and other parts of the
257 standard C++ library.</li>
259 <li><a href="http://www.tempest-sw.com/cpp/">C++ In a Nutshell</a> - This is an
260 O'Reilly book in the making. It has a decent Standard Library
261 Reference that rivals Dinkumware's, and is unfortunately no longer free since the
262 book has been published.</li>
264 <li><a href="http://www.parashift.com/c++-faq-lite/">C++ Frequently Asked
267 <li><a href="http://www.sgi.com/tech/stl/">SGI's STL Programmer's Guide</a> -
269 href="http://www.sgi.com/tech/stl/stl_introduction.html">Introduction to the
272 <li><a href="http://www.research.att.com/%7Ebs/C++.html">Bjarne Stroustrup's C++
275 <li><a href="http://64.78.49.204/">
276 Bruce Eckel's Thinking in C++, 2nd ed. Volume 2 Revision 4.0 (even better, get
281 <p>You are also encouraged to take a look at the <a
282 href="CodingStandards.html">LLVM Coding Standards</a> guide which focuses on how
283 to write maintainable code more than where to put your curly braces.</p>
287 <!-- ======================================================================= -->
288 <div class="doc_subsection">
289 <a name="stl">Other useful references</a>
292 <div class="doc_text">
295 <li><a href="http://www.psc.edu/%7Esemke/cvs_branches.html">CVS
296 Branch and Tag Primer</a></li>
297 <li><a href="http://www.fortran-2000.com/ArnaudRecipes/sharedlib.html">Using
298 static and shared libraries across platforms</a></li>
303 <!-- *********************************************************************** -->
304 <div class="doc_section">
305 <a name="apis">Important and useful LLVM APIs</a>
307 <!-- *********************************************************************** -->
309 <div class="doc_text">
311 <p>Here we highlight some LLVM APIs that are generally useful and good to
312 know about when writing transformations.</p>
316 <!-- ======================================================================= -->
317 <div class="doc_subsection">
318 <a name="isa">The <tt>isa<></tt>, <tt>cast<></tt> and
319 <tt>dyn_cast<></tt> templates</a>
322 <div class="doc_text">
324 <p>The LLVM source-base makes extensive use of a custom form of RTTI.
325 These templates have many similarities to the C++ <tt>dynamic_cast<></tt>
326 operator, but they don't have some drawbacks (primarily stemming from
327 the fact that <tt>dynamic_cast<></tt> only works on classes that
328 have a v-table). Because they are used so often, you must know what they
329 do and how they work. All of these templates are defined in the <a
330 href="/doxygen/Casting_8h-source.html"><tt>llvm/Support/Casting.h</tt></a>
331 file (note that you very rarely have to include this file directly).</p>
334 <dt><tt>isa<></tt>: </dt>
336 <dd><p>The <tt>isa<></tt> operator works exactly like the Java
337 "<tt>instanceof</tt>" operator. It returns true or false depending on whether
338 a reference or pointer points to an instance of the specified class. This can
339 be very useful for constraint checking of various sorts (example below).</p>
342 <dt><tt>cast<></tt>: </dt>
344 <dd><p>The <tt>cast<></tt> operator is a "checked cast" operation. It
345 converts a pointer or reference from a base class to a derived class, causing
346 an assertion failure if it is not really an instance of the right type. This
347 should be used in cases where you have some information that makes you believe
348 that something is of the right type. An example of the <tt>isa<></tt>
349 and <tt>cast<></tt> template is:</p>
351 <div class="doc_code">
353 static bool isLoopInvariant(const <a href="#Value">Value</a> *V, const Loop *L) {
354 if (isa<<a href="#Constant">Constant</a>>(V) || isa<<a href="#Argument">Argument</a>>(V) || isa<<a href="#GlobalValue">GlobalValue</a>>(V))
357 // <i>Otherwise, it must be an instruction...</i>
358 return !L->contains(cast<<a href="#Instruction">Instruction</a>>(V)->getParent());
363 <p>Note that you should <b>not</b> use an <tt>isa<></tt> test followed
364 by a <tt>cast<></tt>, for that use the <tt>dyn_cast<></tt>
369 <dt><tt>dyn_cast<></tt>:</dt>
371 <dd><p>The <tt>dyn_cast<></tt> operator is a "checking cast" operation.
372 It checks to see if the operand is of the specified type, and if so, returns a
373 pointer to it (this operator does not work with references). If the operand is
374 not of the correct type, a null pointer is returned. Thus, this works very
375 much like the <tt>dynamic_cast<></tt> operator in C++, and should be
376 used in the same circumstances. Typically, the <tt>dyn_cast<></tt>
377 operator is used in an <tt>if</tt> statement or some other flow control
378 statement like this:</p>
380 <div class="doc_code">
382 if (<a href="#AllocationInst">AllocationInst</a> *AI = dyn_cast<<a href="#AllocationInst">AllocationInst</a>>(Val)) {
388 <p>This form of the <tt>if</tt> statement effectively combines together a call
389 to <tt>isa<></tt> and a call to <tt>cast<></tt> into one
390 statement, which is very convenient.</p>
392 <p>Note that the <tt>dyn_cast<></tt> operator, like C++'s
393 <tt>dynamic_cast<></tt> or Java's <tt>instanceof</tt> operator, can be
394 abused. In particular, you should not use big chained <tt>if/then/else</tt>
395 blocks to check for lots of different variants of classes. If you find
396 yourself wanting to do this, it is much cleaner and more efficient to use the
397 <tt>InstVisitor</tt> class to dispatch over the instruction type directly.</p>
401 <dt><tt>cast_or_null<></tt>: </dt>
403 <dd><p>The <tt>cast_or_null<></tt> operator works just like the
404 <tt>cast<></tt> operator, except that it allows for a null pointer as an
405 argument (which it then propagates). This can sometimes be useful, allowing
406 you to combine several null checks into one.</p></dd>
408 <dt><tt>dyn_cast_or_null<></tt>: </dt>
410 <dd><p>The <tt>dyn_cast_or_null<></tt> operator works just like the
411 <tt>dyn_cast<></tt> operator, except that it allows for a null pointer
412 as an argument (which it then propagates). This can sometimes be useful,
413 allowing you to combine several null checks into one.</p></dd>
417 <p>These five templates can be used with any classes, whether they have a
418 v-table or not. To add support for these templates, you simply need to add
419 <tt>classof</tt> static methods to the class you are interested casting
420 to. Describing this is currently outside the scope of this document, but there
421 are lots of examples in the LLVM source base.</p>
425 <!-- ======================================================================= -->
426 <div class="doc_subsection">
427 <a name="DEBUG">The <tt>DEBUG()</tt> macro and <tt>-debug</tt> option</a>
430 <div class="doc_text">
432 <p>Often when working on your pass you will put a bunch of debugging printouts
433 and other code into your pass. After you get it working, you want to remove
434 it, but you may need it again in the future (to work out new bugs that you run
437 <p> Naturally, because of this, you don't want to delete the debug printouts,
438 but you don't want them to always be noisy. A standard compromise is to comment
439 them out, allowing you to enable them if you need them in the future.</p>
441 <p>The "<tt><a href="/doxygen/Debug_8h-source.html">llvm/Support/Debug.h</a></tt>"
442 file provides a macro named <tt>DEBUG()</tt> that is a much nicer solution to
443 this problem. Basically, you can put arbitrary code into the argument of the
444 <tt>DEBUG</tt> macro, and it is only executed if '<tt>opt</tt>' (or any other
445 tool) is run with the '<tt>-debug</tt>' command line argument:</p>
447 <div class="doc_code">
449 DOUT << "I am here!\n";
453 <p>Then you can run your pass like this:</p>
455 <div class="doc_code">
457 $ opt < a.bc > /dev/null -mypass
458 <i><no output></i>
459 $ opt < a.bc > /dev/null -mypass -debug
464 <p>Using the <tt>DEBUG()</tt> macro instead of a home-brewed solution allows you
465 to not have to create "yet another" command line option for the debug output for
466 your pass. Note that <tt>DEBUG()</tt> macros are disabled for optimized builds,
467 so they do not cause a performance impact at all (for the same reason, they
468 should also not contain side-effects!).</p>
470 <p>One additional nice thing about the <tt>DEBUG()</tt> macro is that you can
471 enable or disable it directly in gdb. Just use "<tt>set DebugFlag=0</tt>" or
472 "<tt>set DebugFlag=1</tt>" from the gdb if the program is running. If the
473 program hasn't been started yet, you can always just run it with
478 <!-- _______________________________________________________________________ -->
479 <div class="doc_subsubsection">
480 <a name="DEBUG_TYPE">Fine grained debug info with <tt>DEBUG_TYPE</tt> and
481 the <tt>-debug-only</tt> option</a>
484 <div class="doc_text">
486 <p>Sometimes you may find yourself in a situation where enabling <tt>-debug</tt>
487 just turns on <b>too much</b> information (such as when working on the code
488 generator). If you want to enable debug information with more fine-grained
489 control, you define the <tt>DEBUG_TYPE</tt> macro and the <tt>-debug</tt> only
490 option as follows:</p>
492 <div class="doc_code">
494 DOUT << "No debug type\n";
496 #define DEBUG_TYPE "foo"
497 DOUT << "'foo' debug type\n";
499 #define DEBUG_TYPE "bar"
500 DOUT << "'bar' debug type\n";
502 #define DEBUG_TYPE ""
503 DOUT << "No debug type (2)\n";
507 <p>Then you can run your pass like this:</p>
509 <div class="doc_code">
511 $ opt < a.bc > /dev/null -mypass
512 <i><no output></i>
513 $ opt < a.bc > /dev/null -mypass -debug
518 $ opt < a.bc > /dev/null -mypass -debug-only=foo
520 $ opt < a.bc > /dev/null -mypass -debug-only=bar
525 <p>Of course, in practice, you should only set <tt>DEBUG_TYPE</tt> at the top of
526 a file, to specify the debug type for the entire module (if you do this before
527 you <tt>#include "llvm/Support/Debug.h"</tt>, you don't have to insert the ugly
528 <tt>#undef</tt>'s). Also, you should use names more meaningful than "foo" and
529 "bar", because there is no system in place to ensure that names do not
530 conflict. If two different modules use the same string, they will all be turned
531 on when the name is specified. This allows, for example, all debug information
532 for instruction scheduling to be enabled with <tt>-debug-type=InstrSched</tt>,
533 even if the source lives in multiple files.</p>
537 <!-- ======================================================================= -->
538 <div class="doc_subsection">
539 <a name="Statistic">The <tt>Statistic</tt> class & <tt>-stats</tt>
543 <div class="doc_text">
546 href="/doxygen/Statistic_8h-source.html">llvm/ADT/Statistic.h</a></tt>" file
547 provides a class named <tt>Statistic</tt> that is used as a unified way to
548 keep track of what the LLVM compiler is doing and how effective various
549 optimizations are. It is useful to see what optimizations are contributing to
550 making a particular program run faster.</p>
552 <p>Often you may run your pass on some big program, and you're interested to see
553 how many times it makes a certain transformation. Although you can do this with
554 hand inspection, or some ad-hoc method, this is a real pain and not very useful
555 for big programs. Using the <tt>Statistic</tt> class makes it very easy to
556 keep track of this information, and the calculated information is presented in a
557 uniform manner with the rest of the passes being executed.</p>
559 <p>There are many examples of <tt>Statistic</tt> uses, but the basics of using
560 it are as follows:</p>
563 <li><p>Define your statistic like this:</p>
565 <div class="doc_code">
567 #define <a href="#DEBUG_TYPE">DEBUG_TYPE</a> "mypassname" <i>// This goes before any #includes.</i>
568 STATISTIC(NumXForms, "The # of times I did stuff");
572 <p>The <tt>STATISTIC</tt> macro defines a static variable, whose name is
573 specified by the first argument. The pass name is taken from the DEBUG_TYPE
574 macro, and the description is taken from the second argument. The variable
575 defined ("NumXForms" in this case) acts like an unsigned integer.</p></li>
577 <li><p>Whenever you make a transformation, bump the counter:</p>
579 <div class="doc_code">
581 ++NumXForms; // <i>I did stuff!</i>
588 <p>That's all you have to do. To get '<tt>opt</tt>' to print out the
589 statistics gathered, use the '<tt>-stats</tt>' option:</p>
591 <div class="doc_code">
593 $ opt -stats -mypassname < program.bc > /dev/null
594 <i>... statistics output ...</i>
598 <p> When running <tt>opt</tt> on a C file from the SPEC benchmark
599 suite, it gives a report that looks like this:</p>
601 <div class="doc_code">
603 7646 bitcodewriter - Number of normal instructions
604 725 bitcodewriter - Number of oversized instructions
605 129996 bitcodewriter - Number of bitcode bytes written
606 2817 raise - Number of insts DCEd or constprop'd
607 3213 raise - Number of cast-of-self removed
608 5046 raise - Number of expression trees converted
609 75 raise - Number of other getelementptr's formed
610 138 raise - Number of load/store peepholes
611 42 deadtypeelim - Number of unused typenames removed from symtab
612 392 funcresolve - Number of varargs functions resolved
613 27 globaldce - Number of global variables removed
614 2 adce - Number of basic blocks removed
615 134 cee - Number of branches revectored
616 49 cee - Number of setcc instruction eliminated
617 532 gcse - Number of loads removed
618 2919 gcse - Number of instructions removed
619 86 indvars - Number of canonical indvars added
620 87 indvars - Number of aux indvars removed
621 25 instcombine - Number of dead inst eliminate
622 434 instcombine - Number of insts combined
623 248 licm - Number of load insts hoisted
624 1298 licm - Number of insts hoisted to a loop pre-header
625 3 licm - Number of insts hoisted to multiple loop preds (bad, no loop pre-header)
626 75 mem2reg - Number of alloca's promoted
627 1444 cfgsimplify - Number of blocks simplified
631 <p>Obviously, with so many optimizations, having a unified framework for this
632 stuff is very nice. Making your pass fit well into the framework makes it more
633 maintainable and useful.</p>
637 <!-- ======================================================================= -->
638 <div class="doc_subsection">
639 <a name="ViewGraph">Viewing graphs while debugging code</a>
642 <div class="doc_text">
644 <p>Several of the important data structures in LLVM are graphs: for example
645 CFGs made out of LLVM <a href="#BasicBlock">BasicBlock</a>s, CFGs made out of
646 LLVM <a href="CodeGenerator.html#machinebasicblock">MachineBasicBlock</a>s, and
647 <a href="CodeGenerator.html#selectiondag_intro">Instruction Selection
648 DAGs</a>. In many cases, while debugging various parts of the compiler, it is
649 nice to instantly visualize these graphs.</p>
651 <p>LLVM provides several callbacks that are available in a debug build to do
652 exactly that. If you call the <tt>Function::viewCFG()</tt> method, for example,
653 the current LLVM tool will pop up a window containing the CFG for the function
654 where each basic block is a node in the graph, and each node contains the
655 instructions in the block. Similarly, there also exists
656 <tt>Function::viewCFGOnly()</tt> (does not include the instructions), the
657 <tt>MachineFunction::viewCFG()</tt> and <tt>MachineFunction::viewCFGOnly()</tt>,
658 and the <tt>SelectionDAG::viewGraph()</tt> methods. Within GDB, for example,
659 you can usually use something like <tt>call DAG.viewGraph()</tt> to pop
660 up a window. Alternatively, you can sprinkle calls to these functions in your
661 code in places you want to debug.</p>
663 <p>Getting this to work requires a small amount of configuration. On Unix
664 systems with X11, install the <a href="http://www.graphviz.org">graphviz</a>
665 toolkit, and make sure 'dot' and 'gv' are in your path. If you are running on
666 Mac OS/X, download and install the Mac OS/X <a
667 href="http://www.pixelglow.com/graphviz/">Graphviz program</a>, and add
668 <tt>/Applications/Graphviz.app/Contents/MacOS/</tt> (or wherever you install
669 it) to your path. Once in your system and path are set up, rerun the LLVM
670 configure script and rebuild LLVM to enable this functionality.</p>
672 <p><tt>SelectionDAG</tt> has been extended to make it easier to locate
673 <i>interesting</i> nodes in large complex graphs. From gdb, if you
674 <tt>call DAG.setGraphColor(<i>node</i>, "<i>color</i>")</tt>, then the
675 next <tt>call DAG.viewGraph()</tt> would highlight the node in the
676 specified color (choices of colors can be found at <a
677 href="http://www.graphviz.org/doc/info/colors.html">colors</a>.) More
678 complex node attributes can be provided with <tt>call
679 DAG.setGraphAttrs(<i>node</i>, "<i>attributes</i>")</tt> (choices can be
680 found at <a href="http://www.graphviz.org/doc/info/attrs.html">Graph
681 Attributes</a>.) If you want to restart and clear all the current graph
682 attributes, then you can <tt>call DAG.clearGraphAttrs()</tt>. </p>
686 <!-- *********************************************************************** -->
687 <div class="doc_section">
688 <a name="datastructure">Picking the Right Data Structure for a Task</a>
690 <!-- *********************************************************************** -->
692 <div class="doc_text">
694 <p>LLVM has a plethora of data structures in the <tt>llvm/ADT/</tt> directory,
695 and we commonly use STL data structures. This section describes the trade-offs
696 you should consider when you pick one.</p>
699 The first step is a choose your own adventure: do you want a sequential
700 container, a set-like container, or a map-like container? The most important
701 thing when choosing a container is the algorithmic properties of how you plan to
702 access the container. Based on that, you should use:</p>
705 <li>a <a href="#ds_map">map-like</a> container if you need efficient look-up
706 of an value based on another value. Map-like containers also support
707 efficient queries for containment (whether a key is in the map). Map-like
708 containers generally do not support efficient reverse mapping (values to
709 keys). If you need that, use two maps. Some map-like containers also
710 support efficient iteration through the keys in sorted order. Map-like
711 containers are the most expensive sort, only use them if you need one of
712 these capabilities.</li>
714 <li>a <a href="#ds_set">set-like</a> container if you need to put a bunch of
715 stuff into a container that automatically eliminates duplicates. Some
716 set-like containers support efficient iteration through the elements in
717 sorted order. Set-like containers are more expensive than sequential
721 <li>a <a href="#ds_sequential">sequential</a> container provides
722 the most efficient way to add elements and keeps track of the order they are
723 added to the collection. They permit duplicates and support efficient
724 iteration, but do not support efficient look-up based on a key.
