1 <!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.01//EN"
2 "http://www.w3.org/TR/html4/strict.dtd">
5 <meta http-equiv="Content-type" content="text/html;charset=UTF-8">
6 <title>LLVM Programmer's Manual</title>
7 <link rel="stylesheet" href="llvm.css" type="text/css">
11 <div class="doc_title">
12 LLVM Programmer's Manual
16 <li><a href="#introduction">Introduction</a></li>
17 <li><a href="#general">General Information</a>
19 <li><a href="#stl">The C++ Standard Template Library</a></li>
21 <li>The <tt>-time-passes</tt> option</li>
22 <li>How to use the LLVM Makefile system</li>
23 <li>How to write a regression test</li>
28 <li><a href="#apis">Important and useful LLVM APIs</a>
30 <li><a href="#isa">The <tt>isa<></tt>, <tt>cast<></tt>
31 and <tt>dyn_cast<></tt> templates</a> </li>
32 <li><a href="#DEBUG">The <tt>DEBUG()</tt> macro and <tt>-debug</tt>
35 <li><a href="#DEBUG_TYPE">Fine grained debug info with <tt>DEBUG_TYPE</tt>
36 and the <tt>-debug-only</tt> option</a> </li>
39 <li><a href="#Statistic">The <tt>Statistic</tt> class & <tt>-stats</tt>
42 <li>The <tt>InstVisitor</tt> template
43 <li>The general graph API
45 <li><a href="#ViewGraph">Viewing graphs while debugging code</a></li>
48 <li><a href="#datastructure">Picking the Right Data Structure for a Task</a>
50 <li><a href="#ds_sequential">Sequential Containers (std::vector, std::list, etc)</a>
52 <li><a href="#dss_fixedarrays">Fixed Size Arrays</a></li>
53 <li><a href="#dss_heaparrays">Heap Allocated Arrays</a></li>
54 <li><a href="#dss_smallvector">"llvm/ADT/SmallVector.h"</a></li>
55 <li><a href="#dss_vector"><vector></a></li>
56 <li><a href="#dss_deque"><deque></a></li>
57 <li><a href="#dss_list"><list></a></li>
58 <li><a href="#dss_ilist">llvm/ADT/ilist.h</a></li>
59 <li><a href="#dss_other">Other Sequential Container Options</a></li>
61 <li><a href="#ds_set">Set-Like Containers (std::set, SmallSet, SetVector, etc)</a>
63 <li><a href="#dss_sortedvectorset">A sorted 'vector'</a></li>
64 <li><a href="#dss_smallset">"llvm/ADT/SmallSet.h"</a></li>
65 <li><a href="#dss_smallptrset">"llvm/ADT/SmallPtrSet.h"</a></li>
66 <li><a href="#dss_denseset">"llvm/ADT/DenseSet.h"</a></li>
67 <li><a href="#dss_FoldingSet">"llvm/ADT/FoldingSet.h"</a></li>
68 <li><a href="#dss_set"><set></a></li>
69 <li><a href="#dss_setvector">"llvm/ADT/SetVector.h"</a></li>
70 <li><a href="#dss_uniquevector">"llvm/ADT/UniqueVector.h"</a></li>
71 <li><a href="#dss_otherset">Other Set-Like ContainerOptions</a></li>
73 <li><a href="#ds_map">Map-Like Containers (std::map, DenseMap, etc)</a>
75 <li><a href="#dss_sortedvectormap">A sorted 'vector'</a></li>
76 <li><a href="#dss_stringmap">"llvm/ADT/StringMap.h"</a></li>
77 <li><a href="#dss_indexedmap">"llvm/ADT/IndexedMap.h"</a></li>
78 <li><a href="#dss_densemap">"llvm/ADT/DenseMap.h"</a></li>
79 <li><a href="#dss_map"><map></a></li>
80 <li><a href="#dss_othermap">Other Map-Like Container Options</a></li>
82 <li><a href="#ds_bit">BitVector-like containers</a>
84 <li><a href="#dss_bitvector">A dense bitvector</a></li>
85 <li><a href="#dss_sparsebitvector">A sparse bitvector</a></li>
89 <li><a href="#common">Helpful Hints for Common Operations</a>
91 <li><a href="#inspection">Basic Inspection and Traversal Routines</a>
93 <li><a href="#iterate_function">Iterating over the <tt>BasicBlock</tt>s
94 in a <tt>Function</tt></a> </li>
95 <li><a href="#iterate_basicblock">Iterating over the <tt>Instruction</tt>s
96 in a <tt>BasicBlock</tt></a> </li>
97 <li><a href="#iterate_institer">Iterating over the <tt>Instruction</tt>s
98 in a <tt>Function</tt></a> </li>
99 <li><a href="#iterate_convert">Turning an iterator into a
100 class pointer</a> </li>
101 <li><a href="#iterate_complex">Finding call sites: a more
102 complex example</a> </li>
103 <li><a href="#calls_and_invokes">Treating calls and invokes
104 the same way</a> </li>
105 <li><a href="#iterate_chains">Iterating over def-use &
106 use-def chains</a> </li>
107 <li><a href="#iterate_preds">Iterating over predecessors &
108 successors of blocks</a></li>
111 <li><a href="#simplechanges">Making simple changes</a>
113 <li><a href="#schanges_creating">Creating and inserting new
114 <tt>Instruction</tt>s</a> </li>
115 <li><a href="#schanges_deleting">Deleting <tt>Instruction</tt>s</a> </li>
116 <li><a href="#schanges_replacing">Replacing an <tt>Instruction</tt>
117 with another <tt>Value</tt></a> </li>
118 <li><a href="#schanges_deletingGV">Deleting <tt>GlobalVariable</tt>s</a> </li>
121 <li><a href="#create_types">How to Create Types</a></li>
123 <li>Working with the Control Flow Graph
125 <li>Accessing predecessors and successors of a <tt>BasicBlock</tt>
133 <li><a href="#threading">Threads and LLVM</a>
135 <li><a href="#startmultithreaded">Entering threaded mode with <tt>llvm_start_multithreaded()</tt></a></li>
136 <li><a href="#shutdown">Ending execution with <tt>llvm_shutdown()</tt></a></li>
137 <li><a href="#managedstatic">Lazy initialization with <tt>ManagedStatic</tt></a></li>
141 <li><a href="#advanced">Advanced Topics</a>
143 <li><a href="#TypeResolve">LLVM Type Resolution</a>
145 <li><a href="#BuildRecType">Basic Recursive Type Construction</a></li>
146 <li><a href="#refineAbstractTypeTo">The <tt>refineAbstractTypeTo</tt> method</a></li>
147 <li><a href="#PATypeHolder">The PATypeHolder Class</a></li>
148 <li><a href="#AbstractTypeUser">The AbstractTypeUser Class</a></li>
151 <li><a href="#SymbolTable">The <tt>ValueSymbolTable</tt> and <tt>TypeSymbolTable</tt> classes</a></li>
152 <li><a href="#UserLayout">The <tt>User</tt> and owned <tt>Use</tt> classes' memory layout</a></li>
155 <li><a href="#coreclasses">The Core LLVM Class Hierarchy Reference</a>
157 <li><a href="#Type">The <tt>Type</tt> class</a> </li>
158 <li><a href="#Module">The <tt>Module</tt> class</a></li>
159 <li><a href="#Value">The <tt>Value</tt> class</a>
161 <li><a href="#User">The <tt>User</tt> class</a>
163 <li><a href="#Instruction">The <tt>Instruction</tt> class</a></li>
164 <li><a href="#Constant">The <tt>Constant</tt> class</a>
166 <li><a href="#GlobalValue">The <tt>GlobalValue</tt> class</a>
168 <li><a href="#Function">The <tt>Function</tt> class</a></li>
169 <li><a href="#GlobalVariable">The <tt>GlobalVariable</tt> class</a></li>
176 <li><a href="#BasicBlock">The <tt>BasicBlock</tt> class</a></li>
177 <li><a href="#Argument">The <tt>Argument</tt> class</a></li>
184 <div class="doc_author">
185 <p>Written by <a href="mailto:sabre@nondot.org">Chris Lattner</a>,
186 <a href="mailto:dhurjati@cs.uiuc.edu">Dinakar Dhurjati</a>,
187 <a href="mailto:ggreif@gmail.com">Gabor Greif</a>,
188 <a href="mailto:jstanley@cs.uiuc.edu">Joel Stanley</a>,
189 <a href="mailto:rspencer@x10sys.com">Reid Spencer</a> and
190 <a href="mailto:owen@apple.com">Owen Anderson</a></p>
193 <!-- *********************************************************************** -->
194 <div class="doc_section">
195 <a name="introduction">Introduction </a>
197 <!-- *********************************************************************** -->
199 <div class="doc_text">
201 <p>This document is meant to highlight some of the important classes and
202 interfaces available in the LLVM source-base. This manual is not
203 intended to explain what LLVM is, how it works, and what LLVM code looks
204 like. It assumes that you know the basics of LLVM and are interested
205 in writing transformations or otherwise analyzing or manipulating the
208 <p>This document should get you oriented so that you can find your
209 way in the continuously growing source code that makes up the LLVM
210 infrastructure. Note that this manual is not intended to serve as a
211 replacement for reading the source code, so if you think there should be
212 a method in one of these classes to do something, but it's not listed,
213 check the source. Links to the <a href="/doxygen/">doxygen</a> sources
214 are provided to make this as easy as possible.</p>
216 <p>The first section of this document describes general information that is
217 useful to know when working in the LLVM infrastructure, and the second describes
218 the Core LLVM classes. In the future this manual will be extended with
219 information describing how to use extension libraries, such as dominator
220 information, CFG traversal routines, and useful utilities like the <tt><a
221 href="/doxygen/InstVisitor_8h-source.html">InstVisitor</a></tt> template.</p>
225 <!-- *********************************************************************** -->
226 <div class="doc_section">
227 <a name="general">General Information</a>
229 <!-- *********************************************************************** -->
231 <div class="doc_text">
233 <p>This section contains general information that is useful if you are working
234 in the LLVM source-base, but that isn't specific to any particular API.</p>
238 <!-- ======================================================================= -->
239 <div class="doc_subsection">
240 <a name="stl">The C++ Standard Template Library</a>
243 <div class="doc_text">
245 <p>LLVM makes heavy use of the C++ Standard Template Library (STL),
246 perhaps much more than you are used to, or have seen before. Because of
247 this, you might want to do a little background reading in the
248 techniques used and capabilities of the library. There are many good
249 pages that discuss the STL, and several books on the subject that you
250 can get, so it will not be discussed in this document.</p>
252 <p>Here are some useful links:</p>
256 <li><a href="http://www.dinkumware.com/refxcpp.html">Dinkumware C++ Library
257 reference</a> - an excellent reference for the STL and other parts of the
258 standard C++ library.</li>
260 <li><a href="http://www.tempest-sw.com/cpp/">C++ In a Nutshell</a> - This is an
261 O'Reilly book in the making. It has a decent Standard Library
262 Reference that rivals Dinkumware's, and is unfortunately no longer free since the
263 book has been published.</li>
265 <li><a href="http://www.parashift.com/c++-faq-lite/">C++ Frequently Asked
268 <li><a href="http://www.sgi.com/tech/stl/">SGI's STL Programmer's Guide</a> -
270 href="http://www.sgi.com/tech/stl/stl_introduction.html">Introduction to the
273 <li><a href="http://www.research.att.com/%7Ebs/C++.html">Bjarne Stroustrup's C++
276 <li><a href="http://64.78.49.204/">
277 Bruce Eckel's Thinking in C++, 2nd ed. Volume 2 Revision 4.0 (even better, get
282 <p>You are also encouraged to take a look at the <a
283 href="CodingStandards.html">LLVM Coding Standards</a> guide which focuses on how
284 to write maintainable code more than where to put your curly braces.</p>
288 <!-- ======================================================================= -->
289 <div class="doc_subsection">
290 <a name="stl">Other useful references</a>
293 <div class="doc_text">
296 <li><a href="http://www.psc.edu/%7Esemke/cvs_branches.html">CVS
297 Branch and Tag Primer</a></li>
298 <li><a href="http://www.fortran-2000.com/ArnaudRecipes/sharedlib.html">Using
299 static and shared libraries across platforms</a></li>
304 <!-- *********************************************************************** -->
305 <div class="doc_section">
306 <a name="apis">Important and useful LLVM APIs</a>
308 <!-- *********************************************************************** -->
310 <div class="doc_text">
312 <p>Here we highlight some LLVM APIs that are generally useful and good to
313 know about when writing transformations.</p>
317 <!-- ======================================================================= -->
318 <div class="doc_subsection">
319 <a name="isa">The <tt>isa<></tt>, <tt>cast<></tt> and
320 <tt>dyn_cast<></tt> templates</a>
323 <div class="doc_text">
325 <p>The LLVM source-base makes extensive use of a custom form of RTTI.