727 <li>a <a href="#ds_bit">bit</a> container provides an efficient way to store and
728 perform set operations on sets of numeric id's, while automatically
729 eliminating duplicates. Bit containers require a maximum of 1 bit for each
730 identifier you want to store.
735 Once the proper category of container is determined, you can fine tune the
736 memory use, constant factors, and cache behaviors of access by intelligently
737 picking a member of the category. Note that constant factors and cache behavior
738 can be a big deal. If you have a vector that usually only contains a few
739 elements (but could contain many), for example, it's much better to use
740 <a href="#dss_smallvector">SmallVector</a> than <a href="#dss_vector">vector</a>
741 . Doing so avoids (relatively) expensive malloc/free calls, which dwarf the
742 cost of adding the elements to the container. </p>
746 <!-- ======================================================================= -->
747 <div class="doc_subsection">
748 <a name="ds_sequential">Sequential Containers (std::vector, std::list, etc)</a>
751 <div class="doc_text">
752 There are a variety of sequential containers available for you, based on your
753 needs. Pick the first in this section that will do what you want.
756 <!-- _______________________________________________________________________ -->
757 <div class="doc_subsubsection">
758 <a name="dss_fixedarrays">Fixed Size Arrays</a>
761 <div class="doc_text">
762 <p>Fixed size arrays are very simple and very fast. They are good if you know
763 exactly how many elements you have, or you have a (low) upper bound on how many
767 <!-- _______________________________________________________________________ -->
768 <div class="doc_subsubsection">
769 <a name="dss_heaparrays">Heap Allocated Arrays</a>
772 <div class="doc_text">
773 <p>Heap allocated arrays (new[] + delete[]) are also simple. They are good if
774 the number of elements is variable, if you know how many elements you will need
775 before the array is allocated, and if the array is usually large (if not,
776 consider a <a href="#dss_smallvector">SmallVector</a>). The cost of a heap
777 allocated array is the cost of the new/delete (aka malloc/free). Also note that
778 if you are allocating an array of a type with a constructor, the constructor and
779 destructors will be run for every element in the array (re-sizable vectors only
780 construct those elements actually used).</p>
783 <!-- _______________________________________________________________________ -->
784 <div class="doc_subsubsection">
785 <a name="dss_smallvector">"llvm/ADT/SmallVector.h"</a>
788 <div class="doc_text">
789 <p><tt>SmallVector<Type, N></tt> is a simple class that looks and smells
790 just like <tt>vector<Type></tt>:
791 it supports efficient iteration, lays out elements in memory order (so you can
792 do pointer arithmetic between elements), supports efficient push_back/pop_back
793 operations, supports efficient random access to its elements, etc.</p>
795 <p>The advantage of SmallVector is that it allocates space for
796 some number of elements (N) <b>in the object itself</b>. Because of this, if
797 the SmallVector is dynamically smaller than N, no malloc is performed. This can
798 be a big win in cases where the malloc/free call is far more expensive than the
799 code that fiddles around with the elements.</p>
801 <p>This is good for vectors that are "usually small" (e.g. the number of
802 predecessors/successors of a block is usually less than 8). On the other hand,
803 this makes the size of the SmallVector itself large, so you don't want to
804 allocate lots of them (doing so will waste a lot of space). As such,
805 SmallVectors are most useful when on the stack.</p>
807 <p>SmallVector also provides a nice portable and efficient replacement for
812 <!-- _______________________________________________________________________ -->
813 <div class="doc_subsubsection">
814 <a name="dss_vector"><vector></a>
817 <div class="doc_text">
819 std::vector is well loved and respected. It is useful when SmallVector isn't:
820 when the size of the vector is often large (thus the small optimization will
821 rarely be a benefit) or if you will be allocating many instances of the vector
822 itself (which would waste space for elements that aren't in the container).
823 vector is also useful when interfacing with code that expects vectors :).
826 <p>One worthwhile note about std::vector: avoid code like this:</p>
828 <div class="doc_code">
831 std::vector<foo> V;
837 <p>Instead, write this as:</p>
839 <div class="doc_code">
841 std::vector<foo> V;
849 <p>Doing so will save (at least) one heap allocation and free per iteration of
854 <!-- _______________________________________________________________________ -->
855 <div class="doc_subsubsection">
856 <a name="dss_deque"><deque></a>
859 <div class="doc_text">
860 <p>std::deque is, in some senses, a generalized version of std::vector. Like
861 std::vector, it provides constant time random access and other similar
862 properties, but it also provides efficient access to the front of the list. It
863 does not guarantee continuity of elements within memory.</p>
865 <p>In exchange for this extra flexibility, std::deque has significantly higher
866 constant factor costs than std::vector. If possible, use std::vector or
867 something cheaper.</p>
870 <!-- _______________________________________________________________________ -->
871 <div class="doc_subsubsection">
872 <a name="dss_list"><list></a>
875 <div class="doc_text">
876 <p>std::list is an extremely inefficient class that is rarely useful.
877 It performs a heap allocation for every element inserted into it, thus having an
878 extremely high constant factor, particularly for small data types. std::list
879 also only supports bidirectional iteration, not random access iteration.</p>
881 <p>In exchange for this high cost, std::list supports efficient access to both
882 ends of the list (like std::deque, but unlike std::vector or SmallVector). In
883 addition, the iterator invalidation characteristics of std::list are stronger
884 than that of a vector class: inserting or removing an element into the list does
885 not invalidate iterator or pointers to other elements in the list.</p>
888 <!-- _______________________________________________________________________ -->
889 <div class="doc_subsubsection">
890 <a name="dss_ilist">llvm/ADT/ilist.h</a>
893 <div class="doc_text">
894 <p><tt>ilist<T></tt> implements an 'intrusive' doubly-linked list. It is
895 intrusive, because it requires the element to store and provide access to the
896 prev/next pointers for the list.</p>
898 <p><tt>ilist</tt> has the same drawbacks as <tt>std::list</tt>, and additionally
899 requires an <tt>ilist_traits</tt> implementation for the element type, but it
900 provides some novel characteristics. In particular, it can efficiently store
901 polymorphic objects, the traits class is informed when an element is inserted or
902 removed from the list, and <tt>ilist</tt>s are guaranteed to support a
903 constant-time splice operation.</p>
905 <p>These properties are exactly what we want for things like
906 <tt>Instruction</tt>s and basic blocks, which is why these are implemented with
909 Related classes of interest are explained in the following subsections:
911 <li><a href="#dss_ilist_traits">ilist_traits</a></li>
912 <li><a href="#dss_iplist">iplist</a></li>
913 <li><a href="#dss_ilist_node">llvm/ADT/ilist_node.h</a></li>
914 <li><a href="#dss_ilist_sentinel">Sentinels</a></li>
918 <!-- _______________________________________________________________________ -->
919 <div class="doc_subsubsection">
920 <a name="dss_ilist_traits">ilist_traits</a>
923 <div class="doc_text">
924 <p><tt>ilist_traits<T></tt> is <tt>ilist<T></tt>'s customization
925 mechanism. <tt>iplist<T></tt> (and consequently <tt>ilist<T></tt>)
926 publicly derive from this traits class.</p>
929 <!-- _______________________________________________________________________ -->
930 <div class="doc_subsubsection">
931 <a name="dss_iplist">iplist</a>
934 <div class="doc_text">
935 <p><tt>iplist<T></tt> is <tt>ilist<T></tt>'s base and as such
936 supports a slightly narrower interface. Notably, inserters from
937 <tt>T&</tt> are absent.</p>
939 <p><tt>ilist_traits<T></tt> is a public base of this class and can be
940 used for a wide variety of customizations.</p>
943 <!-- _______________________________________________________________________ -->
944 <div class="doc_subsubsection">
945 <a name="dss_ilist_node">llvm/ADT/ilist_node.h</a>
948 <div class="doc_text">
949 <p><tt>ilist_node<T></tt> implements a the forward and backward links
950 that are expected by the <tt>ilist<T></tt> (and analogous containers)
951 in the default manner.</p>
953 <p><tt>ilist_node<T></tt>s are meant to be embedded in the node type
954 <tt>T</tt>, usually <tt>T</tt> publicly derives from
955 <tt>ilist_node<T></tt>.</p>
958 <!-- _______________________________________________________________________ -->
959 <div class="doc_subsubsection">
960 <a name="dss_ilist_sentinel">Sentinels</a>
963 <div class="doc_text">
964 <p><tt>ilist</tt>s have another speciality that must be considered. To be a good
965 citizen in the C++ ecosystem, it needs to support the standard container
966 operations, such as <tt>begin</tt> and <tt>end</tt> iterators, etc. Also, the
967 <tt>operator--</tt> must work correctly on the <tt>end</tt> iterator in the
968 case of non-empty <tt>ilist</tt>s.</p>
970 <p>The only sensible solution to this problem is to allocate a so-called
971 <i>sentinel</i> along with the intrusive list, which serves as the <tt>end</tt>
972 iterator, providing the back-link to the last element. However conforming to the
973 C++ convention it is illegal to <tt>operator++</tt> beyond the sentinel and it
974 also must not be dereferenced.</p>
976 <p>These constraints allow for some implementation freedom to the <tt>ilist</tt>
977 how to allocate and store the sentinel. The corresponding policy is dictated
978 by <tt>ilist_traits<T></tt>. By default a <tt>T</tt> gets heap-allocated
979 whenever the need for a sentinel arises.</p>
981 <p>While the default policy is sufficient in most cases, it may break down when
982 <tt>T</tt> does not provide a default constructor. Also, in the case of many
983 instances of <tt>ilist</tt>s, the memory overhead of the associated sentinels
984 is wasted. To alleviate the situation with numerous and voluminous
985 <tt>T</tt>-sentinels, sometimes a trick is employed, leading to <i>ghostly
988 <p>Ghostly sentinels are obtained by specially-crafted <tt>ilist_traits<T></tt>
989 which superpose the sentinel with the <tt>ilist</tt> instance in memory. Pointer
990 arithmetic is used to obtain the sentinel, which is relative to the
991 <tt>ilist</tt>'s <tt>this</tt> pointer. The <tt>ilist</tt> is augmented by an
992 extra pointer, which serves as the back-link of the sentinel. This is the only
993 field in the ghostly sentinel which can be legally accessed.</p>
996 <!-- _______________________________________________________________________ -->
997 <div class="doc_subsubsection">
998 <a name="dss_other">Other Sequential Container options</a>
1001 <div class="doc_text">
1002 <p>Other STL containers are available, such as std::string.</p>
1004 <p>There are also various STL adapter classes such as std::queue,
1005 std::priority_queue, std::stack, etc. These provide simplified access to an
1006 underlying container but don't affect the cost of the container itself.</p>
1011 <!-- ======================================================================= -->
1012 <div class="doc_subsection">
1013 <a name="ds_set">Set-Like Containers (std::set, SmallSet, SetVector, etc)</a>
1016 <div class="doc_text">
1018 <p>Set-like containers are useful when you need to canonicalize multiple values
1019 into a single representation. There are several different choices for how to do
1020 this, providing various trade-offs.</p>
1025 <!-- _______________________________________________________________________ -->
1026 <div class="doc_subsubsection">
1027 <a name="dss_sortedvectorset">A sorted 'vector'</a>
1030 <div class="doc_text">
1032 <p>If you intend to insert a lot of elements, then do a lot of queries, a
1033 great approach is to use a vector (or other sequential container) with
1034 std::sort+std::unique to remove duplicates. This approach works really well if
1035 your usage pattern has these two distinct phases (insert then query), and can be
1036 coupled with a good choice of <a href="#ds_sequential">sequential container</a>.
1040 This combination provides the several nice properties: the result data is
1041 contiguous in memory (good for cache locality), has few allocations, is easy to
1042 address (iterators in the final vector are just indices or pointers), and can be
1043 efficiently queried with a standard binary or radix search.</p>
1047 <!-- _______________________________________________________________________ -->
1048 <div class="doc_subsubsection">
1049 <a name="dss_smallset">"llvm/ADT/SmallSet.h"</a>
1052 <div class="doc_text">
1054 <p>If you have a set-like data structure that is usually small and whose elements
1055 are reasonably small, a <tt>SmallSet<Type, N></tt> is a good choice. This set
1056 has space for N elements in place (thus, if the set is dynamically smaller than
1057 N, no malloc traffic is required) and accesses them with a simple linear search.
1058 When the set grows beyond 'N' elements, it allocates a more expensive representation that
1059 guarantees efficient access (for most types, it falls back to std::set, but for
1060 pointers it uses something far better, <a
1061 href="#dss_smallptrset">SmallPtrSet</a>).</p>
1063 <p>The magic of this class is that it handles small sets extremely efficiently,
1064 but gracefully handles extremely large sets without loss of efficiency. The
1065 drawback is that the interface is quite small: it supports insertion, queries
1066 and erasing, but does not support iteration.</p>
1070 <!-- _______________________________________________________________________ -->
1071 <div class="doc_subsubsection">
1072 <a name="dss_smallptrset">"llvm/ADT/SmallPtrSet.h"</a>
1075 <div class="doc_text">
1077 <p>SmallPtrSet has all the advantages of SmallSet (and a SmallSet of pointers is
1078 transparently implemented with a SmallPtrSet), but also supports iterators. If
1079 more than 'N' insertions are performed, a single quadratically
1080 probed hash table is allocated and grows as needed, providing extremely
1081 efficient access (constant time insertion/deleting/queries with low constant
1082 factors) and is very stingy with malloc traffic.</p>
1084 <p>Note that, unlike std::set, the iterators of SmallPtrSet are invalidated
1085 whenever an insertion occurs. Also, the values visited by the iterators are not
1086 visited in sorted order.</p>
1090 <!-- _______________________________________________________________________ -->
1091 <div class="doc_subsubsection">
1092 <a name="dss_denseset">"llvm/ADT/DenseSet.h"</a>
1095 <div class="doc_text">
1098 DenseSet is a simple quadratically probed hash table. It excels at supporting
1099 small values: it uses a single allocation to hold all of the pairs that
1100 are currently inserted in the set. DenseSet is a great way to unique small
1101 values that are not simple pointers (use <a
1102 href="#dss_smallptrset">SmallPtrSet</a> for pointers). Note that DenseSet has
1103 the same requirements for the value type that <a
1104 href="#dss_densemap">DenseMap</a> has.
1109 <!-- _______________________________________________________________________ -->
1110 <div class="doc_subsubsection">
1111 <a name="dss_FoldingSet">"llvm/ADT/FoldingSet.h"</a>
1114 <div class="doc_text">
1117 FoldingSet is an aggregate class that is really good at uniquing
1118 expensive-to-create or polymorphic objects. It is a combination of a chained
1119 hash table with intrusive links (uniqued objects are required to inherit from
1120 FoldingSetNode) that uses <a href="#dss_smallvector">SmallVector</a> as part of
1123 <p>Consider a case where you want to implement a "getOrCreateFoo" method for
1124 a complex object (for example, a node in the code generator). The client has a
1125 description of *what* it wants to generate (it knows the opcode and all the
1126 operands), but we don't want to 'new' a node, then try inserting it into a set
1127 only to find out it already exists, at which point we would have to delete it
1128 and return the node that already exists.
1131 <p>To support this style of client, FoldingSet perform a query with a
1132 FoldingSetNodeID (which wraps SmallVector) that can be used to describe the
1133 element that we want to query for. The query either returns the element
1134 matching the ID or it returns an opaque ID that indicates where insertion should
1135 take place. Construction of the ID usually does not require heap traffic.</p>
1137 <p>Because FoldingSet uses intrusive links, it can support polymorphic objects
1138 in the set (for example, you can have SDNode instances mixed with LoadSDNodes).
1139 Because the elements are individually allocated, pointers to the elements are
1140 stable: inserting or removing elements does not invalidate any pointers to other
1146 <!-- _______________________________________________________________________ -->
1147 <div class="doc_subsubsection">
1148 <a name="dss_set"><set></a>
1151 <div class="doc_text">
1153 <p><tt>std::set</tt> is a reasonable all-around set class, which is decent at
1154 many things but great at nothing. std::set allocates memory for each element
1155 inserted (thus it is very malloc intensive) and typically stores three pointers
1156 per element in the set (thus adding a large amount of per-element space
1157 overhead). It offers guaranteed log(n) performance, which is not particularly
1158 fast from a complexity standpoint (particularly if the elements of the set are
1159 expensive to compare, like strings), and has extremely high constant factors for
1160 lookup, insertion and removal.</p>
1162 <p>The advantages of std::set are that its iterators are stable (deleting or
1163 inserting an element from the set does not affect iterators or pointers to other
1164 elements) and that iteration over the set is guaranteed to be in sorted order.
1165 If the elements in the set are large, then the relative overhead of the pointers
1166 and malloc traffic is not a big deal, but if the elements of the set are small,
1167 std::set is almost never a good choice.</p>
1171 <!-- _______________________________________________________________________ -->
1172 <div class="doc_subsubsection">
1173 <a name="dss_setvector">"llvm/ADT/SetVector.h"</a>
1176 <div class="doc_text">
1177 <p>LLVM's SetVector<Type> is an adapter class that combines your choice of
1178 a set-like container along with a <a href="#ds_sequential">Sequential
1179 Container</a>. The important property
1180 that this provides is efficient insertion with uniquing (duplicate elements are
1181 ignored) with iteration support. It implements this by inserting elements into
1182 both a set-like container and the sequential container, using the set-like
1183 container for uniquing and the sequential container for iteration.
1186 <p>The difference between SetVector and other sets is that the order of
1187 iteration is guaranteed to match the order of insertion into the SetVector.
1188 This property is really important for things like sets of pointers. Because
1189 pointer values are non-deterministic (e.g. vary across runs of the program on
1190 different machines), iterating over the pointers in the set will
1191 not be in a well-defined order.</p>
1194 The drawback of SetVector is that it requires twice as much space as a normal
1195 set and has the sum of constant factors from the set-like container and the
1196 sequential container that it uses. Use it *only* if you need to iterate over
1197 the elements in a deterministic order. SetVector is also expensive to delete
1198 elements out of (linear time), unless you use it's "pop_back" method, which is
1202 <p>SetVector is an adapter class that defaults to using std::vector and std::set
1203 for the underlying containers, so it is quite expensive. However,
1204 <tt>"llvm/ADT/SetVector.h"</tt> also provides a SmallSetVector class, which
1205 defaults to using a SmallVector and SmallSet of a specified size. If you use
1206 this, and if your sets are dynamically smaller than N, you will save a lot of
1211 <!-- _______________________________________________________________________ -->
1212 <div class="doc_subsubsection">
1213 <a name="dss_uniquevector">"llvm/ADT/UniqueVector.h"</a>
1216 <div class="doc_text">
1219 UniqueVector is similar to <a href="#dss_setvector">SetVector</a>, but it
1220 retains a unique ID for each element inserted into the set. It internally
1221 contains a map and a vector, and it assigns a unique ID for each value inserted
1224 <p>UniqueVector is very expensive: its cost is the sum of the cost of
1225 maintaining both the map and vector, it has high complexity, high constant
1226 factors, and produces a lot of malloc traffic. It should be avoided.</p>
1231 <!-- _______________________________________________________________________ -->
1232 <div class="doc_subsubsection">
1233 <a name="dss_otherset">Other Set-Like Container Options</a>
1236 <div class="doc_text">
1239 The STL provides several other options, such as std::multiset and the various
1240 "hash_set" like containers (whether from C++ TR1 or from the SGI library). We
1241 never use hash_set and unordered_set because they are generally very expensive
1242 (each insertion requires a malloc) and very non-portable.
1245 <p>std::multiset is useful if you're not interested in elimination of
1246 duplicates, but has all the drawbacks of std::set. A sorted vector (where you
1247 don't delete duplicate entries) or some other approach is almost always
1252 <!-- ======================================================================= -->
1253 <div class="doc_subsection">
1254 <a name="ds_map">Map-Like Containers (std::map, DenseMap, etc)</a>
1257 <div class="doc_text">
1258 Map-like containers are useful when you want to associate data to a key. As
1259 usual, there are a lot of different ways to do this. :)
1262 <!-- _______________________________________________________________________ -->
1263 <div class="doc_subsubsection">
1264 <a name="dss_sortedvectormap">A sorted 'vector'</a>
1267 <div class="doc_text">
1270 If your usage pattern follows a strict insert-then-query approach, you can
1271 trivially use the same approach as <a href="#dss_sortedvectorset">sorted vectors
1272 for set-like containers</a>. The only difference is that your query function
1273 (which uses std::lower_bound to get efficient log(n) lookup) should only compare
1274 the key, not both the key and value. This yields the same advantages as sorted
1279 <!-- _______________________________________________________________________ -->
1280 <div class="doc_subsubsection">
1281 <a name="dss_stringmap">"llvm/ADT/StringMap.h"</a>
1284 <div class="doc_text">
1287 Strings are commonly used as keys in maps, and they are difficult to support
1288 efficiently: they are variable length, inefficient to hash and compare when
1289 long, expensive to copy, etc. StringMap is a specialized container designed to
1290 cope with these issues. It supports mapping an arbitrary range of bytes to an
1291 arbitrary other object.</p>
1293 <p>The StringMap implementation uses a quadratically-probed hash table, where
1294 the buckets store a pointer to the heap allocated entries (and some other
1295 stuff). The entries in the map must be heap allocated because the strings are
1296 variable length. The string data (key) and the element object (value) are
1297 stored in the same allocation with the string data immediately after the element
1298 object. This container guarantees the "<tt>(char*)(&Value+1)</tt>" points
1299 to the key string for a value.</p>
1301 <p>The StringMap is very fast for several reasons: quadratic probing is very
1302 cache efficient for lookups, the hash value of strings in buckets is not
1303 recomputed when lookup up an element, StringMap rarely has to touch the
1304 memory for unrelated objects when looking up a value (even when hash collisions
1305 happen), hash table growth does not recompute the hash values for strings
1306 already in the table, and each pair in the map is store in a single allocation
1307 (the string data is stored in the same allocation as the Value of a pair).</p>
1309 <p>StringMap also provides query methods that take byte ranges, so it only ever
1310 copies a string if a value is inserted into the table.</p>
1313 <!-- _______________________________________________________________________ -->
1314 <div class="doc_subsubsection">
1315 <a name="dss_indexedmap">"llvm/ADT/IndexedMap.h"</a>
1318 <div class="doc_text">
1320 IndexedMap is a specialized container for mapping small dense integers (or
1321 values that can be mapped to small dense integers) to some other type. It is
1322 internally implemented as a vector with a mapping function that maps the keys to
1323 the dense integer range.
1327 This is useful for cases like virtual registers in the LLVM code generator: they
1328 have a dense mapping that is offset by a compile-time constant (the first
1329 virtual register ID).</p>
1333 <!-- _______________________________________________________________________ -->
1334 <div class="doc_subsubsection">
1335 <a name="dss_densemap">"llvm/ADT/DenseMap.h"</a>
1338 <div class="doc_text">
1341 DenseMap is a simple quadratically probed hash table. It excels at supporting
1342 small keys and values: it uses a single allocation to hold all of the pairs that
1343 are currently inserted in the map. DenseMap is a great way to map pointers to
1344 pointers, or map other small types to each other.
1348 There are several aspects of DenseMap that you should be aware of, however. The
1349 iterators in a densemap are invalidated whenever an insertion occurs, unlike
1350 map. Also, because DenseMap allocates space for a large number of key/value
1351 pairs (it starts with 64 by default), it will waste a lot of space if your keys
1352 or values are large. Finally, you must implement a partial specialization of
1353 DenseMapInfo for the key that you want, if it isn't already supported. This
1354 is required to tell DenseMap about two special marker values (which can never be
1355 inserted into the map) that it needs internally.</p>
1359 <!-- _______________________________________________________________________ -->
1360 <div class="doc_subsubsection">
1361 <a name="dss_map"><map></a>
1364 <div class="doc_text">
1367 std::map has similar characteristics to <a href="#dss_set">std::set</a>: it uses
1368 a single allocation per pair inserted into the map, it offers log(n) lookup with
1369 an extremely large constant factor, imposes a space penalty of 3 pointers per
1370 pair in the map, etc.</p>
1372 <p>std::map is most useful when your keys or values are very large, if you need
1373 to iterate over the collection in sorted order, or if you need stable iterators
1374 into the map (i.e. they don't get invalidated if an insertion or deletion of
1375 another element takes place).</p>
1379 <!-- _______________________________________________________________________ -->
1380 <div class="doc_subsubsection">
1381 <a name="dss_othermap">Other Map-Like Container Options</a>
1384 <div class="doc_text">
1387 The STL provides several other options, such as std::multimap and the various
1388 "hash_map" like containers (whether from C++ TR1 or from the SGI library). We
1389 never use hash_set and unordered_set because they are generally very expensive
1390 (each insertion requires a malloc) and very non-portable.</p>
1392 <p>std::multimap is useful if you want to map a key to multiple values, but has
1393 all the drawbacks of std::map. A sorted vector or some other approach is almost
1398 <!-- ======================================================================= -->
1399 <div class="doc_subsection">
1400 <a name="ds_bit">Bit storage containers (BitVector, SparseBitVector)</a>
1403 <div class="doc_text">
1404 <p>Unlike the other containers, there are only two bit storage containers, and
1405 choosing when to use each is relatively straightforward.</p>
1407 <p>One additional option is
1408 <tt>std::vector<bool></tt>: we discourage its use for two reasons 1) the
1409 implementation in many common compilers (e.g. commonly available versions of
1410 GCC) is extremely inefficient and 2) the C++ standards committee is likely to
1411 deprecate this container and/or change it significantly somehow. In any case,
1412 please don't use it.</p>
1415 <!-- _______________________________________________________________________ -->
1416 <div class="doc_subsubsection">
1417 <a name="dss_bitvector">BitVector</a>
1420 <div class="doc_text">
1421 <p> The BitVector container provides a fixed size set of bits for manipulation.
1422 It supports individual bit setting/testing, as well as set operations. The set
1423 operations take time O(size of bitvector), but operations are performed one word
1424 at a time, instead of one bit at a time. This makes the BitVector very fast for
1425 set operations compared to other containers. Use the BitVector when you expect
1426 the number of set bits to be high (IE a dense set).
1430 <!-- _______________________________________________________________________ -->
1431 <div class="doc_subsubsection">
1432 <a name="dss_sparsebitvector">SparseBitVector</a>
1435 <div class="doc_text">
1436 <p> The SparseBitVector container is much like BitVector, with one major
1437 difference: Only the bits that are set, are stored. This makes the
1438 SparseBitVector much more space efficient than BitVector when the set is sparse,
1439 as well as making set operations O(number of set bits) instead of O(size of
1440 universe). The downside to the SparseBitVector is that setting and testing of random bits is O(N), and on large SparseBitVectors, this can be slower than BitVector. In our implementation, setting or testing bits in sorted order
1441 (either forwards or reverse) is O(1) worst case. Testing and setting bits within 128 bits (depends on size) of the current bit is also O(1). As a general statement, testing/setting bits in a SparseBitVector is O(distance away from last set bit).
1445 <!-- *********************************************************************** -->
1446 <div class="doc_section">
1447 <a name="common">Helpful Hints for Common Operations</a>
1449 <!-- *********************************************************************** -->
1451 <div class="doc_text">
1453 <p>This section describes how to perform some very simple transformations of
1454 LLVM code. This is meant to give examples of common idioms used, showing the
1455 practical side of LLVM transformations. <p> Because this is a "how-to" section,
1456 you should also read about the main classes that you will be working with. The
1457 <a href="#coreclasses">Core LLVM Class Hierarchy Reference</a> contains details
1458 and descriptions of the main classes that you should know about.</p>
1462 <!-- NOTE: this section should be heavy on example code -->
1463 <!-- ======================================================================= -->
1464 <div class="doc_subsection">
1465 <a name="inspection">Basic Inspection and Traversal Routines</a>
1468 <div class="doc_text">
1470 <p>The LLVM compiler infrastructure have many different data structures that may
1471 be traversed. Following the example of the C++ standard template library, the
1472 techniques used to traverse these various data structures are all basically the
1473 same. For a enumerable sequence of values, the <tt>XXXbegin()</tt> function (or
1474 method) returns an iterator to the start of the sequence, the <tt>XXXend()</tt>
1475 function returns an iterator pointing to one past the last valid element of the
1476 sequence, and there is some <tt>XXXiterator</tt> data type that is common
1477 between the two operations.</p>
1479 <p>Because the pattern for iteration is common across many different aspects of
1480 the program representation, the standard template library algorithms may be used
1481 on them, and it is easier to remember how to iterate. First we show a few common
1482 examples of the data structures that need to be traversed. Other data
1483 structures are traversed in very similar ways.</p>
1487 <!-- _______________________________________________________________________ -->
1488 <div class="doc_subsubsection">
1489 <a name="iterate_function">Iterating over the </a><a
1490 href="#BasicBlock"><tt>BasicBlock</tt></a>s in a <a
1491 href="#Function"><tt>Function</tt></a>
1494 <div class="doc_text">
1496 <p>It's quite common to have a <tt>Function</tt> instance that you'd like to
1497 transform in some way; in particular, you'd like to manipulate its
1498 <tt>BasicBlock</tt>s. To facilitate this, you'll need to iterate over all of
1499 the <tt>BasicBlock</tt>s that constitute the <tt>Function</tt>. The following is
1500 an example that prints the name of a <tt>BasicBlock</tt> and the number of
1501 <tt>Instruction</tt>s it contains:</p>
1503 <div class="doc_code">
1505 // <i>func is a pointer to a Function instance</i>
1506 for (Function::iterator i = func->begin(), e = func->end(); i != e; ++i)
1507 // <i>Print out the name of the basic block if it has one, and then the</i>
1508 // <i>number of instructions that it contains</i>
1509 llvm::cerr << "Basic block (name=" << i->getName() << ") has "
1510 << i->size() << " instructions.\n";
1514 <p>Note that i can be used as if it were a pointer for the purposes of
1515 invoking member functions of the <tt>Instruction</tt> class. This is
1516 because the indirection operator is overloaded for the iterator
1517 classes. In the above code, the expression <tt>i->size()</tt> is
1518 exactly equivalent to <tt>(*i).size()</tt> just like you'd expect.</p>
1522 <!-- _______________________________________________________________________ -->
1523 <div class="doc_subsubsection">
1524 <a name="iterate_basicblock">Iterating over the </a><a
1525 href="#Instruction"><tt>Instruction</tt></a>s in a <a
1526 href="#BasicBlock"><tt>BasicBlock</tt></a>
1529 <div class="doc_text">
1531 <p>Just like when dealing with <tt>BasicBlock</tt>s in <tt>Function</tt>s, it's
1532 easy to iterate over the individual instructions that make up
1533 <tt>BasicBlock</tt>s. Here's a code snippet that prints out each instruction in
1534 a <tt>BasicBlock</tt>:</p>
1536 <div class="doc_code">
1538 // <i>blk is a pointer to a BasicBlock instance</i>
1539 for (BasicBlock::iterator i = blk->begin(), e = blk->end(); i != e; ++i)
1540 // <i>The next statement works since operator<<(ostream&,...)</i>
1541 // <i>is overloaded for Instruction&</i>
1542 llvm::cerr << *i << "\n";
1546 <p>However, this isn't really the best way to print out the contents of a
1547 <tt>BasicBlock</tt>! Since the ostream operators are overloaded for virtually
1548 anything you'll care about, you could have just invoked the print routine on the
1549 basic block itself: <tt>llvm::cerr << *blk << "\n";</tt>.</p>
1553 <!-- _______________________________________________________________________ -->
1554 <div class="doc_subsubsection">
1555 <a name="iterate_institer">Iterating over the </a><a
1556 href="#Instruction"><tt>Instruction</tt></a>s in a <a
1557 href="#Function"><tt>Function</tt></a>
1560 <div class="doc_text">
1562 <p>If you're finding that you commonly iterate over a <tt>Function</tt>'s
1563 <tt>BasicBlock</tt>s and then that <tt>BasicBlock</tt>'s <tt>Instruction</tt>s,
1564 <tt>InstIterator</tt> should be used instead. You'll need to include <a
1565 href="/doxygen/InstIterator_8h-source.html"><tt>llvm/Support/InstIterator.h</tt></a>,
1566 and then instantiate <tt>InstIterator</tt>s explicitly in your code. Here's a
1567 small example that shows how to dump all instructions in a function to the standard error stream:<p>
1569 <div class="doc_code">
1571 #include "<a href="/doxygen/InstIterator_8h-source.html">llvm/Support/InstIterator.h</a>"
1573 // <i>F is a pointer to a Function instance</i>
1574 for (inst_iterator I = inst_begin(F), E = inst_end(F); I != E; ++I)
1575 llvm::cerr << *I << "\n";
1579 <p>Easy, isn't it? You can also use <tt>InstIterator</tt>s to fill a
1580 work list with its initial contents. For example, if you wanted to
1581 initialize a work list to contain all instructions in a <tt>Function</tt>
1582 F, all you would need to do is something like:</p>
1584 <div class="doc_code">
1586 std::set<Instruction*> worklist;
1587 // or better yet, SmallPtrSet<Instruction*, 64> worklist;
1589 for (inst_iterator I = inst_begin(F), E = inst_end(F); I != E; ++I)
1590 worklist.insert(&*I);
1594 <p>The STL set <tt>worklist</tt> would now contain all instructions in the
1595 <tt>Function</tt> pointed to by F.</p>
1599 <!-- _______________________________________________________________________ -->
1600 <div class="doc_subsubsection">
1601 <a name="iterate_convert">Turning an iterator into a class pointer (and
1605 <div class="doc_text">
1607 <p>Sometimes, it'll be useful to grab a reference (or pointer) to a class
1608 instance when all you've got at hand is an iterator. Well, extracting
1609 a reference or a pointer from an iterator is very straight-forward.