326 These templates have many similarities to the C++ <tt>dynamic_cast<></tt>
327 operator, but they don't have some drawbacks (primarily stemming from
328 the fact that <tt>dynamic_cast<></tt> only works on classes that
329 have a v-table). Because they are used so often, you must know what they
330 do and how they work. All of these templates are defined in the <a
331 href="/doxygen/Casting_8h-source.html"><tt>llvm/Support/Casting.h</tt></a>
332 file (note that you very rarely have to include this file directly).</p>
335 <dt><tt>isa<></tt>: </dt>
337 <dd><p>The <tt>isa<></tt> operator works exactly like the Java
338 "<tt>instanceof</tt>" operator. It returns true or false depending on whether
339 a reference or pointer points to an instance of the specified class. This can
340 be very useful for constraint checking of various sorts (example below).</p>
343 <dt><tt>cast<></tt>: </dt>
345 <dd><p>The <tt>cast<></tt> operator is a "checked cast" operation. It
346 converts a pointer or reference from a base class to a derived class, causing
347 an assertion failure if it is not really an instance of the right type. This
348 should be used in cases where you have some information that makes you believe
349 that something is of the right type. An example of the <tt>isa<></tt>
350 and <tt>cast<></tt> template is:</p>
352 <div class="doc_code">
354 static bool isLoopInvariant(const <a href="#Value">Value</a> *V, const Loop *L) {
355 if (isa<<a href="#Constant">Constant</a>>(V) || isa<<a href="#Argument">Argument</a>>(V) || isa<<a href="#GlobalValue">GlobalValue</a>>(V))
358 // <i>Otherwise, it must be an instruction...</i>
359 return !L->contains(cast<<a href="#Instruction">Instruction</a>>(V)->getParent());
364 <p>Note that you should <b>not</b> use an <tt>isa<></tt> test followed
365 by a <tt>cast<></tt>, for that use the <tt>dyn_cast<></tt>
370 <dt><tt>dyn_cast<></tt>:</dt>
372 <dd><p>The <tt>dyn_cast<></tt> operator is a "checking cast" operation.
373 It checks to see if the operand is of the specified type, and if so, returns a
374 pointer to it (this operator does not work with references). If the operand is
375 not of the correct type, a null pointer is returned. Thus, this works very
376 much like the <tt>dynamic_cast<></tt> operator in C++, and should be
377 used in the same circumstances. Typically, the <tt>dyn_cast<></tt>
378 operator is used in an <tt>if</tt> statement or some other flow control
379 statement like this:</p>
381 <div class="doc_code">
383 if (<a href="#AllocationInst">AllocationInst</a> *AI = dyn_cast<<a href="#AllocationInst">AllocationInst</a>>(Val)) {
389 <p>This form of the <tt>if</tt> statement effectively combines together a call
390 to <tt>isa<></tt> and a call to <tt>cast<></tt> into one
391 statement, which is very convenient.</p>
393 <p>Note that the <tt>dyn_cast<></tt> operator, like C++'s
394 <tt>dynamic_cast<></tt> or Java's <tt>instanceof</tt> operator, can be
395 abused. In particular, you should not use big chained <tt>if/then/else</tt>
396 blocks to check for lots of different variants of classes. If you find
397 yourself wanting to do this, it is much cleaner and more efficient to use the
398 <tt>InstVisitor</tt> class to dispatch over the instruction type directly.</p>
402 <dt><tt>cast_or_null<></tt>: </dt>
404 <dd><p>The <tt>cast_or_null<></tt> operator works just like the
405 <tt>cast<></tt> operator, except that it allows for a null pointer as an
406 argument (which it then propagates). This can sometimes be useful, allowing
407 you to combine several null checks into one.</p></dd>
409 <dt><tt>dyn_cast_or_null<></tt>: </dt>
411 <dd><p>The <tt>dyn_cast_or_null<></tt> operator works just like the
412 <tt>dyn_cast<></tt> operator, except that it allows for a null pointer
413 as an argument (which it then propagates). This can sometimes be useful,
414 allowing you to combine several null checks into one.</p></dd>
418 <p>These five templates can be used with any classes, whether they have a
419 v-table or not. To add support for these templates, you simply need to add
420 <tt>classof</tt> static methods to the class you are interested casting
421 to. Describing this is currently outside the scope of this document, but there
422 are lots of examples in the LLVM source base.</p>
426 <!-- ======================================================================= -->
427 <div class="doc_subsection">
428 <a name="DEBUG">The <tt>DEBUG()</tt> macro and <tt>-debug</tt> option</a>
431 <div class="doc_text">
433 <p>Often when working on your pass you will put a bunch of debugging printouts
434 and other code into your pass. After you get it working, you want to remove
435 it, but you may need it again in the future (to work out new bugs that you run
438 <p> Naturally, because of this, you don't want to delete the debug printouts,
439 but you don't want them to always be noisy. A standard compromise is to comment
440 them out, allowing you to enable them if you need them in the future.</p>
442 <p>The "<tt><a href="/doxygen/Debug_8h-source.html">llvm/Support/Debug.h</a></tt>"
443 file provides a macro named <tt>DEBUG()</tt> that is a much nicer solution to
444 this problem. Basically, you can put arbitrary code into the argument of the
445 <tt>DEBUG</tt> macro, and it is only executed if '<tt>opt</tt>' (or any other
446 tool) is run with the '<tt>-debug</tt>' command line argument:</p>
448 <div class="doc_code">
450 DOUT << "I am here!\n";
454 <p>Then you can run your pass like this:</p>
456 <div class="doc_code">
458 $ opt < a.bc > /dev/null -mypass
459 <i><no output></i>
460 $ opt < a.bc > /dev/null -mypass -debug
465 <p>Using the <tt>DEBUG()</tt> macro instead of a home-brewed solution allows you
466 to not have to create "yet another" command line option for the debug output for
467 your pass. Note that <tt>DEBUG()</tt> macros are disabled for optimized builds,
468 so they do not cause a performance impact at all (for the same reason, they
469 should also not contain side-effects!).</p>
471 <p>One additional nice thing about the <tt>DEBUG()</tt> macro is that you can
472 enable or disable it directly in gdb. Just use "<tt>set DebugFlag=0</tt>" or
473 "<tt>set DebugFlag=1</tt>" from the gdb if the program is running. If the
474 program hasn't been started yet, you can always just run it with
479 <!-- _______________________________________________________________________ -->
480 <div class="doc_subsubsection">
481 <a name="DEBUG_TYPE">Fine grained debug info with <tt>DEBUG_TYPE</tt> and
482 the <tt>-debug-only</tt> option</a>
485 <div class="doc_text">
487 <p>Sometimes you may find yourself in a situation where enabling <tt>-debug</tt>
488 just turns on <b>too much</b> information (such as when working on the code
489 generator). If you want to enable debug information with more fine-grained
490 control, you define the <tt>DEBUG_TYPE</tt> macro and the <tt>-debug</tt> only
491 option as follows:</p>
493 <div class="doc_code">
495 DOUT << "No debug type\n";
497 #define DEBUG_TYPE "foo"
498 DOUT << "'foo' debug type\n";
500 #define DEBUG_TYPE "bar"
501 DOUT << "'bar' debug type\n";
503 #define DEBUG_TYPE ""
504 DOUT << "No debug type (2)\n";
508 <p>Then you can run your pass like this:</p>
510 <div class="doc_code">
512 $ opt < a.bc > /dev/null -mypass
513 <i><no output></i>
514 $ opt < a.bc > /dev/null -mypass -debug
519 $ opt < a.bc > /dev/null -mypass -debug-only=foo
521 $ opt < a.bc > /dev/null -mypass -debug-only=bar
526 <p>Of course, in practice, you should only set <tt>DEBUG_TYPE</tt> at the top of
527 a file, to specify the debug type for the entire module (if you do this before
528 you <tt>#include "llvm/Support/Debug.h"</tt>, you don't have to insert the ugly
529 <tt>#undef</tt>'s). Also, you should use names more meaningful than "foo" and
530 "bar", because there is no system in place to ensure that names do not
531 conflict. If two different modules use the same string, they will all be turned
532 on when the name is specified. This allows, for example, all debug information
533 for instruction scheduling to be enabled with <tt>-debug-type=InstrSched</tt>,
534 even if the source lives in multiple files.</p>
538 <!-- ======================================================================= -->
539 <div class="doc_subsection">
540 <a name="Statistic">The <tt>Statistic</tt> class & <tt>-stats</tt>
544 <div class="doc_text">
547 href="/doxygen/Statistic_8h-source.html">llvm/ADT/Statistic.h</a></tt>" file
548 provides a class named <tt>Statistic</tt> that is used as a unified way to
549 keep track of what the LLVM compiler is doing and how effective various
550 optimizations are. It is useful to see what optimizations are contributing to
551 making a particular program run faster.</p>
553 <p>Often you may run your pass on some big program, and you're interested to see
554 how many times it makes a certain transformation. Although you can do this with
555 hand inspection, or some ad-hoc method, this is a real pain and not very useful
556 for big programs. Using the <tt>Statistic</tt> class makes it very easy to
557 keep track of this information, and the calculated information is presented in a
558 uniform manner with the rest of the passes being executed.</p>
560 <p>There are many examples of <tt>Statistic</tt> uses, but the basics of using
561 it are as follows:</p>
564 <li><p>Define your statistic like this:</p>
566 <div class="doc_code">
568 #define <a href="#DEBUG_TYPE">DEBUG_TYPE</a> "mypassname" <i>// This goes before any #includes.</i>
569 STATISTIC(NumXForms, "The # of times I did stuff");
573 <p>The <tt>STATISTIC</tt> macro defines a static variable, whose name is
574 specified by the first argument. The pass name is taken from the DEBUG_TYPE
575 macro, and the description is taken from the second argument. The variable
576 defined ("NumXForms" in this case) acts like an unsigned integer.</p></li>
578 <li><p>Whenever you make a transformation, bump the counter:</p>
580 <div class="doc_code">
582 ++NumXForms; // <i>I did stuff!</i>
589 <p>That's all you have to do. To get '<tt>opt</tt>' to print out the
590 statistics gathered, use the '<tt>-stats</tt>' option:</p>
592 <div class="doc_code">
594 $ opt -stats -mypassname < program.bc > /dev/null
595 <i>... statistics output ...</i>
599 <p> When running <tt>opt</tt> on a C file from the SPEC benchmark
600 suite, it gives a report that looks like this:</p>
602 <div class="doc_code">
604 7646 bitcodewriter - Number of normal instructions
605 725 bitcodewriter - Number of oversized instructions
606 129996 bitcodewriter - Number of bitcode bytes written
607 2817 raise - Number of insts DCEd or constprop'd
608 3213 raise - Number of cast-of-self removed
609 5046 raise - Number of expression trees converted
610 75 raise - Number of other getelementptr's formed
611 138 raise - Number of load/store peepholes
612 42 deadtypeelim - Number of unused typenames removed from symtab
613 392 funcresolve - Number of varargs functions resolved
614 27 globaldce - Number of global variables removed
615 2 adce - Number of basic blocks removed
616 134 cee - Number of branches revectored
617 49 cee - Number of setcc instruction eliminated
618 532 gcse - Number of loads removed
619 2919 gcse - Number of instructions removed
620 86 indvars - Number of canonical indvars added
621 87 indvars - Number of aux indvars removed
622 25 instcombine - Number of dead inst eliminate
623 434 instcombine - Number of insts combined
624 248 licm - Number of load insts hoisted
625 1298 licm - Number of insts hoisted to a loop pre-header
626 3 licm - Number of insts hoisted to multiple loop preds (bad, no loop pre-header)
627 75 mem2reg - Number of alloca's promoted
628 1444 cfgsimplify - Number of blocks simplified
632 <p>Obviously, with so many optimizations, having a unified framework for this
633 stuff is very nice. Making your pass fit well into the framework makes it more
634 maintainable and useful.</p>
638 <!-- ======================================================================= -->
639 <div class="doc_subsection">
640 <a name="ViewGraph">Viewing graphs while debugging code</a>
643 <div class="doc_text">
645 <p>Several of the important data structures in LLVM are graphs: for example
646 CFGs made out of LLVM <a href="#BasicBlock">BasicBlock</a>s, CFGs made out of
647 LLVM <a href="CodeGenerator.html#machinebasicblock">MachineBasicBlock</a>s, and
648 <a href="CodeGenerator.html#selectiondag_intro">Instruction Selection
649 DAGs</a>. In many cases, while debugging various parts of the compiler, it is
650 nice to instantly visualize these graphs.</p>
652 <p>LLVM provides several callbacks that are available in a debug build to do
653 exactly that. If you call the <tt>Function::viewCFG()</tt> method, for example,
654 the current LLVM tool will pop up a window containing the CFG for the function
655 where each basic block is a node in the graph, and each node contains the
656 instructions in the block. Similarly, there also exists
657 <tt>Function::viewCFGOnly()</tt> (does not include the instructions), the
658 <tt>MachineFunction::viewCFG()</tt> and <tt>MachineFunction::viewCFGOnly()</tt>,
659 and the <tt>SelectionDAG::viewGraph()</tt> methods. Within GDB, for example,
660 you can usually use something like <tt>call DAG.viewGraph()</tt> to pop
661 up a window. Alternatively, you can sprinkle calls to these functions in your
662 code in places you want to debug.</p>
664 <p>Getting this to work requires a small amount of configuration. On Unix
665 systems with X11, install the <a href="http://www.graphviz.org">graphviz</a>
666 toolkit, and make sure 'dot' and 'gv' are in your path. If you are running on
667 Mac OS/X, download and install the Mac OS/X <a
668 href="http://www.pixelglow.com/graphviz/">Graphviz program</a>, and add
669 <tt>/Applications/Graphviz.app/Contents/MacOS/</tt> (or wherever you install
670 it) to your path. Once in your system and path are set up, rerun the LLVM
671 configure script and rebuild LLVM to enable this functionality.</p>
673 <p><tt>SelectionDAG</tt> has been extended to make it easier to locate
674 <i>interesting</i> nodes in large complex graphs. From gdb, if you
675 <tt>call DAG.setGraphColor(<i>node</i>, "<i>color</i>")</tt>, then the
676 next <tt>call DAG.viewGraph()</tt> would highlight the node in the
677 specified color (choices of colors can be found at <a
678 href="http://www.graphviz.org/doc/info/colors.html">colors</a>.) More
679 complex node attributes can be provided with <tt>call
680 DAG.setGraphAttrs(<i>node</i>, "<i>attributes</i>")</tt> (choices can be
681 found at <a href="http://www.graphviz.org/doc/info/attrs.html">Graph
682 Attributes</a>.) If you want to restart and clear all the current graph
683 attributes, then you can <tt>call DAG.clearGraphAttrs()</tt>. </p>
687 <!-- *********************************************************************** -->
688 <div class="doc_section">
689 <a name="datastructure">Picking the Right Data Structure for a Task</a>
691 <!-- *********************************************************************** -->
693 <div class="doc_text">
695 <p>LLVM has a plethora of data structures in the <tt>llvm/ADT/</tt> directory,
696 and we commonly use STL data structures. This section describes the trade-offs
697 you should consider when you pick one.</p>
700 The first step is a choose your own adventure: do you want a sequential
701 container, a set-like container, or a map-like container? The most important
702 thing when choosing a container is the algorithmic properties of how you plan to
703 access the container. Based on that, you should use:</p>
706 <li>a <a href="#ds_map">map-like</a> container if you need efficient look-up
707 of an value based on another value. Map-like containers also support
708 efficient queries for containment (whether a key is in the map). Map-like
709 containers generally do not support efficient reverse mapping (values to
710 keys). If you need that, use two maps. Some map-like containers also
711 support efficient iteration through the keys in sorted order. Map-like
712 containers are the most expensive sort, only use them if you need one of
713 these capabilities.</li>
715 <li>a <a href="#ds_set">set-like</a> container if you need to put a bunch of
716 stuff into a container that automatically eliminates duplicates. Some
717 set-like containers support efficient iteration through the elements in
718 sorted order. Set-like containers are more expensive than sequential
722 <li>a <a href="#ds_sequential">sequential</a> container provides
723 the most efficient way to add elements and keeps track of the order they are
724 added to the collection. They permit duplicates and support efficient
725 iteration, but do not support efficient look-up based on a key.