1610 Assuming that <tt>i</tt> is a <tt>BasicBlock::iterator</tt> and <tt>j</tt>
1611 is a <tt>BasicBlock::const_iterator</tt>:</p>
1613 <div class="doc_code">
1615 Instruction& inst = *i; // <i>Grab reference to instruction reference</i>
1616 Instruction* pinst = &*i; // <i>Grab pointer to instruction reference</i>
1617 const Instruction& inst = *j;
1621 <p>However, the iterators you'll be working with in the LLVM framework are
1622 special: they will automatically convert to a ptr-to-instance type whenever they
1623 need to. Instead of dereferencing the iterator and then taking the address of
1624 the result, you can simply assign the iterator to the proper pointer type and
1625 you get the dereference and address-of operation as a result of the assignment
1626 (behind the scenes, this is a result of overloading casting mechanisms). Thus
1627 the last line of the last example,</p>
1629 <div class="doc_code">
1631 Instruction *pinst = &*i;
1635 <p>is semantically equivalent to</p>
1637 <div class="doc_code">
1639 Instruction *pinst = i;
1643 <p>It's also possible to turn a class pointer into the corresponding iterator,
1644 and this is a constant time operation (very efficient). The following code
1645 snippet illustrates use of the conversion constructors provided by LLVM
1646 iterators. By using these, you can explicitly grab the iterator of something
1647 without actually obtaining it via iteration over some structure:</p>
1649 <div class="doc_code">
1651 void printNextInstruction(Instruction* inst) {
1652 BasicBlock::iterator it(inst);
1653 ++it; // <i>After this line, it refers to the instruction after *inst</i>
1654 if (it != inst->getParent()->end()) llvm::cerr << *it << "\n";
1661 <!--_______________________________________________________________________-->
1662 <div class="doc_subsubsection">
1663 <a name="iterate_complex">Finding call sites: a slightly more complex
1667 <div class="doc_text">
1669 <p>Say that you're writing a FunctionPass and would like to count all the
1670 locations in the entire module (that is, across every <tt>Function</tt>) where a
1671 certain function (i.e., some <tt>Function</tt>*) is already in scope. As you'll
1672 learn later, you may want to use an <tt>InstVisitor</tt> to accomplish this in a
1673 much more straight-forward manner, but this example will allow us to explore how
1674 you'd do it if you didn't have <tt>InstVisitor</tt> around. In pseudo-code, this
1675 is what we want to do:</p>
1677 <div class="doc_code">
1679 initialize callCounter to zero
1680 for each Function f in the Module
1681 for each BasicBlock b in f
1682 for each Instruction i in b
1683 if (i is a CallInst and calls the given function)
1684 increment callCounter
1688 <p>And the actual code is (remember, because we're writing a
1689 <tt>FunctionPass</tt>, our <tt>FunctionPass</tt>-derived class simply has to
1690 override the <tt>runOnFunction</tt> method):</p>
1692 <div class="doc_code">
1694 Function* targetFunc = ...;
1696 class OurFunctionPass : public FunctionPass {
1698 OurFunctionPass(): callCounter(0) { }
1700 virtual runOnFunction(Function& F) {
1701 for (Function::iterator b = F.begin(), be = F.end(); b != be; ++b) {
1702 for (BasicBlock::iterator i = b->begin(), ie = b->end(); i != ie; ++i) {
1703 if (<a href="#CallInst">CallInst</a>* callInst = <a href="#isa">dyn_cast</a><<a
1704 href="#CallInst">CallInst</a>>(&*i)) {
1705 // <i>We know we've encountered a call instruction, so we</i>
1706 // <i>need to determine if it's a call to the</i>
1707 // <i>function pointed to by m_func or not.</i>
1708 if (callInst->getCalledFunction() == targetFunc)
1716 unsigned callCounter;
1723 <!--_______________________________________________________________________-->
1724 <div class="doc_subsubsection">
1725 <a name="calls_and_invokes">Treating calls and invokes the same way</a>
1728 <div class="doc_text">
1730 <p>You may have noticed that the previous example was a bit oversimplified in
1731 that it did not deal with call sites generated by 'invoke' instructions. In
1732 this, and in other situations, you may find that you want to treat
1733 <tt>CallInst</tt>s and <tt>InvokeInst</tt>s the same way, even though their
1734 most-specific common base class is <tt>Instruction</tt>, which includes lots of
1735 less closely-related things. For these cases, LLVM provides a handy wrapper
1737 href="http://llvm.org/doxygen/classllvm_1_1CallSite.html"><tt>CallSite</tt></a>.
1738 It is essentially a wrapper around an <tt>Instruction</tt> pointer, with some
1739 methods that provide functionality common to <tt>CallInst</tt>s and
1740 <tt>InvokeInst</tt>s.</p>
1742 <p>This class has "value semantics": it should be passed by value, not by
1743 reference and it should not be dynamically allocated or deallocated using
1744 <tt>operator new</tt> or <tt>operator delete</tt>. It is efficiently copyable,
1745 assignable and constructable, with costs equivalents to that of a bare pointer.
1746 If you look at its definition, it has only a single pointer member.</p>
1750 <!--_______________________________________________________________________-->
1751 <div class="doc_subsubsection">
1752 <a name="iterate_chains">Iterating over def-use & use-def chains</a>
1755 <div class="doc_text">
1757 <p>Frequently, we might have an instance of the <a
1758 href="/doxygen/classllvm_1_1Value.html">Value Class</a> and we want to
1759 determine which <tt>User</tt>s use the <tt>Value</tt>. The list of all
1760 <tt>User</tt>s of a particular <tt>Value</tt> is called a <i>def-use</i> chain.
1761 For example, let's say we have a <tt>Function*</tt> named <tt>F</tt> to a
1762 particular function <tt>foo</tt>. Finding all of the instructions that
1763 <i>use</i> <tt>foo</tt> is as simple as iterating over the <i>def-use</i> chain
1766 <div class="doc_code">
1770 for (Value::use_iterator i = F->use_begin(), e = F->use_end(); i != e; ++i)
1771 if (Instruction *Inst = dyn_cast<Instruction>(*i)) {
1772 llvm::cerr << "F is used in instruction:\n";
1773 llvm::cerr << *Inst << "\n";
1778 <p>Alternately, it's common to have an instance of the <a
1779 href="/doxygen/classllvm_1_1User.html">User Class</a> and need to know what
1780 <tt>Value</tt>s are used by it. The list of all <tt>Value</tt>s used by a
1781 <tt>User</tt> is known as a <i>use-def</i> chain. Instances of class
1782 <tt>Instruction</tt> are common <tt>User</tt>s, so we might want to iterate over
1783 all of the values that a particular instruction uses (that is, the operands of
1784 the particular <tt>Instruction</tt>):</p>
1786 <div class="doc_code">
1788 Instruction *pi = ...;
1790 for (User::op_iterator i = pi->op_begin(), e = pi->op_end(); i != e; ++i) {
1798 def-use chains ("finding all users of"): Value::use_begin/use_end
1799 use-def chains ("finding all values used"): User::op_begin/op_end [op=operand]
1804 <!--_______________________________________________________________________-->
1805 <div class="doc_subsubsection">
1806 <a name="iterate_preds">Iterating over predecessors &
1807 successors of blocks</a>
1810 <div class="doc_text">
1812 <p>Iterating over the predecessors and successors of a block is quite easy
1813 with the routines defined in <tt>"llvm/Support/CFG.h"</tt>. Just use code like
1814 this to iterate over all predecessors of BB:</p>
1816 <div class="doc_code">
1818 #include "llvm/Support/CFG.h"
1819 BasicBlock *BB = ...;
1821 for (pred_iterator PI = pred_begin(BB), E = pred_end(BB); PI != E; ++PI) {
1822 BasicBlock *Pred = *PI;
1828 <p>Similarly, to iterate over successors use
1829 succ_iterator/succ_begin/succ_end.</p>
1834 <!-- ======================================================================= -->
1835 <div class="doc_subsection">
1836 <a name="simplechanges">Making simple changes</a>
1839 <div class="doc_text">
1841 <p>There are some primitive transformation operations present in the LLVM
1842 infrastructure that are worth knowing about. When performing
1843 transformations, it's fairly common to manipulate the contents of basic
1844 blocks. This section describes some of the common methods for doing so
1845 and gives example code.</p>
1849 <!--_______________________________________________________________________-->
1850 <div class="doc_subsubsection">
1851 <a name="schanges_creating">Creating and inserting new
1852 <tt>Instruction</tt>s</a>
1855 <div class="doc_text">
1857 <p><i>Instantiating Instructions</i></p>
1859 <p>Creation of <tt>Instruction</tt>s is straight-forward: simply call the
1860 constructor for the kind of instruction to instantiate and provide the necessary
1861 parameters. For example, an <tt>AllocaInst</tt> only <i>requires</i> a
1862 (const-ptr-to) <tt>Type</tt>. Thus:</p>
1864 <div class="doc_code">
1866 AllocaInst* ai = new AllocaInst(Type::Int32Ty);
1870 <p>will create an <tt>AllocaInst</tt> instance that represents the allocation of
1871 one integer in the current stack frame, at run time. Each <tt>Instruction</tt>
1872 subclass is likely to have varying default parameters which change the semantics
1873 of the instruction, so refer to the <a
1874 href="/doxygen/classllvm_1_1Instruction.html">doxygen documentation for the subclass of
1875 Instruction</a> that you're interested in instantiating.</p>
1877 <p><i>Naming values</i></p>
1879 <p>It is very useful to name the values of instructions when you're able to, as
1880 this facilitates the debugging of your transformations. If you end up looking
1881 at generated LLVM machine code, you definitely want to have logical names
1882 associated with the results of instructions! By supplying a value for the
1883 <tt>Name</tt> (default) parameter of the <tt>Instruction</tt> constructor, you
1884 associate a logical name with the result of the instruction's execution at
1885 run time. For example, say that I'm writing a transformation that dynamically
1886 allocates space for an integer on the stack, and that integer is going to be
1887 used as some kind of index by some other code. To accomplish this, I place an
1888 <tt>AllocaInst</tt> at the first point in the first <tt>BasicBlock</tt> of some
1889 <tt>Function</tt>, and I'm intending to use it within the same
1890 <tt>Function</tt>. I might do:</p>
1892 <div class="doc_code">
1894 AllocaInst* pa = new AllocaInst(Type::Int32Ty, 0, "indexLoc");
1898 <p>where <tt>indexLoc</tt> is now the logical name of the instruction's
1899 execution value, which is a pointer to an integer on the run time stack.</p>
1901 <p><i>Inserting instructions</i></p>
1903 <p>There are essentially two ways to insert an <tt>Instruction</tt>
1904 into an existing sequence of instructions that form a <tt>BasicBlock</tt>:</p>
1907 <li>Insertion into an explicit instruction list
1909 <p>Given a <tt>BasicBlock* pb</tt>, an <tt>Instruction* pi</tt> within that
1910 <tt>BasicBlock</tt>, and a newly-created instruction we wish to insert
1911 before <tt>*pi</tt>, we do the following: </p>
1913 <div class="doc_code">
1915 BasicBlock *pb = ...;
1916 Instruction *pi = ...;
1917 Instruction *newInst = new Instruction(...);
1919 pb->getInstList().insert(pi, newInst); // <i>Inserts newInst before pi in pb</i>
1923 <p>Appending to the end of a <tt>BasicBlock</tt> is so common that
1924 the <tt>Instruction</tt> class and <tt>Instruction</tt>-derived
1925 classes provide constructors which take a pointer to a
1926 <tt>BasicBlock</tt> to be appended to. For example code that
1929 <div class="doc_code">
1931 BasicBlock *pb = ...;
1932 Instruction *newInst = new Instruction(...);
1934 pb->getInstList().push_back(newInst); // <i>Appends newInst to pb</i>
1940 <div class="doc_code">
1942 BasicBlock *pb = ...;
1943 Instruction *newInst = new Instruction(..., pb);
1947 <p>which is much cleaner, especially if you are creating
1948 long instruction streams.</p></li>
1950 <li>Insertion into an implicit instruction list
1952 <p><tt>Instruction</tt> instances that are already in <tt>BasicBlock</tt>s
1953 are implicitly associated with an existing instruction list: the instruction
1954 list of the enclosing basic block. Thus, we could have accomplished the same
1955 thing as the above code without being given a <tt>BasicBlock</tt> by doing:
1958 <div class="doc_code">
1960 Instruction *pi = ...;
1961 Instruction *newInst = new Instruction(...);
1963 pi->getParent()->getInstList().insert(pi, newInst);
1967 <p>In fact, this sequence of steps occurs so frequently that the
1968 <tt>Instruction</tt> class and <tt>Instruction</tt>-derived classes provide
1969 constructors which take (as a default parameter) a pointer to an
1970 <tt>Instruction</tt> which the newly-created <tt>Instruction</tt> should
1971 precede. That is, <tt>Instruction</tt> constructors are capable of
1972 inserting the newly-created instance into the <tt>BasicBlock</tt> of a
1973 provided instruction, immediately before that instruction. Using an
1974 <tt>Instruction</tt> constructor with a <tt>insertBefore</tt> (default)
1975 parameter, the above code becomes:</p>
1977 <div class="doc_code">
1979 Instruction* pi = ...;
1980 Instruction* newInst = new Instruction(..., pi);
1984 <p>which is much cleaner, especially if you're creating a lot of
1985 instructions and adding them to <tt>BasicBlock</tt>s.</p></li>
1990 <!--_______________________________________________________________________-->
1991 <div class="doc_subsubsection">
1992 <a name="schanges_deleting">Deleting <tt>Instruction</tt>s</a>
1995 <div class="doc_text">
1997 <p>Deleting an instruction from an existing sequence of instructions that form a
1998 <a href="#BasicBlock"><tt>BasicBlock</tt></a> is very straight-forward. First,
1999 you must have a pointer to the instruction that you wish to delete. Second, you
2000 need to obtain the pointer to that instruction's basic block. You use the
2001 pointer to the basic block to get its list of instructions and then use the
2002 erase function to remove your instruction. For example:</p>
2004 <div class="doc_code">
2006 <a href="#Instruction">Instruction</a> *I = .. ;
2007 I->eraseFromParent();
2013 <!--_______________________________________________________________________-->
2014 <div class="doc_subsubsection">
2015 <a name="schanges_replacing">Replacing an <tt>Instruction</tt> with another
2019 <div class="doc_text">
2021 <p><i>Replacing individual instructions</i></p>
2023 <p>Including "<a href="/doxygen/BasicBlockUtils_8h-source.html">llvm/Transforms/Utils/BasicBlockUtils.h</a>"
2024 permits use of two very useful replace functions: <tt>ReplaceInstWithValue</tt>
2025 and <tt>ReplaceInstWithInst</tt>.</p>
2027 <h4><a name="schanges_deleting">Deleting <tt>Instruction</tt>s</a></h4>
2030 <li><tt>ReplaceInstWithValue</tt>
2032 <p>This function replaces all uses of a given instruction with a value,
2033 and then removes the original instruction. The following example
2034 illustrates the replacement of the result of a particular
2035 <tt>AllocaInst</tt> that allocates memory for a single integer with a null
2036 pointer to an integer.</p>
2038 <div class="doc_code">
2040 AllocaInst* instToReplace = ...;
2041 BasicBlock::iterator ii(instToReplace);
2043 ReplaceInstWithValue(instToReplace->getParent()->getInstList(), ii,
2044 Constant::getNullValue(PointerType::getUnqual(Type::Int32Ty)));
2047 <li><tt>ReplaceInstWithInst</tt>
2049 <p>This function replaces a particular instruction with another
2050 instruction, inserting the new instruction into the basic block at the
2051 location where the old instruction was, and replacing any uses of the old
2052 instruction with the new instruction. The following example illustrates
2053 the replacement of one <tt>AllocaInst</tt> with another.</p>
2055 <div class="doc_code">
2057 AllocaInst* instToReplace = ...;
2058 BasicBlock::iterator ii(instToReplace);
2060 ReplaceInstWithInst(instToReplace->getParent()->getInstList(), ii,
2061 new AllocaInst(Type::Int32Ty, 0, "ptrToReplacedInt"));
2065 <p><i>Replacing multiple uses of <tt>User</tt>s and <tt>Value</tt>s</i></p>
2067 <p>You can use <tt>Value::replaceAllUsesWith</tt> and
2068 <tt>User::replaceUsesOfWith</tt> to change more than one use at a time. See the
2069 doxygen documentation for the <a href="/doxygen/classllvm_1_1Value.html">Value Class</a>
2070 and <a href="/doxygen/classllvm_1_1User.html">User Class</a>, respectively, for more
2073 <!-- Value::replaceAllUsesWith User::replaceUsesOfWith Point out:
2074 include/llvm/Transforms/Utils/ especially BasicBlockUtils.h with:
2075 ReplaceInstWithValue, ReplaceInstWithInst -->
2079 <!--_______________________________________________________________________-->
2080 <div class="doc_subsubsection">
2081 <a name="schanges_deletingGV">Deleting <tt>GlobalVariable</tt>s</a>
2084 <div class="doc_text">
2086 <p>Deleting a global variable from a module is just as easy as deleting an
2087 Instruction. First, you must have a pointer to the global variable that you wish
2088 to delete. You use this pointer to erase it from its parent, the module.