728 <li>a <a href="#ds_bit">bit</a> container provides an efficient way to store and
729 perform set operations on sets of numeric id's, while automatically
730 eliminating duplicates. Bit containers require a maximum of 1 bit for each
731 identifier you want to store.
736 Once the proper category of container is determined, you can fine tune the
737 memory use, constant factors, and cache behaviors of access by intelligently
738 picking a member of the category. Note that constant factors and cache behavior
739 can be a big deal. If you have a vector that usually only contains a few
740 elements (but could contain many), for example, it's much better to use
741 <a href="#dss_smallvector">SmallVector</a> than <a href="#dss_vector">vector</a>
742 . Doing so avoids (relatively) expensive malloc/free calls, which dwarf the
743 cost of adding the elements to the container. </p>
747 <!-- ======================================================================= -->
748 <div class="doc_subsection">
749 <a name="ds_sequential">Sequential Containers (std::vector, std::list, etc)</a>
752 <div class="doc_text">
753 There are a variety of sequential containers available for you, based on your
754 needs. Pick the first in this section that will do what you want.
757 <!-- _______________________________________________________________________ -->
758 <div class="doc_subsubsection">
759 <a name="dss_fixedarrays">Fixed Size Arrays</a>
762 <div class="doc_text">
763 <p>Fixed size arrays are very simple and very fast. They are good if you know
764 exactly how many elements you have, or you have a (low) upper bound on how many
768 <!-- _______________________________________________________________________ -->
769 <div class="doc_subsubsection">
770 <a name="dss_heaparrays">Heap Allocated Arrays</a>
773 <div class="doc_text">
774 <p>Heap allocated arrays (new[] + delete[]) are also simple. They are good if
775 the number of elements is variable, if you know how many elements you will need
776 before the array is allocated, and if the array is usually large (if not,
777 consider a <a href="#dss_smallvector">SmallVector</a>). The cost of a heap
778 allocated array is the cost of the new/delete (aka malloc/free). Also note that
779 if you are allocating an array of a type with a constructor, the constructor and
780 destructors will be run for every element in the array (re-sizable vectors only
781 construct those elements actually used).</p>
784 <!-- _______________________________________________________________________ -->
785 <div class="doc_subsubsection">
786 <a name="dss_smallvector">"llvm/ADT/SmallVector.h"</a>
789 <div class="doc_text">
790 <p><tt>SmallVector<Type, N></tt> is a simple class that looks and smells
791 just like <tt>vector<Type></tt>:
792 it supports efficient iteration, lays out elements in memory order (so you can
793 do pointer arithmetic between elements), supports efficient push_back/pop_back
794 operations, supports efficient random access to its elements, etc.</p>
796 <p>The advantage of SmallVector is that it allocates space for
797 some number of elements (N) <b>in the object itself</b>. Because of this, if
798 the SmallVector is dynamically smaller than N, no malloc is performed. This can
799 be a big win in cases where the malloc/free call is far more expensive than the
800 code that fiddles around with the elements.</p>
802 <p>This is good for vectors that are "usually small" (e.g. the number of
803 predecessors/successors of a block is usually less than 8). On the other hand,
804 this makes the size of the SmallVector itself large, so you don't want to
805 allocate lots of them (doing so will waste a lot of space). As such,
806 SmallVectors are most useful when on the stack.</p>
808 <p>SmallVector also provides a nice portable and efficient replacement for
813 <!-- _______________________________________________________________________ -->
814 <div class="doc_subsubsection">
815 <a name="dss_vector"><vector></a>
818 <div class="doc_text">
820 std::vector is well loved and respected. It is useful when SmallVector isn't:
821 when the size of the vector is often large (thus the small optimization will
822 rarely be a benefit) or if you will be allocating many instances of the vector
823 itself (which would waste space for elements that aren't in the container).
824 vector is also useful when interfacing with code that expects vectors :).
827 <p>One worthwhile note about std::vector: avoid code like this:</p>
829 <div class="doc_code">
832 std::vector<foo> V;
838 <p>Instead, write this as:</p>
840 <div class="doc_code">
842 std::vector<foo> V;
850 <p>Doing so will save (at least) one heap allocation and free per iteration of
855 <!-- _______________________________________________________________________ -->
856 <div class="doc_subsubsection">
857 <a name="dss_deque"><deque></a>
860 <div class="doc_text">
861 <p>std::deque is, in some senses, a generalized version of std::vector. Like
862 std::vector, it provides constant time random access and other similar
863 properties, but it also provides efficient access to the front of the list. It
864 does not guarantee continuity of elements within memory.</p>
866 <p>In exchange for this extra flexibility, std::deque has significantly higher
867 constant factor costs than std::vector. If possible, use std::vector or
868 something cheaper.</p>
871 <!-- _______________________________________________________________________ -->
872 <div class="doc_subsubsection">
873 <a name="dss_list"><list></a>
876 <div class="doc_text">
877 <p>std::list is an extremely inefficient class that is rarely useful.
878 It performs a heap allocation for every element inserted into it, thus having an
879 extremely high constant factor, particularly for small data types. std::list
880 also only supports bidirectional iteration, not random access iteration.</p>
882 <p>In exchange for this high cost, std::list supports efficient access to both
883 ends of the list (like std::deque, but unlike std::vector or SmallVector). In
884 addition, the iterator invalidation characteristics of std::list are stronger
885 than that of a vector class: inserting or removing an element into the list does
886 not invalidate iterator or pointers to other elements in the list.</p>
889 <!-- _______________________________________________________________________ -->
890 <div class="doc_subsubsection">
891 <a name="dss_ilist">llvm/ADT/ilist.h</a>
894 <div class="doc_text">
895 <p><tt>ilist<T></tt> implements an 'intrusive' doubly-linked list. It is
896 intrusive, because it requires the element to store and provide access to the
897 prev/next pointers for the list.</p>
899 <p><tt>ilist</tt> has the same drawbacks as <tt>std::list</tt>, and additionally
900 requires an <tt>ilist_traits</tt> implementation for the element type, but it
901 provides some novel characteristics. In particular, it can efficiently store
902 polymorphic objects, the traits class is informed when an element is inserted or
903 removed from the list, and <tt>ilist</tt>s are guaranteed to support a
904 constant-time splice operation.</p>
906 <p>These properties are exactly what we want for things like
907 <tt>Instruction</tt>s and basic blocks, which is why these are implemented with
910 Related classes of interest are explained in the following subsections:
912 <li><a href="#dss_ilist_traits">ilist_traits</a></li>
913 <li><a href="#dss_iplist">iplist</a></li>
914 <li><a href="#dss_ilist_node">llvm/ADT/ilist_node.h</a></li>
915 <li><a href="#dss_ilist_sentinel">Sentinels</a></li>
919 <!-- _______________________________________________________________________ -->
920 <div class="doc_subsubsection">
921 <a name="dss_ilist_traits">ilist_traits</a>
924 <div class="doc_text">
925 <p><tt>ilist_traits<T></tt> is <tt>ilist<T></tt>'s customization
926 mechanism. <tt>iplist<T></tt> (and consequently <tt>ilist<T></tt>)
927 publicly derive from this traits class.</p>
930 <!-- _______________________________________________________________________ -->
931 <div class="doc_subsubsection">
932 <a name="dss_iplist">iplist</a>
935 <div class="doc_text">
936 <p><tt>iplist<T></tt> is <tt>ilist<T></tt>'s base and as such
937 supports a slightly narrower interface. Notably, inserters from
938 <tt>T&</tt> are absent.</p>
940 <p><tt>ilist_traits<T></tt> is a public base of this class and can be
941 used for a wide variety of customizations.</p>
944 <!-- _______________________________________________________________________ -->
945 <div class="doc_subsubsection">
946 <a name="dss_ilist_node">llvm/ADT/ilist_node.h</a>
949 <div class="doc_text">
950 <p><tt>ilist_node<T></tt> implements a the forward and backward links
951 that are expected by the <tt>ilist<T></tt> (and analogous containers)
952 in the default manner.</p>
954 <p><tt>ilist_node<T></tt>s are meant to be embedded in the node type
955 <tt>T</tt>, usually <tt>T</tt> publicly derives from
956 <tt>ilist_node<T></tt>.</p>
959 <!-- _______________________________________________________________________ -->
960 <div class="doc_subsubsection">
961 <a name="dss_ilist_sentinel">Sentinels</a>
964 <div class="doc_text">
965 <p><tt>ilist</tt>s have another speciality that must be considered. To be a good
966 citizen in the C++ ecosystem, it needs to support the standard container
967 operations, such as <tt>begin</tt> and <tt>end</tt> iterators, etc. Also, the
968 <tt>operator--</tt> must work correctly on the <tt>end</tt> iterator in the
969 case of non-empty <tt>ilist</tt>s.</p>
971 <p>The only sensible solution to this problem is to allocate a so-called
972 <i>sentinel</i> along with the intrusive list, which serves as the <tt>end</tt>
973 iterator, providing the back-link to the last element. However conforming to the
974 C++ convention it is illegal to <tt>operator++</tt> beyond the sentinel and it
975 also must not be dereferenced.</p>
977 <p>These constraints allow for some implementation freedom to the <tt>ilist</tt>
978 how to allocate and store the sentinel. The corresponding policy is dictated
979 by <tt>ilist_traits<T></tt>. By default a <tt>T</tt> gets heap-allocated
980 whenever the need for a sentinel arises.</p>
982 <p>While the default policy is sufficient in most cases, it may break down when
983 <tt>T</tt> does not provide a default constructor. Also, in the case of many
984 instances of <tt>ilist</tt>s, the memory overhead of the associated sentinels
985 is wasted. To alleviate the situation with numerous and voluminous
986 <tt>T</tt>-sentinels, sometimes a trick is employed, leading to <i>ghostly
989 <p>Ghostly sentinels are obtained by specially-crafted <tt>ilist_traits<T></tt>
990 which superpose the sentinel with the <tt>ilist</tt> instance in memory. Pointer
991 arithmetic is used to obtain the sentinel, which is relative to the
992 <tt>ilist</tt>'s <tt>this</tt> pointer. The <tt>ilist</tt> is augmented by an
993 extra pointer, which serves as the back-link of the sentinel. This is the only
994 field in the ghostly sentinel which can be legally accessed.</p>
997 <!-- _______________________________________________________________________ -->
998 <div class="doc_subsubsection">
999 <a name="dss_other">Other Sequential Container options</a>
1002 <div class="doc_text">
1003 <p>Other STL containers are available, such as std::string.</p>
1005 <p>There are also various STL adapter classes such as std::queue,
1006 std::priority_queue, std::stack, etc. These provide simplified access to an
1007 underlying container but don't affect the cost of the container itself.</p>
1012 <!-- ======================================================================= -->
1013 <div class="doc_subsection">
1014 <a name="ds_set">Set-Like Containers (std::set, SmallSet, SetVector, etc)</a>
1017 <div class="doc_text">
1019 <p>Set-like containers are useful when you need to canonicalize multiple values
1020 into a single representation. There are several different choices for how to do
1021 this, providing various trade-offs.</p>
1026 <!-- _______________________________________________________________________ -->
1027 <div class="doc_subsubsection">
1028 <a name="dss_sortedvectorset">A sorted 'vector'</a>
1031 <div class="doc_text">
1033 <p>If you intend to insert a lot of elements, then do a lot of queries, a
1034 great approach is to use a vector (or other sequential container) with
1035 std::sort+std::unique to remove duplicates. This approach works really well if
1036 your usage pattern has these two distinct phases (insert then query), and can be
1037 coupled with a good choice of <a href="#ds_sequential">sequential container</a>.
1041 This combination provides the several nice properties: the result data is
1042 contiguous in memory (good for cache locality), has few allocations, is easy to
1043 address (iterators in the final vector are just indices or pointers), and can be
1044 efficiently queried with a standard binary or radix search.</p>
1048 <!-- _______________________________________________________________________ -->
1049 <div class="doc_subsubsection">
1050 <a name="dss_smallset">"llvm/ADT/SmallSet.h"</a>
1053 <div class="doc_text">
1055 <p>If you have a set-like data structure that is usually small and whose elements
1056 are reasonably small, a <tt>SmallSet<Type, N></tt> is a good choice. This set
1057 has space for N elements in place (thus, if the set is dynamically smaller than
1058 N, no malloc traffic is required) and accesses them with a simple linear search.
1059 When the set grows beyond 'N' elements, it allocates a more expensive representation that
1060 guarantees efficient access (for most types, it falls back to std::set, but for
1061 pointers it uses something far better, <a
1062 href="#dss_smallptrset">SmallPtrSet</a>).</p>
1064 <p>The magic of this class is that it handles small sets extremely efficiently,
1065 but gracefully handles extremely large sets without loss of efficiency. The
1066 drawback is that the interface is quite small: it supports insertion, queries
1067 and erasing, but does not support iteration.</p>
1071 <!-- _______________________________________________________________________ -->
1072 <div class="doc_subsubsection">
1073 <a name="dss_smallptrset">"llvm/ADT/SmallPtrSet.h"</a>
1076 <div class="doc_text">
1078 <p>SmallPtrSet has all the advantages of SmallSet (and a SmallSet of pointers is
1079 transparently implemented with a SmallPtrSet), but also supports iterators. If
1080 more than 'N' insertions are performed, a single quadratically
1081 probed hash table is allocated and grows as needed, providing extremely
1082 efficient access (constant time insertion/deleting/queries with low constant
1083 factors) and is very stingy with malloc traffic.</p>
1085 <p>Note that, unlike std::set, the iterators of SmallPtrSet are invalidated
1086 whenever an insertion occurs. Also, the values visited by the iterators are not
1087 visited in sorted order.</p>
1091 <!-- _______________________________________________________________________ -->
1092 <div class="doc_subsubsection">
1093 <a name="dss_denseset">"llvm/ADT/DenseSet.h"</a>
1096 <div class="doc_text">
1099 DenseSet is a simple quadratically probed hash table. It excels at supporting
1100 small values: it uses a single allocation to hold all of the pairs that
1101 are currently inserted in the set. DenseSet is a great way to unique small
1102 values that are not simple pointers (use <a
1103 href="#dss_smallptrset">SmallPtrSet</a> for pointers). Note that DenseSet has
1104 the same requirements for the value type that <a
1105 href="#dss_densemap">DenseMap</a> has.
1110 <!-- _______________________________________________________________________ -->
1111 <div class="doc_subsubsection">
1112 <a name="dss_FoldingSet">"llvm/ADT/FoldingSet.h"</a>
1115 <div class="doc_text">
1118 FoldingSet is an aggregate class that is really good at uniquing
1119 expensive-to-create or polymorphic objects. It is a combination of a chained
1120 hash table with intrusive links (uniqued objects are required to inherit from
1121 FoldingSetNode) that uses <a href="#dss_smallvector">SmallVector</a> as part of
1124 <p>Consider a case where you want to implement a "getOrCreateFoo" method for
1125 a complex object (for example, a node in the code generator). The client has a
1126 description of *what* it wants to generate (it knows the opcode and all the
1127 operands), but we don't want to 'new' a node, then try inserting it into a set
1128 only to find out it already exists, at which point we would have to delete it
1129 and return the node that already exists.
1132 <p>To support this style of client, FoldingSet perform a query with a
1133 FoldingSetNodeID (which wraps SmallVector) that can be used to describe the
1134 element that we want to query for. The query either returns the element
1135 matching the ID or it returns an opaque ID that indicates where insertion should
1136 take place. Construction of the ID usually does not require heap traffic.</p>
1138 <p>Because FoldingSet uses intrusive links, it can support polymorphic objects
1139 in the set (for example, you can have SDNode instances mixed with LoadSDNodes).
1140 Because the elements are individually allocated, pointers to the elements are
1141 stable: inserting or removing elements does not invalidate any pointers to other
1147 <!-- _______________________________________________________________________ -->
1148 <div class="doc_subsubsection">
1149 <a name="dss_set"><set></a>
1152 <div class="doc_text">
1154 <p><tt>std::set</tt> is a reasonable all-around set class, which is decent at
1155 many things but great at nothing. std::set allocates memory for each element
1156 inserted (thus it is very malloc intensive) and typically stores three pointers
1157 per element in the set (thus adding a large amount of per-element space
1158 overhead). It offers guaranteed log(n) performance, which is not particularly
1159 fast from a complexity standpoint (particularly if the elements of the set are
1160 expensive to compare, like strings), and has extremely high constant factors for
1161 lookup, insertion and removal.</p>
1163 <p>The advantages of std::set are that its iterators are stable (deleting or
1164 inserting an element from the set does not affect iterators or pointers to other
1165 elements) and that iteration over the set is guaranteed to be in sorted order.
1166 If the elements in the set are large, then the relative overhead of the pointers
1167 and malloc traffic is not a big deal, but if the elements of the set are small,
1168 std::set is almost never a good choice.</p>
1172 <!-- _______________________________________________________________________ -->
1173 <div class="doc_subsubsection">
1174 <a name="dss_setvector">"llvm/ADT/SetVector.h"</a>
1177 <div class="doc_text">
1178 <p>LLVM's SetVector<Type> is an adapter class that combines your choice of
1179 a set-like container along with a <a href="#ds_sequential">Sequential
1180 Container</a>. The important property
1181 that this provides is efficient insertion with uniquing (duplicate elements are
1182 ignored) with iteration support. It implements this by inserting elements into
1183 both a set-like container and the sequential container, using the set-like
1184 container for uniquing and the sequential container for iteration.
1187 <p>The difference between SetVector and other sets is that the order of
1188 iteration is guaranteed to match the order of insertion into the SetVector.
1189 This property is really important for things like sets of pointers. Because
1190 pointer values are non-deterministic (e.g. vary across runs of the program on
1191 different machines), iterating over the pointers in the set will
1192 not be in a well-defined order.</p>
1195 The drawback of SetVector is that it requires twice as much space as a normal
1196 set and has the sum of constant factors from the set-like container and the
1197 sequential container that it uses. Use it *only* if you need to iterate over
1198 the elements in a deterministic order. SetVector is also expensive to delete
1199 elements out of (linear time), unless you use it's "pop_back" method, which is
1203 <p>SetVector is an adapter class that defaults to using std::vector and std::set
1204 for the underlying containers, so it is quite expensive. However,
1205 <tt>"llvm/ADT/SetVector.h"</tt> also provides a SmallSetVector class, which
1206 defaults to using a SmallVector and SmallSet of a specified size. If you use
1207 this, and if your sets are dynamically smaller than N, you will save a lot of
1212 <!-- _______________________________________________________________________ -->
1213 <div class="doc_subsubsection">
1214 <a name="dss_uniquevector">"llvm/ADT/UniqueVector.h"</a>
1217 <div class="doc_text">
1220 UniqueVector is similar to <a href="#dss_setvector">SetVector</a>, but it
1221 retains a unique ID for each element inserted into the set. It internally
1222 contains a map and a vector, and it assigns a unique ID for each value inserted
1225 <p>UniqueVector is very expensive: its cost is the sum of the cost of
1226 maintaining both the map and vector, it has high complexity, high constant
1227 factors, and produces a lot of malloc traffic. It should be avoided.</p>
1232 <!-- _______________________________________________________________________ -->
1233 <div class="doc_subsubsection">
1234 <a name="dss_otherset">Other Set-Like Container Options</a>
1237 <div class="doc_text">
1240 The STL provides several other options, such as std::multiset and the various
1241 "hash_set" like containers (whether from C++ TR1 or from the SGI library). We
1242 never use hash_set and unordered_set because they are generally very expensive
1243 (each insertion requires a malloc) and very non-portable.
1246 <p>std::multiset is useful if you're not interested in elimination of
1247 duplicates, but has all the drawbacks of std::set. A sorted vector (where you
1248 don't delete duplicate entries) or some other approach is almost always
1253 <!-- ======================================================================= -->
1254 <div class="doc_subsection">
1255 <a name="ds_map">Map-Like Containers (std::map, DenseMap, etc)</a>
1258 <div class="doc_text">
1259 Map-like containers are useful when you want to associate data to a key. As
1260 usual, there are a lot of different ways to do this. :)
1263 <!-- _______________________________________________________________________ -->
1264 <div class="doc_subsubsection">
1265 <a name="dss_sortedvectormap">A sorted 'vector'</a>
1268 <div class="doc_text">
1271 If your usage pattern follows a strict insert-then-query approach, you can
1272 trivially use the same approach as <a href="#dss_sortedvectorset">sorted vectors
1273 for set-like containers</a>. The only difference is that your query function
1274 (which uses std::lower_bound to get efficient log(n) lookup) should only compare
1275 the key, not both the key and value. This yields the same advantages as sorted
1280 <!-- _______________________________________________________________________ -->
1281 <div class="doc_subsubsection">
1282 <a name="dss_stringmap">"llvm/ADT/StringMap.h"</a>
1285 <div class="doc_text">
1288 Strings are commonly used as keys in maps, and they are difficult to support
1289 efficiently: they are variable length, inefficient to hash and compare when
1290 long, expensive to copy, etc. StringMap is a specialized container designed to
1291 cope with these issues. It supports mapping an arbitrary range of bytes to an
1292 arbitrary other object.</p>
1294 <p>The StringMap implementation uses a quadratically-probed hash table, where
1295 the buckets store a pointer to the heap allocated entries (and some other
1296 stuff). The entries in the map must be heap allocated because the strings are
1297 variable length. The string data (key) and the element object (value) are
1298 stored in the same allocation with the string data immediately after the element
1299 object. This container guarantees the "<tt>(char*)(&Value+1)</tt>" points
1300 to the key string for a value.</p>
1302 <p>The StringMap is very fast for several reasons: quadratic probing is very
1303 cache efficient for lookups, the hash value of strings in buckets is not
1304 recomputed when lookup up an element, StringMap rarely has to touch the
1305 memory for unrelated objects when looking up a value (even when hash collisions
1306 happen), hash table growth does not recompute the hash values for strings
1307 already in the table, and each pair in the map is store in a single allocation
1308 (the string data is stored in the same allocation as the Value of a pair).</p>
1310 <p>StringMap also provides query methods that take byte ranges, so it only ever
1311 copies a string if a value is inserted into the table.</p>
1314 <!-- _______________________________________________________________________ -->
1315 <div class="doc_subsubsection">
1316 <a name="dss_indexedmap">"llvm/ADT/IndexedMap.h"</a>
1319 <div class="doc_text">
1321 IndexedMap is a specialized container for mapping small dense integers (or
1322 values that can be mapped to small dense integers) to some other type. It is
1323 internally implemented as a vector with a mapping function that maps the keys to
1324 the dense integer range.