2091 <div class="doc_code">
2093 <a href="#GlobalVariable">GlobalVariable</a> *GV = .. ;
2095 GV->eraseFromParent();
2101 <!-- ======================================================================= -->
2102 <div class="doc_subsection">
2103 <a name="create_types">How to Create Types</a>
2106 <div class="doc_text">
2108 <p>In generating IR, you may need some complex types. If you know these types
2109 statically, you can use <tt>TypeBuilder<...>::get()</tt>, defined
2110 in <tt>llvm/Support/TypeBuilder.h</tt>, to retrieve them. <tt>TypeBuilder</tt>
2111 has two forms depending on whether you're building types for cross-compilation
2112 or native library use. <tt>TypeBuilder<T, true></tt> requires
2113 that <tt>T</tt> be independent of the host environment, meaning that it's built
2115 the <a href="/doxygen/namespacellvm_1_1types.html"><tt>llvm::types</tt></a>
2116 namespace and pointers, functions, arrays, etc. built of
2117 those. <tt>TypeBuilder<T, false></tt> additionally allows native C types
2118 whose size may depend on the host compiler. For example,</p>
2120 <div class="doc_code">
2122 FunctionType *ft = TypeBuilder<types::i<8>(types::i<32>*), true>::get();
2126 <p>is easier to read and write than the equivalent</p>
2128 <div class="doc_code">
2130 std::vector<const Type*> params;
2131 params.push_back(PointerType::getUnqual(Type::Int32Ty));
2132 FunctionType *ft = FunctionType::get(Type::Int8Ty, params, false);
2136 <p>See the <a href="/doxygen/TypeBuilder_8h-source.html#l00001">class
2137 comment</a> for more details.</p>
2141 <!-- *********************************************************************** -->
2142 <div class="doc_section">
2143 <a name="threading">Threads and LLVM</a>
2145 <!-- *********************************************************************** -->
2147 <div class="doc_text">
2149 This section describes the interaction of the LLVM APIs with multithreading,
2150 both on the part of client applications, and in the JIT, in the hosted
2155 Note that LLVM's support for multithreading is still relatively young. Up
2156 through version 2.5, the execution of threaded hosted applications was
2157 supported, but not threaded client access to the APIs. While this use case is
2158 now supported, clients <em>must</em> adhere to the guidelines specified below to
2159 ensure proper operation in multithreaded mode.
2163 Note that, on Unix-like platforms, LLVM requires the presence of GCC's atomic
2164 intrinsics in order to support threaded operation. If you need a
2165 multhreading-capable LLVM on a platform without a suitably modern system
2166 compiler, consider compiling LLVM and LLVM-GCC in single-threaded mode, and
2167 using the resultant compiler to build a copy of LLVM with multithreading
2172 <!-- ======================================================================= -->
2173 <div class="doc_subsection">
2174 <a name="startmultithreaded">Entering Threaded Mode with
2175 <tt>llvm_start_multithreaded()</tt></a>
2178 <div class="doc_text">
2181 In order to properly protect its internal data structures while avoiding
2182 excessive locking overhead in the single-threaded case, the LLVM APIs require
2183 that the client invoke <tt>llvm_start_multithreaded()</tt>. This call must
2184 complete <em>before</em> any other threads attempt to invoke LLVM APIs. Any
2185 attempts to call LLVM APIs from multiple threads before
2186 <tt>llvm_start_multithreaded</tt> returns can and will cause corruption of
2187 LLVM's internal data.
2191 A caveat: before <tt>llvm_start_multithreaded()</tt> has been invoked, all
2192 <tt>llvm::sys::Mutex</tt> acquisitions and releases will become no-ops. This
2193 means that <tt>llvm_start_multithreaded()</tt> must be invoked before a threaded
2194 application can be executed in the JIT.
2198 <!-- ======================================================================= -->
2199 <div class="doc_subsection">
2200 <a name="shutdown">Ending Execution with <tt>llvm_shutdown()</tt></a>
2203 <div class="doc_text">
2205 When you are done using the LLVM APIs, you should call <tt>llvm_shutdown()</tt>
2206 to deallocate memory used for internal structures. This call must not begin
2207 while any other threads are still issuing LLVM API calls. Doing so is likely
2208 to result in garbage data or crashes.
2212 Note that, if you use scope-based shutdown, you can use the
2213 <tt>llvm_shutdown_obj</tt> class, which calls <tt>llvm_shutdown()</tt> in its
2217 <!-- ======================================================================= -->
2218 <div class="doc_subsection">
2219 <a name="managedstatic">Lazy Initialization with <tt>ManagedStatic</tt></a>
2222 <div class="doc_text">
2224 <tt>ManagedStatic</tt> is a utility class in LLVM used to implement static
2225 initialization of static resources, such as the global type tables. Before the
2226 invocation of <tt>llvm_shutdown()</tt>, it implements a simple lazy
2227 initialization scheme. Once <tt>llvm_start_multithreaded()</tt> returns,
2228 however, it uses double-checked locking to implement thread-safe lazy
2233 Note that, because no other threads are allowed to issue LLVM API calls before
2234 <tt>llvm_start_multithreaded()</tt> returns, it is possible to have
2235 <tt>ManagedStatic</tt>s of <tt>llvm::sys::Mutex</tt>s.
2239 <!-- *********************************************************************** -->
2240 <div class="doc_section">
2241 <a name="advanced">Advanced Topics</a>
2243 <!-- *********************************************************************** -->
2245 <div class="doc_text">
2247 This section describes some of the advanced or obscure API's that most clients
2248 do not need to be aware of. These API's tend manage the inner workings of the
2249 LLVM system, and only need to be accessed in unusual circumstances.
2253 <!-- ======================================================================= -->
2254 <div class="doc_subsection">
2255 <a name="TypeResolve">LLVM Type Resolution</a>
2258 <div class="doc_text">
2261 The LLVM type system has a very simple goal: allow clients to compare types for
2262 structural equality with a simple pointer comparison (aka a shallow compare).
2263 This goal makes clients much simpler and faster, and is used throughout the LLVM
2268 Unfortunately achieving this goal is not a simple matter. In particular,
2269 recursive types and late resolution of opaque types makes the situation very
2270 difficult to handle. Fortunately, for the most part, our implementation makes
2271 most clients able to be completely unaware of the nasty internal details. The
2272 primary case where clients are exposed to the inner workings of it are when
2273 building a recursive type. In addition to this case, the LLVM bitcode reader,
2274 assembly parser, and linker also have to be aware of the inner workings of this
2279 For our purposes below, we need three concepts. First, an "Opaque Type" is
2280 exactly as defined in the <a href="LangRef.html#t_opaque">language
2281 reference</a>. Second an "Abstract Type" is any type which includes an
2282 opaque type as part of its type graph (for example "<tt>{ opaque, i32 }</tt>").
2283 Third, a concrete type is a type that is not an abstract type (e.g. "<tt>{ i32,
2289 <!-- ______________________________________________________________________ -->
2290 <div class="doc_subsubsection">
2291 <a name="BuildRecType">Basic Recursive Type Construction</a>
2294 <div class="doc_text">
2297 Because the most common question is "how do I build a recursive type with LLVM",
2298 we answer it now and explain it as we go. Here we include enough to cause this
2299 to be emitted to an output .ll file:
2302 <div class="doc_code">
2304 %mylist = type { %mylist*, i32 }
2309 To build this, use the following LLVM APIs:
2312 <div class="doc_code">
2314 // <i>Create the initial outer struct</i>
2315 <a href="#PATypeHolder">PATypeHolder</a> StructTy = OpaqueType::get();
2316 std::vector<const Type*> Elts;
2317 Elts.push_back(PointerType::getUnqual(StructTy));
2318 Elts.push_back(Type::Int32Ty);
2319 StructType *NewSTy = StructType::get(Elts);
2321 // <i>At this point, NewSTy = "{ opaque*, i32 }". Tell VMCore that</i>
2322 // <i>the struct and the opaque type are actually the same.</i>
2323 cast<OpaqueType>(StructTy.get())-><a href="#refineAbstractTypeTo">refineAbstractTypeTo</a>(NewSTy);
2325 // <i>NewSTy is potentially invalidated, but StructTy (a <a href="#PATypeHolder">PATypeHolder</a>) is</i>
2326 // <i>kept up-to-date</i>
2327 NewSTy = cast<StructType>(StructTy.get());
2329 // <i>Add a name for the type to the module symbol table (optional)</i>
2330 MyModule->addTypeName("mylist", NewSTy);
2335 This code shows the basic approach used to build recursive types: build a
2336 non-recursive type using 'opaque', then use type unification to close the cycle.
2337 The type unification step is performed by the <tt><a
2338 href="#refineAbstractTypeTo">refineAbstractTypeTo</a></tt> method, which is
2339 described next. After that, we describe the <a
2340 href="#PATypeHolder">PATypeHolder class</a>.
2345 <!-- ______________________________________________________________________ -->
2346 <div class="doc_subsubsection">
2347 <a name="refineAbstractTypeTo">The <tt>refineAbstractTypeTo</tt> method</a>
2350 <div class="doc_text">
2352 The <tt>refineAbstractTypeTo</tt> method starts the type unification process.
2353 While this method is actually a member of the DerivedType class, it is most
2354 often used on OpaqueType instances. Type unification is actually a recursive
2355 process. After unification, types can become structurally isomorphic to
2356 existing types, and all duplicates are deleted (to preserve pointer equality).
2360 In the example above, the OpaqueType object is definitely deleted.
2361 Additionally, if there is an "{ \2*, i32}" type already created in the system,
2362 the pointer and struct type created are <b>also</b> deleted. Obviously whenever
2363 a type is deleted, any "Type*" pointers in the program are invalidated. As
2364 such, it is safest to avoid having <i>any</i> "Type*" pointers to abstract types
2365 live across a call to <tt>refineAbstractTypeTo</tt> (note that non-abstract
2366 types can never move or be deleted). To deal with this, the <a
2367 href="#PATypeHolder">PATypeHolder</a> class is used to maintain a stable
2368 reference to a possibly refined type, and the <a
2369 href="#AbstractTypeUser">AbstractTypeUser</a> class is used to update more
2370 complex datastructures.
2375 <!-- ______________________________________________________________________ -->
2376 <div class="doc_subsubsection">
2377 <a name="PATypeHolder">The PATypeHolder Class</a>
2380 <div class="doc_text">
2382 PATypeHolder is a form of a "smart pointer" for Type objects. When VMCore
2383 happily goes about nuking types that become isomorphic to existing types, it
2384 automatically updates all PATypeHolder objects to point to the new type. In the
2385 example above, this allows the code to maintain a pointer to the resultant
2386 resolved recursive type, even though the Type*'s are potentially invalidated.
2390 PATypeHolder is an extremely light-weight object that uses a lazy union-find
2391 implementation to update pointers. For example the pointer from a Value to its
2392 Type is maintained by PATypeHolder objects.
2397 <!-- ______________________________________________________________________ -->
2398 <div class="doc_subsubsection">
2399 <a name="AbstractTypeUser">The AbstractTypeUser Class</a>
2402 <div class="doc_text">
2405 Some data structures need more to perform more complex updates when types get
2406 resolved. To support this, a class can derive from the AbstractTypeUser class.
2408 allows it to get callbacks when certain types are resolved. To register to get
2409 callbacks for a particular type, the DerivedType::{add/remove}AbstractTypeUser
2410 methods can be called on a type. Note that these methods only work for <i>
2411 abstract</i> types. Concrete types (those that do not include any opaque
2412 objects) can never be refined.
2417 <!-- ======================================================================= -->
2418 <div class="doc_subsection">
2419 <a name="SymbolTable">The <tt>ValueSymbolTable</tt> and
2420 <tt>TypeSymbolTable</tt> classes</a>
2423 <div class="doc_text">
2424 <p>The <tt><a href="http://llvm.org/doxygen/classllvm_1_1ValueSymbolTable.html">
2425 ValueSymbolTable</a></tt> class provides a symbol table that the <a
2426 href="#Function"><tt>Function</tt></a> and <a href="#Module">
2427 <tt>Module</tt></a> classes use for naming value definitions. The symbol table
2428 can provide a name for any <a href="#Value"><tt>Value</tt></a>.
2429 The <tt><a href="http://llvm.org/doxygen/classllvm_1_1TypeSymbolTable.html">
2430 TypeSymbolTable</a></tt> class is used by the <tt>Module</tt> class to store
2431 names for types.</p>
2433 <p>Note that the <tt>SymbolTable</tt> class should not be directly accessed
2434 by most clients. It should only be used when iteration over the symbol table
2435 names themselves are required, which is very special purpose. Note that not
2437 <tt><a href="#Value">Value</a></tt>s have names, and those without names (i.e. they have
2438 an empty name) do not exist in the symbol table.
2441 <p>These symbol tables support iteration over the values/types in the symbol
2442 table with <tt>begin/end/iterator</tt> and supports querying to see if a
2443 specific name is in the symbol table (with <tt>lookup</tt>). The
2444 <tt>ValueSymbolTable</tt> class exposes no public mutator methods, instead,
2445 simply call <tt>setName</tt> on a value, which will autoinsert it into the
2446 appropriate symbol table. For types, use the Module::addTypeName method to
2447 insert entries into the symbol table.</p>
2453 <!-- ======================================================================= -->
2454 <div class="doc_subsection">
2455 <a name="UserLayout">The <tt>User</tt> and owned <tt>Use</tt> classes' memory layout</a>
2458 <div class="doc_text">
2459 <p>The <tt><a href="http://llvm.org/doxygen/classllvm_1_1User.html">
2460 User</a></tt> class provides a basis for expressing the ownership of <tt>User</tt>
2461 towards other <tt><a href="http://llvm.org/doxygen/classllvm_1_1Value.html">
2462 Value</a></tt>s. The <tt><a href="http://llvm.org/doxygen/classllvm_1_1Use.html">
2463 Use</a></tt> helper class is employed to do the bookkeeping and to facilitate <i>O(1)</i>
2464 addition and removal.</p>
2466 <!-- ______________________________________________________________________ -->
2467 <div class="doc_subsubsection">
2468 <a name="Use2User">Interaction and relationship between <tt>User</tt> and <tt>Use</tt> objects</a>
2471 <div class="doc_text">
2473 A subclass of <tt>User</tt> can choose between incorporating its <tt>Use</tt> objects
2474 or refer to them out-of-line by means of a pointer. A mixed variant
2475 (some <tt>Use</tt>s inline others hung off) is impractical and breaks the invariant
2476 that the <tt>Use</tt> objects belonging to the same <tt>User</tt> form a contiguous array.