1328 This is useful for cases like virtual registers in the LLVM code generator: they
1329 have a dense mapping that is offset by a compile-time constant (the first
1330 virtual register ID).</p>
1334 <!-- _______________________________________________________________________ -->
1335 <div class="doc_subsubsection">
1336 <a name="dss_densemap">"llvm/ADT/DenseMap.h"</a>
1339 <div class="doc_text">
1342 DenseMap is a simple quadratically probed hash table. It excels at supporting
1343 small keys and values: it uses a single allocation to hold all of the pairs that
1344 are currently inserted in the map. DenseMap is a great way to map pointers to
1345 pointers, or map other small types to each other.
1349 There are several aspects of DenseMap that you should be aware of, however. The
1350 iterators in a densemap are invalidated whenever an insertion occurs, unlike
1351 map. Also, because DenseMap allocates space for a large number of key/value
1352 pairs (it starts with 64 by default), it will waste a lot of space if your keys
1353 or values are large. Finally, you must implement a partial specialization of
1354 DenseMapInfo for the key that you want, if it isn't already supported. This
1355 is required to tell DenseMap about two special marker values (which can never be
1356 inserted into the map) that it needs internally.</p>
1360 <!-- _______________________________________________________________________ -->
1361 <div class="doc_subsubsection">
1362 <a name="dss_map"><map></a>
1365 <div class="doc_text">
1368 std::map has similar characteristics to <a href="#dss_set">std::set</a>: it uses
1369 a single allocation per pair inserted into the map, it offers log(n) lookup with
1370 an extremely large constant factor, imposes a space penalty of 3 pointers per
1371 pair in the map, etc.</p>
1373 <p>std::map is most useful when your keys or values are very large, if you need
1374 to iterate over the collection in sorted order, or if you need stable iterators
1375 into the map (i.e. they don't get invalidated if an insertion or deletion of
1376 another element takes place).</p>
1380 <!-- _______________________________________________________________________ -->
1381 <div class="doc_subsubsection">
1382 <a name="dss_othermap">Other Map-Like Container Options</a>
1385 <div class="doc_text">
1388 The STL provides several other options, such as std::multimap and the various
1389 "hash_map" like containers (whether from C++ TR1 or from the SGI library). We
1390 never use hash_set and unordered_set because they are generally very expensive
1391 (each insertion requires a malloc) and very non-portable.</p>
1393 <p>std::multimap is useful if you want to map a key to multiple values, but has
1394 all the drawbacks of std::map. A sorted vector or some other approach is almost
1399 <!-- ======================================================================= -->
1400 <div class="doc_subsection">
1401 <a name="ds_bit">Bit storage containers (BitVector, SparseBitVector)</a>
1404 <div class="doc_text">
1405 <p>Unlike the other containers, there are only two bit storage containers, and
1406 choosing when to use each is relatively straightforward.</p>
1408 <p>One additional option is
1409 <tt>std::vector<bool></tt>: we discourage its use for two reasons 1) the
1410 implementation in many common compilers (e.g. commonly available versions of
1411 GCC) is extremely inefficient and 2) the C++ standards committee is likely to
1412 deprecate this container and/or change it significantly somehow. In any case,
1413 please don't use it.</p>
1416 <!-- _______________________________________________________________________ -->
1417 <div class="doc_subsubsection">
1418 <a name="dss_bitvector">BitVector</a>
1421 <div class="doc_text">
1422 <p> The BitVector container provides a fixed size set of bits for manipulation.
1423 It supports individual bit setting/testing, as well as set operations. The set
1424 operations take time O(size of bitvector), but operations are performed one word
1425 at a time, instead of one bit at a time. This makes the BitVector very fast for
1426 set operations compared to other containers. Use the BitVector when you expect
1427 the number of set bits to be high (IE a dense set).
1431 <!-- _______________________________________________________________________ -->
1432 <div class="doc_subsubsection">
1433 <a name="dss_sparsebitvector">SparseBitVector</a>
1436 <div class="doc_text">
1437 <p> The SparseBitVector container is much like BitVector, with one major
1438 difference: Only the bits that are set, are stored. This makes the
1439 SparseBitVector much more space efficient than BitVector when the set is sparse,
1440 as well as making set operations O(number of set bits) instead of O(size of
1441 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
1442 (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).
1446 <!-- *********************************************************************** -->
1447 <div class="doc_section">
1448 <a name="common">Helpful Hints for Common Operations</a>
1450 <!-- *********************************************************************** -->
1452 <div class="doc_text">
1454 <p>This section describes how to perform some very simple transformations of
1455 LLVM code. This is meant to give examples of common idioms used, showing the
1456 practical side of LLVM transformations. <p> Because this is a "how-to" section,
1457 you should also read about the main classes that you will be working with. The
1458 <a href="#coreclasses">Core LLVM Class Hierarchy Reference</a> contains details
1459 and descriptions of the main classes that you should know about.</p>
1463 <!-- NOTE: this section should be heavy on example code -->
1464 <!-- ======================================================================= -->
1465 <div class="doc_subsection">
1466 <a name="inspection">Basic Inspection and Traversal Routines</a>
1469 <div class="doc_text">
1471 <p>The LLVM compiler infrastructure have many different data structures that may
1472 be traversed. Following the example of the C++ standard template library, the
1473 techniques used to traverse these various data structures are all basically the
1474 same. For a enumerable sequence of values, the <tt>XXXbegin()</tt> function (or
1475 method) returns an iterator to the start of the sequence, the <tt>XXXend()</tt>
1476 function returns an iterator pointing to one past the last valid element of the
1477 sequence, and there is some <tt>XXXiterator</tt> data type that is common
1478 between the two operations.</p>
1480 <p>Because the pattern for iteration is common across many different aspects of
1481 the program representation, the standard template library algorithms may be used
1482 on them, and it is easier to remember how to iterate. First we show a few common
1483 examples of the data structures that need to be traversed. Other data
1484 structures are traversed in very similar ways.</p>
1488 <!-- _______________________________________________________________________ -->
1489 <div class="doc_subsubsection">
1490 <a name="iterate_function">Iterating over the </a><a
1491 href="#BasicBlock"><tt>BasicBlock</tt></a>s in a <a
1492 href="#Function"><tt>Function</tt></a>
1495 <div class="doc_text">
1497 <p>It's quite common to have a <tt>Function</tt> instance that you'd like to
1498 transform in some way; in particular, you'd like to manipulate its
1499 <tt>BasicBlock</tt>s. To facilitate this, you'll need to iterate over all of
1500 the <tt>BasicBlock</tt>s that constitute the <tt>Function</tt>. The following is
1501 an example that prints the name of a <tt>BasicBlock</tt> and the number of
1502 <tt>Instruction</tt>s it contains:</p>
1504 <div class="doc_code">
1506 // <i>func is a pointer to a Function instance</i>
1507 for (Function::iterator i = func->begin(), e = func->end(); i != e; ++i)
1508 // <i>Print out the name of the basic block if it has one, and then the</i>
1509 // <i>number of instructions that it contains</i>
1510 llvm::cerr << "Basic block (name=" << i->getName() << ") has "
1511 << i->size() << " instructions.\n";
1515 <p>Note that i can be used as if it were a pointer for the purposes of
1516 invoking member functions of the <tt>Instruction</tt> class. This is
1517 because the indirection operator is overloaded for the iterator
1518 classes. In the above code, the expression <tt>i->size()</tt> is
1519 exactly equivalent to <tt>(*i).size()</tt> just like you'd expect.</p>
1523 <!-- _______________________________________________________________________ -->
1524 <div class="doc_subsubsection">
1525 <a name="iterate_basicblock">Iterating over the </a><a
1526 href="#Instruction"><tt>Instruction</tt></a>s in a <a
1527 href="#BasicBlock"><tt>BasicBlock</tt></a>
1530 <div class="doc_text">
1532 <p>Just like when dealing with <tt>BasicBlock</tt>s in <tt>Function</tt>s, it's
1533 easy to iterate over the individual instructions that make up
1534 <tt>BasicBlock</tt>s. Here's a code snippet that prints out each instruction in
1535 a <tt>BasicBlock</tt>:</p>
1537 <div class="doc_code">
1539 // <i>blk is a pointer to a BasicBlock instance</i>
1540 for (BasicBlock::iterator i = blk->begin(), e = blk->end(); i != e; ++i)
1541 // <i>The next statement works since operator<<(ostream&,...)</i>
1542 // <i>is overloaded for Instruction&</i>
1543 llvm::cerr << *i << "\n";
1547 <p>However, this isn't really the best way to print out the contents of a
1548 <tt>BasicBlock</tt>! Since the ostream operators are overloaded for virtually
1549 anything you'll care about, you could have just invoked the print routine on the
1550 basic block itself: <tt>llvm::cerr << *blk << "\n";</tt>.</p>
1554 <!-- _______________________________________________________________________ -->
1555 <div class="doc_subsubsection">
1556 <a name="iterate_institer">Iterating over the </a><a
1557 href="#Instruction"><tt>Instruction</tt></a>s in a <a
1558 href="#Function"><tt>Function</tt></a>
1561 <div class="doc_text">
1563 <p>If you're finding that you commonly iterate over a <tt>Function</tt>'s
1564 <tt>BasicBlock</tt>s and then that <tt>BasicBlock</tt>'s <tt>Instruction</tt>s,
1565 <tt>InstIterator</tt> should be used instead. You'll need to include <a
1566 href="/doxygen/InstIterator_8h-source.html"><tt>llvm/Support/InstIterator.h</tt></a>,
1567 and then instantiate <tt>InstIterator</tt>s explicitly in your code. Here's a
1568 small example that shows how to dump all instructions in a function to the standard error stream:<p>
1570 <div class="doc_code">
1572 #include "<a href="/doxygen/InstIterator_8h-source.html">llvm/Support/InstIterator.h</a>"
1574 // <i>F is a pointer to a Function instance</i>
1575 for (inst_iterator I = inst_begin(F), E = inst_end(F); I != E; ++I)
1576 llvm::cerr << *I << "\n";
1580 <p>Easy, isn't it? You can also use <tt>InstIterator</tt>s to fill a
1581 work list with its initial contents. For example, if you wanted to
1582 initialize a work list to contain all instructions in a <tt>Function</tt>
1583 F, all you would need to do is something like:</p>
1585 <div class="doc_code">
1587 std::set<Instruction*> worklist;
1588 // or better yet, SmallPtrSet<Instruction*, 64> worklist;
1590 for (inst_iterator I = inst_begin(F), E = inst_end(F); I != E; ++I)
1591 worklist.insert(&*I);
1595 <p>The STL set <tt>worklist</tt> would now contain all instructions in the
1596 <tt>Function</tt> pointed to by F.</p>
1600 <!-- _______________________________________________________________________ -->
1601 <div class="doc_subsubsection">
1602 <a name="iterate_convert">Turning an iterator into a class pointer (and
1606 <div class="doc_text">
1608 <p>Sometimes, it'll be useful to grab a reference (or pointer) to a class
1609 instance when all you've got at hand is an iterator. Well, extracting
1610 a reference or a pointer from an iterator is very straight-forward.
1611 Assuming that <tt>i</tt> is a <tt>BasicBlock::iterator</tt> and <tt>j</tt>
1612 is a <tt>BasicBlock::const_iterator</tt>:</p>
1614 <div class="doc_code">
1616 Instruction& inst = *i; // <i>Grab reference to instruction reference</i>
1617 Instruction* pinst = &*i; // <i>Grab pointer to instruction reference</i>
1618 const Instruction& inst = *j;
1622 <p>However, the iterators you'll be working with in the LLVM framework are
1623 special: they will automatically convert to a ptr-to-instance type whenever they
1624 need to. Instead of dereferencing the iterator and then taking the address of
1625 the result, you can simply assign the iterator to the proper pointer type and
1626 you get the dereference and address-of operation as a result of the assignment
1627 (behind the scenes, this is a result of overloading casting mechanisms). Thus
1628 the last line of the last example,</p>
1630 <div class="doc_code">
1632 Instruction *pinst = &*i;
1636 <p>is semantically equivalent to</p>
1638 <div class="doc_code">
1640 Instruction *pinst = i;
1644 <p>It's also possible to turn a class pointer into the corresponding iterator,
1645 and this is a constant time operation (very efficient). The following code
1646 snippet illustrates use of the conversion constructors provided by LLVM
1647 iterators. By using these, you can explicitly grab the iterator of something
1648 without actually obtaining it via iteration over some structure:</p>
1650 <div class="doc_code">
1652 void printNextInstruction(Instruction* inst) {
1653 BasicBlock::iterator it(inst);
1654 ++it; // <i>After this line, it refers to the instruction after *inst</i>
1655 if (it != inst->getParent()->end()) llvm::cerr << *it << "\n";
1662 <!--_______________________________________________________________________-->
1663 <div class="doc_subsubsection">
1664 <a name="iterate_complex">Finding call sites: a slightly more complex
1668 <div class="doc_text">
1670 <p>Say that you're writing a FunctionPass and would like to count all the
1671 locations in the entire module (that is, across every <tt>Function</tt>) where a
1672 certain function (i.e., some <tt>Function</tt>*) is already in scope. As you'll
1673 learn later, you may want to use an <tt>InstVisitor</tt> to accomplish this in a
1674 much more straight-forward manner, but this example will allow us to explore how
1675 you'd do it if you didn't have <tt>InstVisitor</tt> around. In pseudo-code, this
1676 is what we want to do:</p>
1678 <div class="doc_code">
1680 initialize callCounter to zero
1681 for each Function f in the Module
1682 for each BasicBlock b in f
1683 for each Instruction i in b
1684 if (i is a CallInst and calls the given function)
1685 increment callCounter
1689 <p>And the actual code is (remember, because we're writing a
1690 <tt>FunctionPass</tt>, our <tt>FunctionPass</tt>-derived class simply has to
1691 override the <tt>runOnFunction</tt> method):</p>
1693 <div class="doc_code">
1695 Function* targetFunc = ...;
1697 class OurFunctionPass : public FunctionPass {
1699 OurFunctionPass(): callCounter(0) { }
1701 virtual runOnFunction(Function& F) {
1702 for (Function::iterator b = F.begin(), be = F.end(); b != be; ++b) {
1703 for (BasicBlock::iterator i = b->begin(), ie = b->end(); i != ie; ++i) {
1704 if (<a href="#CallInst">CallInst</a>* callInst = <a href="#isa">dyn_cast</a><<a
1705 href="#CallInst">CallInst</a>>(&*i)) {
1706 // <i>We know we've encountered a call instruction, so we</i>
1707 // <i>need to determine if it's a call to the</i>
1708 // <i>function pointed to by m_func or not.</i>
1709 if (callInst->getCalledFunction() == targetFunc)
1717 unsigned callCounter;
1724 <!--_______________________________________________________________________-->
1725 <div class="doc_subsubsection">
1726 <a name="calls_and_invokes">Treating calls and invokes the same way</a>
1729 <div class="doc_text">
1731 <p>You may have noticed that the previous example was a bit oversimplified in
1732 that it did not deal with call sites generated by 'invoke' instructions. In
1733 this, and in other situations, you may find that you want to treat
1734 <tt>CallInst</tt>s and <tt>InvokeInst</tt>s the same way, even though their
1735 most-specific common base class is <tt>Instruction</tt>, which includes lots of
1736 less closely-related things. For these cases, LLVM provides a handy wrapper
1738 href="http://llvm.org/doxygen/classllvm_1_1CallSite.html"><tt>CallSite</tt></a>.
1739 It is essentially a wrapper around an <tt>Instruction</tt> pointer, with some
1740 methods that provide functionality common to <tt>CallInst</tt>s and
1741 <tt>InvokeInst</tt>s.</p>
1743 <p>This class has "value semantics": it should be passed by value, not by
1744 reference and it should not be dynamically allocated or deallocated using
1745 <tt>operator new</tt> or <tt>operator delete</tt>. It is efficiently copyable,
1746 assignable and constructable, with costs equivalents to that of a bare pointer.
1747 If you look at its definition, it has only a single pointer member.</p>
1751 <!--_______________________________________________________________________-->
1752 <div class="doc_subsubsection">
1753 <a name="iterate_chains">Iterating over def-use & use-def chains</a>
1756 <div class="doc_text">
1758 <p>Frequently, we might have an instance of the <a
1759 href="/doxygen/classllvm_1_1Value.html">Value Class</a> and we want to
1760 determine which <tt>User</tt>s use the <tt>Value</tt>. The list of all
1761 <tt>User</tt>s of a particular <tt>Value</tt> is called a <i>def-use</i> chain.
1762 For example, let's say we have a <tt>Function*</tt> named <tt>F</tt> to a
1763 particular function <tt>foo</tt>. Finding all of the instructions that
1764 <i>use</i> <tt>foo</tt> is as simple as iterating over the <i>def-use</i> chain
1767 <div class="doc_code">
1771 for (Value::use_iterator i = F->use_begin(), e = F->use_end(); i != e; ++i)
1772 if (Instruction *Inst = dyn_cast<Instruction>(*i)) {
1773 llvm::cerr << "F is used in instruction:\n";
1774 llvm::cerr << *Inst << "\n";
1779 <p>Alternately, it's common to have an instance of the <a
1780 href="/doxygen/classllvm_1_1User.html">User Class</a> and need to know what
1781 <tt>Value</tt>s are used by it. The list of all <tt>Value</tt>s used by a
1782 <tt>User</tt> is known as a <i>use-def</i> chain. Instances of class
1783 <tt>Instruction</tt> are common <tt>User</tt>s, so we might want to iterate over
1784 all of the values that a particular instruction uses (that is, the operands of
1785 the particular <tt>Instruction</tt>):</p>
1787 <div class="doc_code">
1789 Instruction *pi = ...;
1791 for (User::op_iterator i = pi->op_begin(), e = pi->op_end(); i != e; ++i) {
1799 def-use chains ("finding all users of"): Value::use_begin/use_end
1800 use-def chains ("finding all values used"): User::op_begin/op_end [op=operand]
1805 <!--_______________________________________________________________________-->
1806 <div class="doc_subsubsection">
1807 <a name="iterate_preds">Iterating over predecessors &
1808 successors of blocks</a>
1811 <div class="doc_text">
1813 <p>Iterating over the predecessors and successors of a block is quite easy
1814 with the routines defined in <tt>"llvm/Support/CFG.h"</tt>. Just use code like
1815 this to iterate over all predecessors of BB:</p>
1817 <div class="doc_code">
1819 #include "llvm/Support/CFG.h"
1820 BasicBlock *BB = ...;
1822 for (pred_iterator PI = pred_begin(BB), E = pred_end(BB); PI != E; ++PI) {
1823 BasicBlock *Pred = *PI;
1829 <p>Similarly, to iterate over successors use
1830 succ_iterator/succ_begin/succ_end.</p>
1835 <!-- ======================================================================= -->
1836 <div class="doc_subsection">
1837 <a name="simplechanges">Making simple changes</a>
1840 <div class="doc_text">
1842 <p>There are some primitive transformation operations present in the LLVM
1843 infrastructure that are worth knowing about. When performing
1844 transformations, it's fairly common to manipulate the contents of basic
1845 blocks. This section describes some of the common methods for doing so
1846 and gives example code.</p>
1850 <!--_______________________________________________________________________-->
1851 <div class="doc_subsubsection">
1852 <a name="schanges_creating">Creating and inserting new
1853 <tt>Instruction</tt>s</a>
1856 <div class="doc_text">
1858 <p><i>Instantiating Instructions</i></p>
1860 <p>Creation of <tt>Instruction</tt>s is straight-forward: simply call the
1861 constructor for the kind of instruction to instantiate and provide the necessary
1862 parameters. For example, an <tt>AllocaInst</tt> only <i>requires</i> a
1863 (const-ptr-to) <tt>Type</tt>. Thus:</p>
1865 <div class="doc_code">
1867 AllocaInst* ai = new AllocaInst(Type::Int32Ty);
1871 <p>will create an <tt>AllocaInst</tt> instance that represents the allocation of
1872 one integer in the current stack frame, at run time. Each <tt>Instruction</tt>
1873 subclass is likely to have varying default parameters which change the semantics
1874 of the instruction, so refer to the <a
1875 href="/doxygen/classllvm_1_1Instruction.html">doxygen documentation for the subclass of
1876 Instruction</a> that you're interested in instantiating.</p>
1878 <p><i>Naming values</i></p>
1880 <p>It is very useful to name the values of instructions when you're able to, as
1881 this facilitates the debugging of your transformations. If you end up looking
1882 at generated LLVM machine code, you definitely want to have logical names
1883 associated with the results of instructions! By supplying a value for the
1884 <tt>Name</tt> (default) parameter of the <tt>Instruction</tt> constructor, you
1885 associate a logical name with the result of the instruction's execution at
1886 run time. For example, say that I'm writing a transformation that dynamically
1887 allocates space for an integer on the stack, and that integer is going to be
1888 used as some kind of index by some other code. To accomplish this, I place an
1889 <tt>AllocaInst</tt> at the first point in the first <tt>BasicBlock</tt> of some
1890 <tt>Function</tt>, and I'm intending to use it within the same
1891 <tt>Function</tt>. I might do:</p>
1893 <div class="doc_code">
1895 AllocaInst* pa = new AllocaInst(Type::Int32Ty, 0, "indexLoc");
1899 <p>where <tt>indexLoc</tt> is now the logical name of the instruction's
1900 execution value, which is a pointer to an integer on the run time stack.</p>
1902 <p><i>Inserting instructions</i></p>
1904 <p>There are essentially two ways to insert an <tt>Instruction</tt>
1905 into an existing sequence of instructions that form a <tt>BasicBlock</tt>:</p>
1908 <li>Insertion into an explicit instruction list
1910 <p>Given a <tt>BasicBlock* pb</tt>, an <tt>Instruction* pi</tt> within that
1911 <tt>BasicBlock</tt>, and a newly-created instruction we wish to insert
1912 before <tt>*pi</tt>, we do the following: </p>
1914 <div class="doc_code">
1916 BasicBlock *pb = ...;
1917 Instruction *pi = ...;
1918 Instruction *newInst = new Instruction(...);
1920 pb->getInstList().insert(pi, newInst); // <i>Inserts newInst before pi in pb</i>
1924 <p>Appending to the end of a <tt>BasicBlock</tt> is so common that
1925 the <tt>Instruction</tt> class and <tt>Instruction</tt>-derived
1926 classes provide constructors which take a pointer to a
1927 <tt>BasicBlock</tt> to be appended to. For example code that
1930 <div class="doc_code">
1932 BasicBlock *pb = ...;
1933 Instruction *newInst = new Instruction(...);
1935 pb->getInstList().push_back(newInst); // <i>Appends newInst to pb</i>
1941 <div class="doc_code">
1943 BasicBlock *pb = ...;
1944 Instruction *newInst = new Instruction(..., pb);
1948 <p>which is much cleaner, especially if you are creating
1949 long instruction streams.</p></li>
1951 <li>Insertion into an implicit instruction list
1953 <p><tt>Instruction</tt> instances that are already in <tt>BasicBlock</tt>s
1954 are implicitly associated with an existing instruction list: the instruction
1955 list of the enclosing basic block. Thus, we could have accomplished the same
1956 thing as the above code without being given a <tt>BasicBlock</tt> by doing:
1959 <div class="doc_code">
1961 Instruction *pi = ...;
1962 Instruction *newInst = new Instruction(...);
1964 pi->getParent()->getInstList().insert(pi, newInst);
1968 <p>In fact, this sequence of steps occurs so frequently that the
1969 <tt>Instruction</tt> class and <tt>Instruction</tt>-derived classes provide
1970 constructors which take (as a default parameter) a pointer to an
1971 <tt>Instruction</tt> which the newly-created <tt>Instruction</tt> should
1972 precede. That is, <tt>Instruction</tt> constructors are capable of
1973 inserting the newly-created instance into the <tt>BasicBlock</tt> of a
1974 provided instruction, immediately before that instruction. Using an
1975 <tt>Instruction</tt> constructor with a <tt>insertBefore</tt> (default)
1976 parameter, the above code becomes:</p>
1978 <div class="doc_code">
1980 Instruction* pi = ...;
1981 Instruction* newInst = new Instruction(..., pi);
1985 <p>which is much cleaner, especially if you're creating a lot of
1986 instructions and adding them to <tt>BasicBlock</tt>s.</p></li>
1991 <!--_______________________________________________________________________-->
1992 <div class="doc_subsubsection">
1993 <a name="schanges_deleting">Deleting <tt>Instruction</tt>s</a>
1996 <div class="doc_text">
1998 <p>Deleting an instruction from an existing sequence of instructions that form a
1999 <a href="#BasicBlock"><tt>BasicBlock</tt></a> is very straight-forward. First,
2000 you must have a pointer to the instruction that you wish to delete. Second, you
2001 need to obtain the pointer to that instruction's basic block. You use the
2002 pointer to the basic block to get its list of instructions and then use the
2003 erase function to remove your instruction. For example:</p>
2005 <div class="doc_code">
2007 <a href="#Instruction">Instruction</a> *I = .. ;
2008 I->eraseFromParent();
2014 <!--_______________________________________________________________________-->
2015 <div class="doc_subsubsection">
2016 <a name="schanges_replacing">Replacing an <tt>Instruction</tt> with another
2020 <div class="doc_text">
2022 <p><i>Replacing individual instructions</i></p>
2024 <p>Including "<a href="/doxygen/BasicBlockUtils_8h-source.html">llvm/Transforms/Utils/BasicBlockUtils.h</a>"
2025 permits use of two very useful replace functions: <tt>ReplaceInstWithValue</tt>
2026 and <tt>ReplaceInstWithInst</tt>.</p>
2028 <h4><a name="schanges_deleting">Deleting <tt>Instruction</tt>s</a></h4>
2031 <li><tt>ReplaceInstWithValue</tt>
2033 <p>This function replaces all uses of a given instruction with a value,
2034 and then removes the original instruction. The following example
2035 illustrates the replacement of the result of a particular
2036 <tt>AllocaInst</tt> that allocates memory for a single integer with a null
2037 pointer to an integer.</p>
2039 <div class="doc_code">
2041 AllocaInst* instToReplace = ...;
2042 BasicBlock::iterator ii(instToReplace);
2044 ReplaceInstWithValue(instToReplace->getParent()->getInstList(), ii,
2045 Constant::getNullValue(PointerType::getUnqual(Type::Int32Ty)));
2048 <li><tt>ReplaceInstWithInst</tt>
2050 <p>This function replaces a particular instruction with another
2051 instruction, inserting the new instruction into the basic block at the
2052 location where the old instruction was, and replacing any uses of the old
2053 instruction with the new instruction. The following example illustrates
2054 the replacement of one <tt>AllocaInst</tt> with another.</p>
2056 <div class="doc_code">
2058 AllocaInst* instToReplace = ...;
2059 BasicBlock::iterator ii(instToReplace);
2061 ReplaceInstWithInst(instToReplace->getParent()->getInstList(), ii,
2062 new AllocaInst(Type::Int32Ty, 0, "ptrToReplacedInt"));
2066 <p><i>Replacing multiple uses of <tt>User</tt>s and <tt>Value</tt>s</i></p>
2068 <p>You can use <tt>Value::replaceAllUsesWith</tt> and
2069 <tt>User::replaceUsesOfWith</tt> to change more than one use at a time. See the
2070 doxygen documentation for the <a href="/doxygen/classllvm_1_1Value.html">Value Class</a>
2071 and <a href="/doxygen/classllvm_1_1User.html">User Class</a>, respectively, for more
2074 <!-- Value::replaceAllUsesWith User::replaceUsesOfWith Point out:
2075 include/llvm/Transforms/Utils/ especially BasicBlockUtils.h with:
2076 ReplaceInstWithValue, ReplaceInstWithInst -->
2080 <!--_______________________________________________________________________-->
2081 <div class="doc_subsubsection">
2082 <a name="schanges_deletingGV">Deleting <tt>GlobalVariable</tt>s</a>
2085 <div class="doc_text">
2087 <p>Deleting a global variable from a module is just as easy as deleting an
2088 Instruction. First, you must have a pointer to the global variable that you wish
2089 to delete. You use this pointer to erase it from its parent, the module.