2481 We have 2 different layouts in the <tt>User</tt> (sub)classes:
2484 The <tt>Use</tt> object(s) are inside (resp. at fixed offset) of the <tt>User</tt>
2485 object and there are a fixed number of them.</p>
2488 The <tt>Use</tt> object(s) are referenced by a pointer to an
2489 array from the <tt>User</tt> object and there may be a variable
2493 As of v2.4 each layout still possesses a direct pointer to the
2494 start of the array of <tt>Use</tt>s. Though not mandatory for layout a),
2495 we stick to this redundancy for the sake of simplicity.
2496 The <tt>User</tt> object also stores the number of <tt>Use</tt> objects it
2497 has. (Theoretically this information can also be calculated
2498 given the scheme presented below.)</p>
2500 Special forms of allocation operators (<tt>operator new</tt>)
2501 enforce the following memory layouts:</p>
2504 <li><p>Layout a) is modelled by prepending the <tt>User</tt> object by the <tt>Use[]</tt> array.</p>
2507 ...---.---.---.---.-------...
2508 | P | P | P | P | User
2509 '''---'---'---'---'-------'''
2512 <li><p>Layout b) is modelled by pointing at the <tt>Use[]</tt> array.</p>
2524 <i>(In the above figures '<tt>P</tt>' stands for the <tt>Use**</tt> that
2525 is stored in each <tt>Use</tt> object in the member <tt>Use::Prev</tt>)</i>
2527 <!-- ______________________________________________________________________ -->
2528 <div class="doc_subsubsection">
2529 <a name="Waymarking">The waymarking algorithm</a>
2532 <div class="doc_text">
2534 Since the <tt>Use</tt> objects are deprived of the direct (back)pointer to
2535 their <tt>User</tt> objects, there must be a fast and exact method to
2536 recover it. This is accomplished by the following scheme:</p>
2539 A bit-encoding in the 2 LSBits (least significant bits) of the <tt>Use::Prev</tt> allows to find the
2540 start of the <tt>User</tt> object:
2542 <li><tt>00</tt> —> binary digit 0</li>
2543 <li><tt>01</tt> —> binary digit 1</li>
2544 <li><tt>10</tt> —> stop and calculate (<tt>s</tt>)</li>
2545 <li><tt>11</tt> —> full stop (<tt>S</tt>)</li>
2548 Given a <tt>Use*</tt>, all we have to do is to walk till we get
2549 a stop and we either have a <tt>User</tt> immediately behind or
2550 we have to walk to the next stop picking up digits
2551 and calculating the offset:</p>
2553 .---.---.---.---.---.---.---.---.---.---.---.---.---.---.---.---.----------------
2554 | 1 | s | 1 | 0 | 1 | 0 | s | 1 | 1 | 0 | s | 1 | 1 | s | 1 | S | User (or User*)
2555 '---'---'---'---'---'---'---'---'---'---'---'---'---'---'---'---'----------------
2556 |+15 |+10 |+6 |+3 |+1
2559 | | |______________________>
2560 | |______________________________________>
2561 |__________________________________________________________>
2564 Only the significant number of bits need to be stored between the
2565 stops, so that the <i>worst case is 20 memory accesses</i> when there are
2566 1000 <tt>Use</tt> objects associated with a <tt>User</tt>.</p>
2568 <!-- ______________________________________________________________________ -->
2569 <div class="doc_subsubsection">
2570 <a name="ReferenceImpl">Reference implementation</a>
2573 <div class="doc_text">
2575 The following literate Haskell fragment demonstrates the concept:</p>
2578 <div class="doc_code">
2580 > import Test.QuickCheck
2582 > digits :: Int -> [Char] -> [Char]
2583 > digits 0 acc = '0' : acc
2584 > digits 1 acc = '1' : acc
2585 > digits n acc = digits (n `div` 2) $ digits (n `mod` 2) acc
2587 > dist :: Int -> [Char] -> [Char]
2590 > dist 1 acc = let r = dist 0 acc in 's' : digits (length r) r
2591 > dist n acc = dist (n - 1) $ dist 1 acc
2593 > takeLast n ss = reverse $ take n $ reverse ss
2595 > test = takeLast 40 $ dist 20 []
2600 Printing <test> gives: <tt>"1s100000s11010s10100s1111s1010s110s11s1S"</tt></p>
2602 The reverse algorithm computes the length of the string just by examining
2603 a certain prefix:</p>
2605 <div class="doc_code">
2607 > pref :: [Char] -> Int
2609 > pref ('s':'1':rest) = decode 2 1 rest
2610 > pref (_:rest) = 1 + pref rest
2612 > decode walk acc ('0':rest) = decode (walk + 1) (acc * 2) rest
2613 > decode walk acc ('1':rest) = decode (walk + 1) (acc * 2 + 1) rest
2614 > decode walk acc _ = walk + acc
2619 Now, as expected, printing <pref test> gives <tt>40</tt>.</p>
2621 We can <i>quickCheck</i> this with following property:</p>
2623 <div class="doc_code">
2625 > testcase = dist 2000 []
2626 > testcaseLength = length testcase
2628 > identityProp n = n > 0 && n <= testcaseLength ==> length arr == pref arr
2629 > where arr = takeLast n testcase
2634 As expected <quickCheck identityProp> gives:</p>
2637 *Main> quickCheck identityProp
2638 OK, passed 100 tests.
2641 Let's be a bit more exhaustive:</p>
2643 <div class="doc_code">
2646 > deepCheck p = check (defaultConfig { configMaxTest = 500 }) p
2651 And here is the result of <deepCheck identityProp>:</p>
2654 *Main> deepCheck identityProp
2655 OK, passed 500 tests.
2658 <!-- ______________________________________________________________________ -->
2659 <div class="doc_subsubsection">
2660 <a name="Tagging">Tagging considerations</a>
2664 To maintain the invariant that the 2 LSBits of each <tt>Use**</tt> in <tt>Use</tt>
2665 never change after being set up, setters of <tt>Use::Prev</tt> must re-tag the
2666 new <tt>Use**</tt> on every modification. Accordingly getters must strip the
2669 For layout b) instead of the <tt>User</tt> we find a pointer (<tt>User*</tt> with LSBit set).
2670 Following this pointer brings us to the <tt>User</tt>. A portable trick ensures
2671 that the first bytes of <tt>User</tt> (if interpreted as a pointer) never has
2672 the LSBit set. (Portability is relying on the fact that all known compilers place the
2673 <tt>vptr</tt> in the first word of the instances.)</p>
2677 <!-- *********************************************************************** -->
2678 <div class="doc_section">
2679 <a name="coreclasses">The Core LLVM Class Hierarchy Reference </a>
2681 <!-- *********************************************************************** -->
2683 <div class="doc_text">
2684 <p><tt>#include "<a href="/doxygen/Type_8h-source.html">llvm/Type.h</a>"</tt>
2685 <br>doxygen info: <a href="/doxygen/classllvm_1_1Type.html">Type Class</a></p>
2687 <p>The Core LLVM classes are the primary means of representing the program
2688 being inspected or transformed. The core LLVM classes are defined in
2689 header files in the <tt>include/llvm/</tt> directory, and implemented in
2690 the <tt>lib/VMCore</tt> directory.</p>
2694 <!-- ======================================================================= -->
2695 <div class="doc_subsection">
2696 <a name="Type">The <tt>Type</tt> class and Derived Types</a>
2699 <div class="doc_text">
2701 <p><tt>Type</tt> is a superclass of all type classes. Every <tt>Value</tt> has
2702 a <tt>Type</tt>. <tt>Type</tt> cannot be instantiated directly but only
2703 through its subclasses. Certain primitive types (<tt>VoidType</tt>,
2704 <tt>LabelType</tt>, <tt>FloatType</tt> and <tt>DoubleType</tt>) have hidden
2705 subclasses. They are hidden because they offer no useful functionality beyond
2706 what the <tt>Type</tt> class offers except to distinguish themselves from
2707 other subclasses of <tt>Type</tt>.</p>
2708 <p>All other types are subclasses of <tt>DerivedType</tt>. Types can be
2709 named, but this is not a requirement. There exists exactly
2710 one instance of a given shape at any one time. This allows type equality to
2711 be performed with address equality of the Type Instance. That is, given two
2712 <tt>Type*</tt> values, the types are identical if the pointers are identical.
2716 <!-- _______________________________________________________________________ -->
2717 <div class="doc_subsubsection">
2718 <a name="m_Type">Important Public Methods</a>
2721 <div class="doc_text">
2724 <li><tt>bool isInteger() const</tt>: Returns true for any integer type.</li>
2726 <li><tt>bool isFloatingPoint()</tt>: Return true if this is one of the two
2727 floating point types.</li>
2729 <li><tt>bool isAbstract()</tt>: Return true if the type is abstract (contains
2730 an OpaqueType anywhere in its definition).</li>
2732 <li><tt>bool isSized()</tt>: Return true if the type has known size. Things
2733 that don't have a size are abstract types, labels and void.</li>
2738 <!-- _______________________________________________________________________ -->
2739 <div class="doc_subsubsection">
2740 <a name="derivedtypes">Important Derived Types</a>
2742 <div class="doc_text">
2744 <dt><tt>IntegerType</tt></dt>
2745 <dd>Subclass of DerivedType that represents integer types of any bit width.
2746 Any bit width between <tt>IntegerType::MIN_INT_BITS</tt> (1) and
2747 <tt>IntegerType::MAX_INT_BITS</tt> (~8 million) can be represented.
2749 <li><tt>static const IntegerType* get(unsigned NumBits)</tt>: get an integer
2750 type of a specific bit width.</li>
2751 <li><tt>unsigned getBitWidth() const</tt>: Get the bit width of an integer
2755 <dt><tt>SequentialType</tt></dt>
2756 <dd>This is subclassed by ArrayType and PointerType
2758 <li><tt>const Type * getElementType() const</tt>: Returns the type of each
2759 of the elements in the sequential type. </li>
2762 <dt><tt>ArrayType</tt></dt>
2763 <dd>This is a subclass of SequentialType and defines the interface for array
2766 <li><tt>unsigned getNumElements() const</tt>: Returns the number of
2767 elements in the array. </li>
2770 <dt><tt>PointerType</tt></dt>
2771 <dd>Subclass of SequentialType for pointer types.</dd>
2772 <dt><tt>VectorType</tt></dt>
2773 <dd>Subclass of SequentialType for vector types. A
2774 vector type is similar to an ArrayType but is distinguished because it is
2775 a first class type wherease ArrayType is not. Vector types are used for
2776 vector operations and are usually small vectors of of an integer or floating
2778 <dt><tt>StructType</tt></dt>
2779 <dd>Subclass of DerivedTypes for struct types.</dd>
2780 <dt><tt><a name="FunctionType">FunctionType</a></tt></dt>
2781 <dd>Subclass of DerivedTypes for function types.
2783 <li><tt>bool isVarArg() const</tt>: Returns true if its a vararg
2785 <li><tt> const Type * getReturnType() const</tt>: Returns the
2786 return type of the function.</li>
2787 <li><tt>const Type * getParamType (unsigned i)</tt>: Returns
2788 the type of the ith parameter.</li>
2789 <li><tt> const unsigned getNumParams() const</tt>: Returns the
2790 number of formal parameters.</li>
2793 <dt><tt>OpaqueType</tt></dt>
2794 <dd>Sublcass of DerivedType for abstract types. This class
2795 defines no content and is used as a placeholder for some other type. Note
2796 that OpaqueType is used (temporarily) during type resolution for forward
2797 references of types. Once the referenced type is resolved, the OpaqueType
2798 is replaced with the actual type. OpaqueType can also be used for data
2799 abstraction. At link time opaque types can be resolved to actual types
2800 of the same name.</dd>
2806 <!-- ======================================================================= -->
2807 <div class="doc_subsection">
2808 <a name="Module">The <tt>Module</tt> class</a>
2811 <div class="doc_text">
2814 href="/doxygen/Module_8h-source.html">llvm/Module.h</a>"</tt><br> doxygen info:
2815 <a href="/doxygen/classllvm_1_1Module.html">Module Class</a></p>
2817 <p>The <tt>Module</tt> class represents the top level structure present in LLVM
2818 programs. An LLVM module is effectively either a translation unit of the
2819 original program or a combination of several translation units merged by the
2820 linker. The <tt>Module</tt> class keeps track of a list of <a
2821 href="#Function"><tt>Function</tt></a>s, a list of <a
2822 href="#GlobalVariable"><tt>GlobalVariable</tt></a>s, and a <a
2823 href="#SymbolTable"><tt>SymbolTable</tt></a>. Additionally, it contains a few
2824 helpful member functions that try to make common operations easy.</p>
2828 <!-- _______________________________________________________________________ -->
2829 <div class="doc_subsubsection">
2830 <a name="m_Module">Important Public Members of the <tt>Module</tt> class</a>
2833 <div class="doc_text">
2836 <li><tt>Module::Module(std::string name = "")</tt></li>
2839 <p>Constructing a <a href="#Module">Module</a> is easy. You can optionally
2840 provide a name for it (probably based on the name of the translation unit).</p>
2843 <li><tt>Module::iterator</tt> - Typedef for function list iterator<br>
2844 <tt>Module::const_iterator</tt> - Typedef for const_iterator.<br>
2846 <tt>begin()</tt>, <tt>end()</tt>
2847 <tt>size()</tt>, <tt>empty()</tt>
2849 <p>These are forwarding methods that make it easy to access the contents of
2850 a <tt>Module</tt> object's <a href="#Function"><tt>Function</tt></a>
2853 <li><tt>Module::FunctionListType &getFunctionList()</tt>
2855 <p> Returns the list of <a href="#Function"><tt>Function</tt></a>s. This is
2856 necessary to use when you need to update the list or perform a complex
2857 action that doesn't have a forwarding method.</p>
2859 <p><!-- Global Variable --></p></li>
2865 <li><tt>Module::global_iterator</tt> - Typedef for global variable list iterator<br>
2867 <tt>Module::const_global_iterator</tt> - Typedef for const_iterator.<br>
2869 <tt>global_begin()</tt>, <tt>global_end()</tt>
2870 <tt>global_size()</tt>, <tt>global_empty()</tt>
2872 <p> These are forwarding methods that make it easy to access the contents of
2873 a <tt>Module</tt> object's <a
2874 href="#GlobalVariable"><tt>GlobalVariable</tt></a> list.</p></li>
2876 <li><tt>Module::GlobalListType &getGlobalList()</tt>
2878 <p>Returns the list of <a
2879 href="#GlobalVariable"><tt>GlobalVariable</tt></a>s. This is necessary to
2880 use when you need to update the list or perform a complex action that
2881 doesn't have a forwarding method.</p>
2883 <p><!-- Symbol table stuff --> </p></li>
2889 <li><tt><a href="#SymbolTable">SymbolTable</a> *getSymbolTable()</tt>
2891 <p>Return a reference to the <a href="#SymbolTable"><tt>SymbolTable</tt></a>
2892 for this <tt>Module</tt>.</p>
2894 <p><!-- Convenience methods --></p></li>
2900 <li><tt><a href="#Function">Function</a> *getFunction(const std::string
2901 &Name, const <a href="#FunctionType">FunctionType</a> *Ty)</tt>
2903 <p>Look up the specified function in the <tt>Module</tt> <a
2904 href="#SymbolTable"><tt>SymbolTable</tt></a>. If it does not exist, return
2905 <tt>null</tt>.</p></li>
2907 <li><tt><a href="#Function">Function</a> *getOrInsertFunction(const
2908 std::string &Name, const <a href="#FunctionType">FunctionType</a> *T)</tt>
2910 <p>Look up the specified function in the <tt>Module</tt> <a
2911 href="#SymbolTable"><tt>SymbolTable</tt></a>. If it does not exist, add an
2912 external declaration for the function and return it.</p></li>
2914 <li><tt>std::string getTypeName(const <a href="#Type">Type</a> *Ty)</tt>
2916 <p>If there is at least one entry in the <a
2917 href="#SymbolTable"><tt>SymbolTable</tt></a> for the specified <a
2918 href="#Type"><tt>Type</tt></a>, return it. Otherwise return the empty
2921 <li><tt>bool addTypeName(const std::string &Name, const <a
2922 href="#Type">Type</a> *Ty)</tt>
2924 <p>Insert an entry in the <a href="#SymbolTable"><tt>SymbolTable</tt></a>
2925 mapping <tt>Name</tt> to <tt>Ty</tt>. If there is already an entry for this
2926 name, true is returned and the <a
2927 href="#SymbolTable"><tt>SymbolTable</tt></a> is not modified.</p></li>
2933 <!-- ======================================================================= -->
2934 <div class="doc_subsection">
2935 <a name="Value">The <tt>Value</tt> class</a>
2938 <div class="doc_text">
2940 <p><tt>#include "<a href="/doxygen/Value_8h-source.html">llvm/Value.h</a>"</tt>
2942 doxygen info: <a href="/doxygen/classllvm_1_1Value.html">Value Class</a></p>
2944 <p>The <tt>Value</tt> class is the most important class in the LLVM Source
2945 base. It represents a typed value that may be used (among other things) as an
2946 operand to an instruction. There are many different types of <tt>Value</tt>s,
2947 such as <a href="#Constant"><tt>Constant</tt></a>s,<a
2948 href="#Argument"><tt>Argument</tt></a>s. Even <a
2949 href="#Instruction"><tt>Instruction</tt></a>s and <a
2950 href="#Function"><tt>Function</tt></a>s are <tt>Value</tt>s.</p>
2952 <p>A particular <tt>Value</tt> may be used many times in the LLVM representation
2953 for a program. For example, an incoming argument to a function (represented
2954 with an instance of the <a href="#Argument">Argument</a> class) is "used" by
2955 every instruction in the function that references the argument. To keep track
2956 of this relationship, the <tt>Value</tt> class keeps a list of all of the <a
2957 href="#User"><tt>User</tt></a>s that is using it (the <a
2958 href="#User"><tt>User</tt></a> class is a base class for all nodes in the LLVM
2959 graph that can refer to <tt>Value</tt>s). This use list is how LLVM represents
2960 def-use information in the program, and is accessible through the <tt>use_</tt>*
2961 methods, shown below.</p>
2963 <p>Because LLVM is a typed representation, every LLVM <tt>Value</tt> is typed,
2964 and this <a href="#Type">Type</a> is available through the <tt>getType()</tt>
2965 method. In addition, all LLVM values can be named. The "name" of the
2966 <tt>Value</tt> is a symbolic string printed in the LLVM code:</p>
2968 <div class="doc_code">
2970 %<b>foo</b> = add i32 1, 2
2974 <p><a name="nameWarning">The name of this instruction is "foo".</a> <b>NOTE</b>
2975 that the name of any value may be missing (an empty string), so names should
2976 <b>ONLY</b> be used for debugging (making the source code easier to read,
2977 debugging printouts), they should not be used to keep track of values or map
2978 between them. For this purpose, use a <tt>std::map</tt> of pointers to the
2979 <tt>Value</tt> itself instead.</p>
2981 <p>One important aspect of LLVM is that there is no distinction between an SSA
2982 variable and the operation that produces it. Because of this, any reference to
2983 the value produced by an instruction (or the value available as an incoming
2984 argument, for example) is represented as a direct pointer to the instance of
2986 represents this value. Although this may take some getting used to, it
2987 simplifies the representation and makes it easier to manipulate.</p>
2991 <!-- _______________________________________________________________________ -->
2992 <div class="doc_subsubsection">
2993 <a name="m_Value">Important Public Members of the <tt>Value</tt> class</a>
2996 <div class="doc_text">
2999 <li><tt>Value::use_iterator</tt> - Typedef for iterator over the
3001 <tt>Value::use_const_iterator</tt> - Typedef for const_iterator over
3003 <tt>unsigned use_size()</tt> - Returns the number of users of the
3005 <tt>bool use_empty()</tt> - Returns true if there are no users.<br>
3006 <tt>use_iterator use_begin()</tt> - Get an iterator to the start of
3008 <tt>use_iterator use_end()</tt> - Get an iterator to the end of the
3010 <tt><a href="#User">User</a> *use_back()</tt> - Returns the last
3011 element in the list.