2092 <div class="doc_code">
2094 <a href="#GlobalVariable">GlobalVariable</a> *GV = .. ;
2096 GV->eraseFromParent();
2102 <!-- ======================================================================= -->
2103 <div class="doc_subsection">
2104 <a name="create_types">How to Create Types</a>
2107 <div class="doc_text">
2109 <p>In generating IR, you may need some complex types. If you know these types
2110 statically, you can use <tt>TypeBuilder<...>::get()</tt>, defined
2111 in <tt>llvm/Support/TypeBuilder.h</tt>, to retrieve them. <tt>TypeBuilder</tt>
2112 has two forms depending on whether you're building types for cross-compilation
2113 or native library use. <tt>TypeBuilder<T, true></tt> requires
2114 that <tt>T</tt> be independent of the host environment, meaning that it's built
2116 the <a href="/doxygen/namespacellvm_1_1types.html"><tt>llvm::types</tt></a>
2117 namespace and pointers, functions, arrays, etc. built of
2118 those. <tt>TypeBuilder<T, false></tt> additionally allows native C types
2119 whose size may depend on the host compiler. For example,</p>
2121 <div class="doc_code">
2123 FunctionType *ft = TypeBuilder<types::i<8>(types::i<32>*), true>::get();
2127 <p>is easier to read and write than the equivalent</p>
2129 <div class="doc_code">
2131 std::vector<const Type*> params;
2132 params.push_back(PointerType::getUnqual(Type::Int32Ty));
2133 FunctionType *ft = FunctionType::get(Type::Int8Ty, params, false);
2137 <p>See the <a href="/doxygen/TypeBuilder_8h-source.html#l00001">class
2138 comment</a> for more details.</p>
2142 <!-- *********************************************************************** -->
2143 <div class="doc_section">
2144 <a name="threading">Threads and LLVM</a>
2146 <!-- *********************************************************************** -->
2148 <div class="doc_text">
2150 This section describes the interaction of the LLVM APIs with multithreading,
2151 both on the part of client applications, and in the JIT, in the hosted
2156 Note that LLVM's support for multithreading is still relatively young. Up
2157 through version 2.5, the execution of threaded hosted applications was
2158 supported, but not threaded client access to the APIs. While this use case is
2159 now supported, clients <em>must</em> adhere to the guidelines specified below to
2160 ensure proper operation in multithreaded mode.
2164 Note that, on Unix-like platforms, LLVM requires the presence of GCC's atomic
2165 intrinsics in order to support threaded operation. If you need a
2166 multhreading-capable LLVM on a platform without a suitably modern system
2167 compiler, consider compiling LLVM and LLVM-GCC in single-threaded mode, and
2168 using the resultant compiler to build a copy of LLVM with multithreading
2173 <!-- ======================================================================= -->
2174 <div class="doc_subsection">
2175 <a name="startmultithreaded">Entering Threaded Mode with
2176 <tt>llvm_start_multithreaded()</tt></a>
2179 <div class="doc_text">
2182 In order to properly protect its internal data structures while avoiding
2183 excessive locking overhead in the single-threaded case, the LLVM APIs require
2184 that the client invoke <tt>llvm_start_multithreaded()</tt>. This call must
2185 complete <em>before</em> any other threads attempt to invoke LLVM APIs. Any
2186 attempts to call LLVM APIs from multiple threads before
2187 <tt>llvm_start_multithreaded</tt> returns can and will cause corruption of
2188 LLVM's internal data.
2192 A caveat: before <tt>llvm_start_multithreaded()</tt> has been invoked, all
2193 <tt>llvm::sys::Mutex</tt> acquisitions and releases will become no-ops. This
2194 means that <tt>llvm_start_multithreaded()</tt> must be invoked before a threaded
2195 application can be executed in the JIT.
2199 <!-- ======================================================================= -->
2200 <div class="doc_subsection">
2201 <a name="shutdown">Ending Execution with <tt>llvm_shutdown()</tt></a>
2204 <div class="doc_text">
2206 When you are done using the LLVM APIs, you should call <tt>llvm_shutdown()</tt>
2207 to deallocate memory used for internal structures. This call must not begin
2208 while any other threads are still issuing LLVM API calls. Doing so is likely
2209 to result in garbage data or crashes.
2213 Note that, if you use scope-based shutdown, you can use the
2214 <tt>llvm_shutdown_obj</tt> class, which calls <tt>llvm_shutdown()</tt> in its
2218 <!-- ======================================================================= -->
2219 <div class="doc_subsection">
2220 <a name="managedstatic">Lazy Initialization with <tt>ManagedStatic</tt></a>
2223 <div class="doc_text">
2225 <tt>ManagedStatic</tt> is a utility class in LLVM used to implement static
2226 initialization of static resources, such as the global type tables. Before the
2227 invocation of <tt>llvm_shutdown()</tt>, it implements a simple lazy
2228 initialization scheme. Once <tt>llvm_start_multithreaded()</tt> returns,
2229 however, it uses double-checked locking to implement thread-safe lazy
2234 Note that, because no other threads are allowed to issue LLVM API calls before
2235 <tt>llvm_start_multithreaded()</tt> returns, it is possible to have
2236 <tt>ManagedStatic</tt>s of <tt>llvm::sys::Mutex</tt>s.
2240 <!-- *********************************************************************** -->
2241 <div class="doc_section">
2242 <a name="advanced">Advanced Topics</a>
2244 <!-- *********************************************************************** -->
2246 <div class="doc_text">
2248 This section describes some of the advanced or obscure API's that most clients
2249 do not need to be aware of. These API's tend manage the inner workings of the
2250 LLVM system, and only need to be accessed in unusual circumstances.
2254 <!-- ======================================================================= -->
2255 <div class="doc_subsection">
2256 <a name="TypeResolve">LLVM Type Resolution</a>
2259 <div class="doc_text">
2262 The LLVM type system has a very simple goal: allow clients to compare types for
2263 structural equality with a simple pointer comparison (aka a shallow compare).
2264 This goal makes clients much simpler and faster, and is used throughout the LLVM
2269 Unfortunately achieving this goal is not a simple matter. In particular,
2270 recursive types and late resolution of opaque types makes the situation very
2271 difficult to handle. Fortunately, for the most part, our implementation makes
2272 most clients able to be completely unaware of the nasty internal details. The
2273 primary case where clients are exposed to the inner workings of it are when
2274 building a recursive type. In addition to this case, the LLVM bitcode reader,
2275 assembly parser, and linker also have to be aware of the inner workings of this
2280 For our purposes below, we need three concepts. First, an "Opaque Type" is
2281 exactly as defined in the <a href="LangRef.html#t_opaque">language
2282 reference</a>. Second an "Abstract Type" is any type which includes an
2283 opaque type as part of its type graph (for example "<tt>{ opaque, i32 }</tt>").
2284 Third, a concrete type is a type that is not an abstract type (e.g. "<tt>{ i32,
2290 <!-- ______________________________________________________________________ -->
2291 <div class="doc_subsubsection">
2292 <a name="BuildRecType">Basic Recursive Type Construction</a>
2295 <div class="doc_text">
2298 Because the most common question is "how do I build a recursive type with LLVM",
2299 we answer it now and explain it as we go. Here we include enough to cause this
2300 to be emitted to an output .ll file:
2303 <div class="doc_code">
2305 %mylist = type { %mylist*, i32 }
2310 To build this, use the following LLVM APIs:
2313 <div class="doc_code">
2315 // <i>Create the initial outer struct</i>
2316 <a href="#PATypeHolder">PATypeHolder</a> StructTy = OpaqueType::get();
2317 std::vector<const Type*> Elts;
2318 Elts.push_back(PointerType::getUnqual(StructTy));
2319 Elts.push_back(Type::Int32Ty);
2320 StructType *NewSTy = StructType::get(Elts);
2322 // <i>At this point, NewSTy = "{ opaque*, i32 }". Tell VMCore that</i>
2323 // <i>the struct and the opaque type are actually the same.</i>
2324 cast<OpaqueType>(StructTy.get())-><a href="#refineAbstractTypeTo">refineAbstractTypeTo</a>(NewSTy);
2326 // <i>NewSTy is potentially invalidated, but StructTy (a <a href="#PATypeHolder">PATypeHolder</a>) is</i>
2327 // <i>kept up-to-date</i>
2328 NewSTy = cast<StructType>(StructTy.get());
2330 // <i>Add a name for the type to the module symbol table (optional)</i>
2331 MyModule->addTypeName("mylist", NewSTy);
2336 This code shows the basic approach used to build recursive types: build a
2337 non-recursive type using 'opaque', then use type unification to close the cycle.
2338 The type unification step is performed by the <tt><a
2339 href="#refineAbstractTypeTo">refineAbstractTypeTo</a></tt> method, which is
2340 described next. After that, we describe the <a
2341 href="#PATypeHolder">PATypeHolder class</a>.
2346 <!-- ______________________________________________________________________ -->
2347 <div class="doc_subsubsection">
2348 <a name="refineAbstractTypeTo">The <tt>refineAbstractTypeTo</tt> method</a>
2351 <div class="doc_text">
2353 The <tt>refineAbstractTypeTo</tt> method starts the type unification process.
2354 While this method is actually a member of the DerivedType class, it is most
2355 often used on OpaqueType instances. Type unification is actually a recursive
2356 process. After unification, types can become structurally isomorphic to
2357 existing types, and all duplicates are deleted (to preserve pointer equality).