3012 <p> These methods are the interface to access the def-use
3013 information in LLVM. As with all other iterators in LLVM, the naming
3014 conventions follow the conventions defined by the <a href="#stl">STL</a>.</p>
3016 <li><tt><a href="#Type">Type</a> *getType() const</tt>
3017 <p>This method returns the Type of the Value.</p>
3019 <li><tt>bool hasName() const</tt><br>
3020 <tt>std::string getName() const</tt><br>
3021 <tt>void setName(const std::string &Name)</tt>
3022 <p> This family of methods is used to access and assign a name to a <tt>Value</tt>,
3023 be aware of the <a href="#nameWarning">precaution above</a>.</p>
3025 <li><tt>void replaceAllUsesWith(Value *V)</tt>
3027 <p>This method traverses the use list of a <tt>Value</tt> changing all <a
3028 href="#User"><tt>User</tt>s</a> of the current value to refer to
3029 "<tt>V</tt>" instead. For example, if you detect that an instruction always
3030 produces a constant value (for example through constant folding), you can
3031 replace all uses of the instruction with the constant like this:</p>
3033 <div class="doc_code">
3035 Inst->replaceAllUsesWith(ConstVal);
3043 <!-- ======================================================================= -->
3044 <div class="doc_subsection">
3045 <a name="User">The <tt>User</tt> class</a>
3048 <div class="doc_text">
3051 <tt>#include "<a href="/doxygen/User_8h-source.html">llvm/User.h</a>"</tt><br>
3052 doxygen info: <a href="/doxygen/classllvm_1_1User.html">User Class</a><br>
3053 Superclass: <a href="#Value"><tt>Value</tt></a></p>
3055 <p>The <tt>User</tt> class is the common base class of all LLVM nodes that may
3056 refer to <a href="#Value"><tt>Value</tt></a>s. It exposes a list of "Operands"
3057 that are all of the <a href="#Value"><tt>Value</tt></a>s that the User is
3058 referring to. The <tt>User</tt> class itself is a subclass of
3061 <p>The operands of a <tt>User</tt> point directly to the LLVM <a
3062 href="#Value"><tt>Value</tt></a> that it refers to. Because LLVM uses Static
3063 Single Assignment (SSA) form, there can only be one definition referred to,
3064 allowing this direct connection. This connection provides the use-def
3065 information in LLVM.</p>
3069 <!-- _______________________________________________________________________ -->
3070 <div class="doc_subsubsection">
3071 <a name="m_User">Important Public Members of the <tt>User</tt> class</a>
3074 <div class="doc_text">
3076 <p>The <tt>User</tt> class exposes the operand list in two ways: through
3077 an index access interface and through an iterator based interface.</p>
3080 <li><tt>Value *getOperand(unsigned i)</tt><br>
3081 <tt>unsigned getNumOperands()</tt>
3082 <p> These two methods expose the operands of the <tt>User</tt> in a
3083 convenient form for direct access.</p></li>
3085 <li><tt>User::op_iterator</tt> - Typedef for iterator over the operand
3087 <tt>op_iterator op_begin()</tt> - Get an iterator to the start of
3088 the operand list.<br>
3089 <tt>op_iterator op_end()</tt> - Get an iterator to the end of the
3091 <p> Together, these methods make up the iterator based interface to
3092 the operands of a <tt>User</tt>.</p></li>
3097 <!-- ======================================================================= -->
3098 <div class="doc_subsection">
3099 <a name="Instruction">The <tt>Instruction</tt> class</a>
3102 <div class="doc_text">
3104 <p><tt>#include "</tt><tt><a
3105 href="/doxygen/Instruction_8h-source.html">llvm/Instruction.h</a>"</tt><br>
3106 doxygen info: <a href="/doxygen/classllvm_1_1Instruction.html">Instruction Class</a><br>
3107 Superclasses: <a href="#User"><tt>User</tt></a>, <a
3108 href="#Value"><tt>Value</tt></a></p>
3110 <p>The <tt>Instruction</tt> class is the common base class for all LLVM
3111 instructions. It provides only a few methods, but is a very commonly used
3112 class. The primary data tracked by the <tt>Instruction</tt> class itself is the
3113 opcode (instruction type) and the parent <a
3114 href="#BasicBlock"><tt>BasicBlock</tt></a> the <tt>Instruction</tt> is embedded
3115 into. To represent a specific type of instruction, one of many subclasses of
3116 <tt>Instruction</tt> are used.</p>
3118 <p> Because the <tt>Instruction</tt> class subclasses the <a
3119 href="#User"><tt>User</tt></a> class, its operands can be accessed in the same
3120 way as for other <a href="#User"><tt>User</tt></a>s (with the
3121 <tt>getOperand()</tt>/<tt>getNumOperands()</tt> and
3122 <tt>op_begin()</tt>/<tt>op_end()</tt> methods).</p> <p> An important file for
3123 the <tt>Instruction</tt> class is the <tt>llvm/Instruction.def</tt> file. This
3124 file contains some meta-data about the various different types of instructions
3125 in LLVM. It describes the enum values that are used as opcodes (for example
3126 <tt>Instruction::Add</tt> and <tt>Instruction::ICmp</tt>), as well as the
3127 concrete sub-classes of <tt>Instruction</tt> that implement the instruction (for
3128 example <tt><a href="#BinaryOperator">BinaryOperator</a></tt> and <tt><a
3129 href="#CmpInst">CmpInst</a></tt>). Unfortunately, the use of macros in
3130 this file confuses doxygen, so these enum values don't show up correctly in the
3131 <a href="/doxygen/classllvm_1_1Instruction.html">doxygen output</a>.</p>
3135 <!-- _______________________________________________________________________ -->
3136 <div class="doc_subsubsection">
3137 <a name="s_Instruction">Important Subclasses of the <tt>Instruction</tt>
3140 <div class="doc_text">
3142 <li><tt><a name="BinaryOperator">BinaryOperator</a></tt>
3143 <p>This subclasses represents all two operand instructions whose operands
3144 must be the same type, except for the comparison instructions.</p></li>
3145 <li><tt><a name="CastInst">CastInst</a></tt>
3146 <p>This subclass is the parent of the 12 casting instructions. It provides
3147 common operations on cast instructions.</p>
3148 <li><tt><a name="CmpInst">CmpInst</a></tt>
3149 <p>This subclass respresents the two comparison instructions,
3150 <a href="LangRef.html#i_icmp">ICmpInst</a> (integer opreands), and
3151 <a href="LangRef.html#i_fcmp">FCmpInst</a> (floating point operands).</p>
3152 <li><tt><a name="TerminatorInst">TerminatorInst</a></tt>
3153 <p>This subclass is the parent of all terminator instructions (those which
3154 can terminate a block).</p>
3158 <!-- _______________________________________________________________________ -->
3159 <div class="doc_subsubsection">
3160 <a name="m_Instruction">Important Public Members of the <tt>Instruction</tt>
3164 <div class="doc_text">
3167 <li><tt><a href="#BasicBlock">BasicBlock</a> *getParent()</tt>
3168 <p>Returns the <a href="#BasicBlock"><tt>BasicBlock</tt></a> that
3169 this <tt>Instruction</tt> is embedded into.</p></li>
3170 <li><tt>bool mayWriteToMemory()</tt>
3171 <p>Returns true if the instruction writes to memory, i.e. it is a
3172 <tt>call</tt>,<tt>free</tt>,<tt>invoke</tt>, or <tt>store</tt>.</p></li>
3173 <li><tt>unsigned getOpcode()</tt>
3174 <p>Returns the opcode for the <tt>Instruction</tt>.</p></li>
3175 <li><tt><a href="#Instruction">Instruction</a> *clone() const</tt>
3176 <p>Returns another instance of the specified instruction, identical
3177 in all ways to the original except that the instruction has no parent
3178 (ie it's not embedded into a <a href="#BasicBlock"><tt>BasicBlock</tt></a>),
3179 and it has no name</p></li>
3184 <!-- ======================================================================= -->
3185 <div class="doc_subsection">
3186 <a name="Constant">The <tt>Constant</tt> class and subclasses</a>
3189 <div class="doc_text">
3191 <p>Constant represents a base class for different types of constants. It
3192 is subclassed by ConstantInt, ConstantArray, etc. for representing
3193 the various types of Constants. <a href="#GlobalValue">GlobalValue</a> is also
3194 a subclass, which represents the address of a global variable or function.
3199 <!-- _______________________________________________________________________ -->
3200 <div class="doc_subsubsection">Important Subclasses of Constant </div>
3201 <div class="doc_text">
3203 <li>ConstantInt : This subclass of Constant represents an integer constant of
3206 <li><tt>const APInt& getValue() const</tt>: Returns the underlying
3207 value of this constant, an APInt value.</li>
3208 <li><tt>int64_t getSExtValue() const</tt>: Converts the underlying APInt
3209 value to an int64_t via sign extension. If the value (not the bit width)
3210 of the APInt is too large to fit in an int64_t, an assertion will result.
3211 For this reason, use of this method is discouraged.</li>
3212 <li><tt>uint64_t getZExtValue() const</tt>: Converts the underlying APInt
3213 value to a uint64_t via zero extension. IF the value (not the bit width)
3214 of the APInt is too large to fit in a uint64_t, an assertion will result.
3215 For this reason, use of this method is discouraged.</li>
3216 <li><tt>static ConstantInt* get(const APInt& Val)</tt>: Returns the
3217 ConstantInt object that represents the value provided by <tt>Val</tt>.
3218 The type is implied as the IntegerType that corresponds to the bit width
3219 of <tt>Val</tt>.</li>
3220 <li><tt>static ConstantInt* get(const Type *Ty, uint64_t Val)</tt>:
3221 Returns the ConstantInt object that represents the value provided by
3222 <tt>Val</tt> for integer type <tt>Ty</tt>.</li>
3225 <li>ConstantFP : This class represents a floating point constant.
3227 <li><tt>double getValue() const</tt>: Returns the underlying value of
3228 this constant. </li>
3231 <li>ConstantArray : This represents a constant array.
3233 <li><tt>const std::vector<Use> &getValues() const</tt>: Returns
3234 a vector of component constants that makeup this array. </li>
3237 <li>ConstantStruct : This represents a constant struct.
3239 <li><tt>const std::vector<Use> &getValues() const</tt>: Returns
3240 a vector of component constants that makeup this array. </li>
3243 <li>GlobalValue : This represents either a global variable or a function. In
3244 either case, the value is a constant fixed address (after linking).
3250 <!-- ======================================================================= -->
3251 <div class="doc_subsection">
3252 <a name="GlobalValue">The <tt>GlobalValue</tt> class</a>
3255 <div class="doc_text">
3258 href="/doxygen/GlobalValue_8h-source.html">llvm/GlobalValue.h</a>"</tt><br>
3259 doxygen info: <a href="/doxygen/classllvm_1_1GlobalValue.html">GlobalValue
3261 Superclasses: <a href="#Constant"><tt>Constant</tt></a>,
3262 <a href="#User"><tt>User</tt></a>, <a href="#Value"><tt>Value</tt></a></p>
3264 <p>Global values (<a href="#GlobalVariable"><tt>GlobalVariable</tt></a>s or <a
3265 href="#Function"><tt>Function</tt></a>s) are the only LLVM values that are
3266 visible in the bodies of all <a href="#Function"><tt>Function</tt></a>s.
3267 Because they are visible at global scope, they are also subject to linking with
3268 other globals defined in different translation units. To control the linking
3269 process, <tt>GlobalValue</tt>s know their linkage rules. Specifically,
3270 <tt>GlobalValue</tt>s know whether they have internal or external linkage, as
3271 defined by the <tt>LinkageTypes</tt> enumeration.</p>
3273 <p>If a <tt>GlobalValue</tt> has internal linkage (equivalent to being
3274 <tt>static</tt> in C), it is not visible to code outside the current translation
3275 unit, and does not participate in linking. If it has external linkage, it is
3276 visible to external code, and does participate in linking. In addition to
3277 linkage information, <tt>GlobalValue</tt>s keep track of which <a
3278 href="#Module"><tt>Module</tt></a> they are currently part of.</p>
3280 <p>Because <tt>GlobalValue</tt>s are memory objects, they are always referred to
3281 by their <b>address</b>. As such, the <a href="#Type"><tt>Type</tt></a> of a
3282 global is always a pointer to its contents. It is important to remember this
3283 when using the <tt>GetElementPtrInst</tt> instruction because this pointer must
3284 be dereferenced first. For example, if you have a <tt>GlobalVariable</tt> (a
3285 subclass of <tt>GlobalValue)</tt> that is an array of 24 ints, type <tt>[24 x
3286 i32]</tt>, then the <tt>GlobalVariable</tt> is a pointer to that array. Although
3287 the address of the first element of this array and the value of the
3288 <tt>GlobalVariable</tt> are the same, they have different types. The
3289 <tt>GlobalVariable</tt>'s type is <tt>[24 x i32]</tt>. The first element's type
3290 is <tt>i32.</tt> Because of this, accessing a global value requires you to
3291 dereference the pointer with <tt>GetElementPtrInst</tt> first, then its elements
3292 can be accessed. This is explained in the <a href="LangRef.html#globalvars">LLVM
3293 Language Reference Manual</a>.</p>
3297 <!-- _______________________________________________________________________ -->
3298 <div class="doc_subsubsection">
3299 <a name="m_GlobalValue">Important Public Members of the <tt>GlobalValue</tt>
3303 <div class="doc_text">
3306 <li><tt>bool hasInternalLinkage() const</tt><br>
3307 <tt>bool hasExternalLinkage() const</tt><br>
3308 <tt>void setInternalLinkage(bool HasInternalLinkage)</tt>
3309 <p> These methods manipulate the linkage characteristics of the <tt>GlobalValue</tt>.</p>
3312 <li><tt><a href="#Module">Module</a> *getParent()</tt>
3313 <p> This returns the <a href="#Module"><tt>Module</tt></a> that the
3314 GlobalValue is currently embedded into.</p></li>
3319 <!-- ======================================================================= -->
3320 <div class="doc_subsection">
3321 <a name="Function">The <tt>Function</tt> class</a>
3324 <div class="doc_text">
3327 href="/doxygen/Function_8h-source.html">llvm/Function.h</a>"</tt><br> doxygen
3328 info: <a href="/doxygen/classllvm_1_1Function.html">Function Class</a><br>
3329 Superclasses: <a href="#GlobalValue"><tt>GlobalValue</tt></a>,
3330 <a href="#Constant"><tt>Constant</tt></a>,
3331 <a href="#User"><tt>User</tt></a>,
3332 <a href="#Value"><tt>Value</tt></a></p>
3334 <p>The <tt>Function</tt> class represents a single procedure in LLVM. It is
3335 actually one of the more complex classes in the LLVM heirarchy because it must
3336 keep track of a large amount of data. The <tt>Function</tt> class keeps track
3337 of a list of <a href="#BasicBlock"><tt>BasicBlock</tt></a>s, a list of formal
3338 <a href="#Argument"><tt>Argument</tt></a>s, and a
3339 <a href="#SymbolTable"><tt>SymbolTable</tt></a>.</p>
3341 <p>The list of <a href="#BasicBlock"><tt>BasicBlock</tt></a>s is the most
3342 commonly used part of <tt>Function</tt> objects. The list imposes an implicit
3343 ordering of the blocks in the function, which indicate how the code will be
3344 layed out by the backend. Additionally, the first <a
3345 href="#BasicBlock"><tt>BasicBlock</tt></a> is the implicit entry node for the
3346 <tt>Function</tt>. It is not legal in LLVM to explicitly branch to this initial
3347 block. There are no implicit exit nodes, and in fact there may be multiple exit
3348 nodes from a single <tt>Function</tt>. If the <a
3349 href="#BasicBlock"><tt>BasicBlock</tt></a> list is empty, this indicates that
3350 the <tt>Function</tt> is actually a function declaration: the actual body of the
3351 function hasn't been linked in yet.</p>
3353 <p>In addition to a list of <a href="#BasicBlock"><tt>BasicBlock</tt></a>s, the
3354 <tt>Function</tt> class also keeps track of the list of formal <a
3355 href="#Argument"><tt>Argument</tt></a>s that the function receives. This
3356 container manages the lifetime of the <a href="#Argument"><tt>Argument</tt></a>
3357 nodes, just like the <a href="#BasicBlock"><tt>BasicBlock</tt></a> list does for
3358 the <a href="#BasicBlock"><tt>BasicBlock</tt></a>s.</p>
3360 <p>The <a href="#SymbolTable"><tt>SymbolTable</tt></a> is a very rarely used
3361 LLVM feature that is only used when you have to look up a value by name. Aside
3362 from that, the <a href="#SymbolTable"><tt>SymbolTable</tt></a> is used
3363 internally to make sure that there are not conflicts between the names of <a
3364 href="#Instruction"><tt>Instruction</tt></a>s, <a
3365 href="#BasicBlock"><tt>BasicBlock</tt></a>s, or <a
3366 href="#Argument"><tt>Argument</tt></a>s in the function body.</p>
3368 <p>Note that <tt>Function</tt> is a <a href="#GlobalValue">GlobalValue</a>
3369 and therefore also a <a href="#Constant">Constant</a>. The value of the function
3370 is its address (after linking) which is guaranteed to be constant.</p>
3373 <!-- _______________________________________________________________________ -->
3374 <div class="doc_subsubsection">
3375 <a name="m_Function">Important Public Members of the <tt>Function</tt>
3379 <div class="doc_text">
3382 <li><tt>Function(const </tt><tt><a href="#FunctionType">FunctionType</a>
3383 *Ty, LinkageTypes Linkage, const std::string &N = "", Module* Parent = 0)</tt>
3385 <p>Constructor used when you need to create new <tt>Function</tt>s to add
3386 the the program. The constructor must specify the type of the function to
3387 create and what type of linkage the function should have. The <a
3388 href="#FunctionType"><tt>FunctionType</tt></a> argument
3389 specifies the formal arguments and return value for the function. The same
3390 <a href="#FunctionType"><tt>FunctionType</tt></a> value can be used to
3391 create multiple functions. The <tt>Parent</tt> argument specifies the Module
3392 in which the function is defined. If this argument is provided, the function
3393 will automatically be inserted into that module's list of
3396 <li><tt>bool isDeclaration()</tt>
3398 <p>Return whether or not the <tt>Function</tt> has a body defined. If the
3399 function is "external", it does not have a body, and thus must be resolved
3400 by linking with a function defined in a different translation unit.</p></li>
3402 <li><tt>Function::iterator</tt> - Typedef for basic block list iterator<br>
3403 <tt>Function::const_iterator</tt> - Typedef for const_iterator.<br>
3405 <tt>begin()</tt>, <tt>end()</tt>
3406 <tt>size()</tt>, <tt>empty()</tt>
3408 <p>These are forwarding methods that make it easy to access the contents of
3409 a <tt>Function</tt> object's <a href="#BasicBlock"><tt>BasicBlock</tt></a>
3412 <li><tt>Function::BasicBlockListType &getBasicBlockList()</tt>
3414 <p>Returns the list of <a href="#BasicBlock"><tt>BasicBlock</tt></a>s. This
3415 is necessary to use when you need to update the list or perform a complex
3416 action that doesn't have a forwarding method.</p></li>
3418 <li><tt>Function::arg_iterator</tt> - Typedef for the argument list
3420 <tt>Function::const_arg_iterator</tt> - Typedef for const_iterator.<br>
3422 <tt>arg_begin()</tt>, <tt>arg_end()</tt>
3423 <tt>arg_size()</tt>, <tt>arg_empty()</tt>
3425 <p>These are forwarding methods that make it easy to access the contents of
3426 a <tt>Function</tt> object's <a href="#Argument"><tt>Argument</tt></a>
3429 <li><tt>Function::ArgumentListType &getArgumentList()</tt>
3431 <p>Returns the list of <a href="#Argument"><tt>Argument</tt></a>s. This is
3432 necessary to use when you need to update the list or perform a complex
3433 action that doesn't have a forwarding method.</p></li>
3435 <li><tt><a href="#BasicBlock">BasicBlock</a> &getEntryBlock()</tt>
3437 <p>Returns the entry <a href="#BasicBlock"><tt>BasicBlock</tt></a> for the
3438 function. Because the entry block for the function is always the first
3439 block, this returns the first block of the <tt>Function</tt>.</p></li>
3441 <li><tt><a href="#Type">Type</a> *getReturnType()</tt><br>
3442 <tt><a href="#FunctionType">FunctionType</a> *getFunctionType()</tt>
3444 <p>This traverses the <a href="#Type"><tt>Type</tt></a> of the
3445 <tt>Function</tt> and returns the return type of the function, or the <a
3446 href="#FunctionType"><tt>FunctionType</tt></a> of the actual
3449 <li><tt><a href="#SymbolTable">SymbolTable</a> *getSymbolTable()</tt>
3451 <p> Return a pointer to the <a href="#SymbolTable"><tt>SymbolTable</tt></a>
3452 for this <tt>Function</tt>.</p></li>
3457 <!-- ======================================================================= -->
3458 <div class="doc_subsection">
3459 <a name="GlobalVariable">The <tt>GlobalVariable</tt> class</a>
3462 <div class="doc_text">
3465 href="/doxygen/GlobalVariable_8h-source.html">llvm/GlobalVariable.h</a>"</tt>
3467 doxygen info: <a href="/doxygen/classllvm_1_1GlobalVariable.html">GlobalVariable
3469 Superclasses: <a href="#GlobalValue"><tt>GlobalValue</tt></a>,
3470 <a href="#Constant"><tt>Constant</tt></a>,
3471 <a href="#User"><tt>User</tt></a>,
3472 <a href="#Value"><tt>Value</tt></a></p>
3474 <p>Global variables are represented with the (suprise suprise)
3475 <tt>GlobalVariable</tt> class. Like functions, <tt>GlobalVariable</tt>s are also
3476 subclasses of <a href="#GlobalValue"><tt>GlobalValue</tt></a>, and as such are
3477 always referenced by their address (global values must live in memory, so their
3478 "name" refers to their constant address). See
3479 <a href="#GlobalValue"><tt>GlobalValue</tt></a> for more on this. Global
3480 variables may have an initial value (which must be a
3481 <a href="#Constant"><tt>Constant</tt></a>), and if they have an initializer,
3482 they may be marked as "constant" themselves (indicating that their contents
3483 never change at runtime).</p>
3486 <!-- _______________________________________________________________________ -->
3487 <div class="doc_subsubsection">
3488 <a name="m_GlobalVariable">Important Public Members of the
3489 <tt>GlobalVariable</tt> class</a>
3492 <div class="doc_text">
3495 <li><tt>GlobalVariable(const </tt><tt><a href="#Type">Type</a> *Ty, bool
3496 isConstant, LinkageTypes& Linkage, <a href="#Constant">Constant</a>
3497 *Initializer = 0, const std::string &Name = "", Module* Parent = 0)</tt>
3499 <p>Create a new global variable of the specified type. If
3500 <tt>isConstant</tt> is true then the global variable will be marked as
3501 unchanging for the program. The Linkage parameter specifies the type of
3502 linkage (internal, external, weak, linkonce, appending) for the variable.
3503 If the linkage is InternalLinkage, WeakAnyLinkage, WeakODRLinkage,
3504 LinkOnceAnyLinkage or LinkOnceODRLinkage, then the resultant
3505 global variable will have internal linkage. AppendingLinkage concatenates
3506 together all instances (in different translation units) of the variable
3507 into a single variable but is only applicable to arrays. See
3508 the <a href="LangRef.html#modulestructure">LLVM Language Reference</a> for
3509 further details on linkage types. Optionally an initializer, a name, and the
3510 module to put the variable into may be specified for the global variable as
3513 <li><tt>bool isConstant() const</tt>
3515 <p>Returns true if this is a global variable that is known not to
3516 be modified at runtime.</p></li>
3518 <li><tt>bool hasInitializer()</tt>
3520 <p>Returns true if this <tt>GlobalVariable</tt> has an intializer.</p></li>
3522 <li><tt><a href="#Constant">Constant</a> *getInitializer()</tt>
3524 <p>Returns the intial value for a <tt>GlobalVariable</tt>. It is not legal
3525 to call this method if there is no initializer.</p></li>
3531 <!-- ======================================================================= -->
3532 <div class="doc_subsection">
3533 <a name="BasicBlock">The <tt>BasicBlock</tt> class</a>
3536 <div class="doc_text">
3539 href="/doxygen/BasicBlock_8h-source.html">llvm/BasicBlock.h</a>"</tt><br>
3540 doxygen info: <a href="/doxygen/structllvm_1_1BasicBlock.html">BasicBlock
3542 Superclass: <a href="#Value"><tt>Value</tt></a></p>
3544 <p>This class represents a single entry multiple exit section of the code,
3545 commonly known as a basic block by the compiler community. The
3546 <tt>BasicBlock</tt> class maintains a list of <a
3547 href="#Instruction"><tt>Instruction</tt></a>s, which form the body of the block.
3548 Matching the language definition, the last element of this list of instructions
3549 is always a terminator instruction (a subclass of the <a
3550 href="#TerminatorInst"><tt>TerminatorInst</tt></a> class).</p>
3552 <p>In addition to tracking the list of instructions that make up the block, the
3553 <tt>BasicBlock</tt> class also keeps track of the <a
3554 href="#Function"><tt>Function</tt></a> that it is embedded into.</p>
3556 <p>Note that <tt>BasicBlock</tt>s themselves are <a
3557 href="#Value"><tt>Value</tt></a>s, because they are referenced by instructions
3558 like branches and can go in the switch tables. <tt>BasicBlock</tt>s have type
3563 <!-- _______________________________________________________________________ -->
3564 <div class="doc_subsubsection">
3565 <a name="m_BasicBlock">Important Public Members of the <tt>BasicBlock</tt>
3569 <div class="doc_text">
3572 <li><tt>BasicBlock(const std::string &Name = "", </tt><tt><a
3573 href="#Function">Function</a> *Parent = 0)</tt>
3575 <p>The <tt>BasicBlock</tt> constructor is used to create new basic blocks for
3576 insertion into a function. The constructor optionally takes a name for the new
3577 block, and a <a href="#Function"><tt>Function</tt></a> to insert it into. If
3578 the <tt>Parent</tt> parameter is specified, the new <tt>BasicBlock</tt> is
3579 automatically inserted at the end of the specified <a
3580 href="#Function"><tt>Function</tt></a>, if not specified, the BasicBlock must be
3581 manually inserted into the <a href="#Function"><tt>Function</tt></a>.</p></li>
3583 <li><tt>BasicBlock::iterator</tt> - Typedef for instruction list iterator<br>
3584 <tt>BasicBlock::const_iterator</tt> - Typedef for const_iterator.<br>
3585 <tt>begin()</tt>, <tt>end()</tt>, <tt>front()</tt>, <tt>back()</tt>,
3586 <tt>size()</tt>, <tt>empty()</tt>
3587 STL-style functions for accessing the instruction list.
3589 <p>These methods and typedefs are forwarding functions that have the same
3590 semantics as the standard library methods of the same names. These methods
3591 expose the underlying instruction list of a basic block in a way that is easy to
3592 manipulate. To get the full complement of container operations (including
3593 operations to update the list), you must use the <tt>getInstList()</tt>
3596 <li><tt>BasicBlock::InstListType &getInstList()</tt>
3598 <p>This method is used to get access to the underlying container that actually
3599 holds the Instructions. This method must be used when there isn't a forwarding
3600 function in the <tt>BasicBlock</tt> class for the operation that you would like
3601 to perform. Because there are no forwarding functions for "updating"
3602 operations, you need to use this if you want to update the contents of a
3603 <tt>BasicBlock</tt>.</p></li>
3605 <li><tt><a href="#Function">Function</a> *getParent()</tt>
3607 <p> Returns a pointer to <a href="#Function"><tt>Function</tt></a> the block is
3608 embedded into, or a null pointer if it is homeless.</p></li>
3610 <li><tt><a href="#TerminatorInst">TerminatorInst</a> *getTerminator()</tt>
3612 <p> Returns a pointer to the terminator instruction that appears at the end of
3613 the <tt>BasicBlock</tt>. If there is no terminator instruction, or if the last
3614 instruction in the block is not a terminator, then a null pointer is
3622 <!-- ======================================================================= -->
3623 <div class="doc_subsection">
3624 <a name="Argument">The <tt>Argument</tt> class</a>
3627 <div class="doc_text">
3629 <p>This subclass of Value defines the interface for incoming formal
3630 arguments to a function. A Function maintains a list of its formal
3631 arguments. An argument has a pointer to the parent Function.</p>
3635 <!-- *********************************************************************** -->
3638 <a href="http://jigsaw.w3.org/css-validator/check/referer"><img
3639 src="http://jigsaw.w3.org/css-validator/images/vcss-blue" alt="Valid CSS"></a>
3640 <a href="http://validator.w3.org/check/referer"><img
3641 src="http://www.w3.org/Icons/valid-html401" alt="Valid HTML 4.01 Strict"></a>
3643 <a href="mailto:dhurjati@cs.uiuc.edu">Dinakar Dhurjati</a> and
3644 <a href="mailto:sabre@nondot.org">Chris Lattner</a><br>
3645 <a href="http://llvm.org">The LLVM Compiler Infrastructure</a><br>
3646 Last modified: $Date$