2361 In the example above, the OpaqueType object is definitely deleted.
2362 Additionally, if there is an "{ \2*, i32}" type already created in the system,
2363 the pointer and struct type created are <b>also</b> deleted. Obviously whenever
2364 a type is deleted, any "Type*" pointers in the program are invalidated. As
2365 such, it is safest to avoid having <i>any</i> "Type*" pointers to abstract types
2366 live across a call to <tt>refineAbstractTypeTo</tt> (note that non-abstract
2367 types can never move or be deleted). To deal with this, the <a
2368 href="#PATypeHolder">PATypeHolder</a> class is used to maintain a stable
2369 reference to a possibly refined type, and the <a
2370 href="#AbstractTypeUser">AbstractTypeUser</a> class is used to update more
2371 complex datastructures.
2376 <!-- ______________________________________________________________________ -->
2377 <div class="doc_subsubsection">
2378 <a name="PATypeHolder">The PATypeHolder Class</a>
2381 <div class="doc_text">
2383 PATypeHolder is a form of a "smart pointer" for Type objects. When VMCore
2384 happily goes about nuking types that become isomorphic to existing types, it
2385 automatically updates all PATypeHolder objects to point to the new type. In the
2386 example above, this allows the code to maintain a pointer to the resultant
2387 resolved recursive type, even though the Type*'s are potentially invalidated.
2391 PATypeHolder is an extremely light-weight object that uses a lazy union-find
2392 implementation to update pointers. For example the pointer from a Value to its
2393 Type is maintained by PATypeHolder objects.
2398 <!-- ______________________________________________________________________ -->
2399 <div class="doc_subsubsection">
2400 <a name="AbstractTypeUser">The AbstractTypeUser Class</a>
2403 <div class="doc_text">
2406 Some data structures need more to perform more complex updates when types get
2407 resolved. To support this, a class can derive from the AbstractTypeUser class.
2409 allows it to get callbacks when certain types are resolved. To register to get
2410 callbacks for a particular type, the DerivedType::{add/remove}AbstractTypeUser
2411 methods can be called on a type. Note that these methods only work for <i>
2412 abstract</i> types. Concrete types (those that do not include any opaque
2413 objects) can never be refined.
2418 <!-- ======================================================================= -->
2419 <div class="doc_subsection">
2420 <a name="SymbolTable">The <tt>ValueSymbolTable</tt> and
2421 <tt>TypeSymbolTable</tt> classes</a>
2424 <div class="doc_text">
2425 <p>The <tt><a href="http://llvm.org/doxygen/classllvm_1_1ValueSymbolTable.html">
2426 ValueSymbolTable</a></tt> class provides a symbol table that the <a
2427 href="#Function"><tt>Function</tt></a> and <a href="#Module">
2428 <tt>Module</tt></a> classes use for naming value definitions. The symbol table
2429 can provide a name for any <a href="#Value"><tt>Value</tt></a>.
2430 The <tt><a href="http://llvm.org/doxygen/classllvm_1_1TypeSymbolTable.html">
2431 TypeSymbolTable</a></tt> class is used by the <tt>Module</tt> class to store
2432 names for types.</p>
2434 <p>Note that the <tt>SymbolTable</tt> class should not be directly accessed
2435 by most clients. It should only be used when iteration over the symbol table
2436 names themselves are required, which is very special purpose. Note that not
2438 <tt><a href="#Value">Value</a></tt>s have names, and those without names (i.e. they have
2439 an empty name) do not exist in the symbol table.
2442 <p>These symbol tables support iteration over the values/types in the symbol
2443 table with <tt>begin/end/iterator</tt> and supports querying to see if a
2444 specific name is in the symbol table (with <tt>lookup</tt>). The
2445 <tt>ValueSymbolTable</tt> class exposes no public mutator methods, instead,
2446 simply call <tt>setName</tt> on a value, which will autoinsert it into the
2447 appropriate symbol table. For types, use the Module::addTypeName method to
2448 insert entries into the symbol table.</p>
2454 <!-- ======================================================================= -->
2455 <div class="doc_subsection">
2456 <a name="UserLayout">The <tt>User</tt> and owned <tt>Use</tt> classes' memory layout</a>
2459 <div class="doc_text">
2460 <p>The <tt><a href="http://llvm.org/doxygen/classllvm_1_1User.html">
2461 User</a></tt> class provides a basis for expressing the ownership of <tt>User</tt>
2462 towards other <tt><a href="http://llvm.org/doxygen/classllvm_1_1Value.html">
2463 Value</a></tt>s. The <tt><a href="http://llvm.org/doxygen/classllvm_1_1Use.html">
2464 Use</a></tt> helper class is employed to do the bookkeeping and to facilitate <i>O(1)</i>
2465 addition and removal.</p>
2467 <!-- ______________________________________________________________________ -->
2468 <div class="doc_subsubsection">
2469 <a name="Use2User">Interaction and relationship between <tt>User</tt> and <tt>Use</tt> objects</a>
2472 <div class="doc_text">
2474 A subclass of <tt>User</tt> can choose between incorporating its <tt>Use</tt> objects
2475 or refer to them out-of-line by means of a pointer. A mixed variant
2476 (some <tt>Use</tt>s inline others hung off) is impractical and breaks the invariant
2477 that the <tt>Use</tt> objects belonging to the same <tt>User</tt> form a contiguous array.
2482 We have 2 different layouts in the <tt>User</tt> (sub)classes:
2485 The <tt>Use</tt> object(s) are inside (resp. at fixed offset) of the <tt>User</tt>
2486 object and there are a fixed number of them.</p>
2489 The <tt>Use</tt> object(s) are referenced by a pointer to an
2490 array from the <tt>User</tt> object and there may be a variable
2494 As of v2.4 each layout still possesses a direct pointer to the
2495 start of the array of <tt>Use</tt>s. Though not mandatory for layout a),
2496 we stick to this redundancy for the sake of simplicity.
2497 The <tt>User</tt> object also stores the number of <tt>Use</tt> objects it
2498 has. (Theoretically this information can also be calculated
2499 given the scheme presented below.)</p>
2501 Special forms of allocation operators (<tt>operator new</tt>)
2502 enforce the following memory layouts:</p>
2505 <li><p>Layout a) is modelled by prepending the <tt>User</tt> object by the <tt>Use[]</tt> array.</p>
2508 ...---.---.---.---.-------...
2509 | P | P | P | P | User
2510 '''---'---'---'---'-------'''
2513 <li><p>Layout b) is modelled by pointing at the <tt>Use[]</tt> array.</p>
2525 <i>(In the above figures '<tt>P</tt>' stands for the <tt>Use**</tt> that
2526 is stored in each <tt>Use</tt> object in the member <tt>Use::Prev</tt>)</i>
2528 <!-- ______________________________________________________________________ -->
2529 <div class="doc_subsubsection">
2530 <a name="Waymarking">The waymarking algorithm</a>
2533 <div class="doc_text">
2535 Since the <tt>Use</tt> objects are deprived of the direct (back)pointer to
2536 their <tt>User</tt> objects, there must be a fast and exact method to
2537 recover it. This is accomplished by the following scheme:</p>
2540 A bit-encoding in the 2 LSBits (least significant bits) of the <tt>Use::Prev</tt> allows to find the
2541 start of the <tt>User</tt> object:
2543 <li><tt>00</tt> —> binary digit 0</li>
2544 <li><tt>01</tt> —> binary digit 1</li>
2545 <li><tt>10</tt> —> stop and calculate (<tt>s</tt>)</li>
2546 <li><tt>11</tt> —> full stop (<tt>S</tt>)</li>
2549 Given a <tt>Use*</tt>, all we have to do is to walk till we get
2550 a stop and we either have a <tt>User</tt> immediately behind or
2551 we have to walk to the next stop picking up digits
2552 and calculating the offset:</p>
2554 .---.---.---.---.---.---.---.---.---.---.---.---.---.---.---.---.----------------
2555 | 1 | s | 1 | 0 | 1 | 0 | s | 1 | 1 | 0 | s | 1 | 1 | s | 1 | S | User (or User*)
2556 '---'---'---'---'---'---'---'---'---'---'---'---'---'---'---'---'----------------
2557 |+15 |+10 |+6 |+3 |+1
2560 | | |______________________>
2561 | |______________________________________>
2562 |__________________________________________________________>
2565 Only the significant number of bits need to be stored between the
2566 stops, so that the <i>worst case is 20 memory accesses</i> when there are
2567 1000 <tt>Use</tt> objects associated with a <tt>User</tt>.</p>
2569 <!-- ______________________________________________________________________ -->
2570 <div class="doc_subsubsection">
2571 <a name="ReferenceImpl">Reference implementation</a>
2574 <div class="doc_text">
2576 The following literate Haskell fragment demonstrates the concept:</p>
2579 <div class="doc_code">
2581 > import Test.QuickCheck
2583 > digits :: Int -> [Char] -> [Char]
2584 > digits 0 acc = '0' : acc
2585 > digits 1 acc = '1' : acc
2586 > digits n acc = digits (n `div` 2) $ digits (n `mod` 2) acc
2588 > dist :: Int -> [Char] -> [Char]
2591 > dist 1 acc = let r = dist 0 acc in 's' : digits (length r) r
2592 > dist n acc = dist (n - 1) $ dist 1 acc
2594 > takeLast n ss = reverse $ take n $ reverse ss
2596 > test = takeLast 40 $ dist 20 []
2601 Printing <test> gives: <tt>"1s100000s11010s10100s1111s1010s110s11s1S"</tt></p>
2603 The reverse algorithm computes the length of the string just by examining
2604 a certain prefix:</p>
2606 <div class="doc_code">
2608 > pref :: [Char] -> Int
2610 > pref ('s':'1':rest) = decode 2 1 rest
2611 > pref (_:rest) = 1 + pref rest
2613 > decode walk acc ('0':rest) = decode (walk + 1) (acc * 2) rest
2614 > decode walk acc ('1':rest) = decode (walk + 1) (acc * 2 + 1) rest
2615 > decode walk acc _ = walk + acc
2620 Now, as expected, printing <pref test> gives <tt>40</tt>.</p>
2622 We can <i>quickCheck</i> this with following property:</p>
2624 <div class="doc_code">
2626 > testcase = dist 2000 []
2627 > testcaseLength = length testcase
2629 > identityProp n = n > 0 && n <= testcaseLength ==> length arr == pref arr
2630 > where arr = takeLast n testcase
2635 As expected <quickCheck identityProp> gives:</p>
2638 *Main> quickCheck identityProp
2639 OK, passed 100 tests.
2642 Let's be a bit more exhaustive:</p>
2644 <div class="doc_code">
2647 > deepCheck p = check (defaultConfig { configMaxTest = 500 }) p
2652 And here is the result of <deepCheck identityProp>:</p>
2655 *Main> deepCheck identityProp
2656 OK, passed 500 tests.
2659 <!-- ______________________________________________________________________ -->
2660 <div class="doc_subsubsection">
2661 <a name="Tagging">Tagging considerations</a>
2665 To maintain the invariant that the 2 LSBits of each <tt>Use**</tt> in <tt>Use</tt>
2666 never change after being set up, setters of <tt>Use::Prev</tt> must re-tag the
2667 new <tt>Use**</tt> on every modification. Accordingly getters must strip the
2670 For layout b) instead of the <tt>User</tt> we find a pointer (<tt>User*</tt> with LSBit set).
2671 Following this pointer brings us to the <tt>User</tt>. A portable trick ensures
2672 that the first bytes of <tt>User</tt> (if interpreted as a pointer) never has
2673 the LSBit set. (Portability is relying on the fact that all known compilers place the
2674 <tt>vptr</tt> in the first word of the instances.)</p>
2678 <!-- *********************************************************************** -->
2679 <div class="doc_section">
2680 <a name="coreclasses">The Core LLVM Class Hierarchy Reference </a>
2682 <!-- *********************************************************************** -->
2684 <div class="doc_text">
2685 <p><tt>#include "<a href="/doxygen/Type_8h-source.html">llvm/Type.h</a>"</tt>
2686 <br>doxygen info: <a href="/doxygen/classllvm_1_1Type.html">Type Class</a></p>
2688 <p>The Core LLVM classes are the primary means of representing the program
2689 being inspected or transformed. The core LLVM classes are defined in
2690 header files in the <tt>include/llvm/</tt> directory, and implemented in
2691 the <tt>lib/VMCore</tt> directory.</p>
2695 <!-- ======================================================================= -->
2696 <div class="doc_subsection">
2697 <a name="Type">The <tt>Type</tt> class and Derived Types</a>
2700 <div class="doc_text">
2702 <p><tt>Type</tt> is a superclass of all type classes. Every <tt>Value</tt> has
2703 a <tt>Type</tt>. <tt>Type</tt> cannot be instantiated directly but only
2704 through its subclasses. Certain primitive types (<tt>VoidType</tt>,
2705 <tt>LabelType</tt>, <tt>FloatType</tt> and <tt>DoubleType</tt>) have hidden
2706 subclasses. They are hidden because they offer no useful functionality beyond
2707 what the <tt>Type</tt> class offers except to distinguish themselves from
2708 other subclasses of <tt>Type</tt>.</p>
2709 <p>All other types are subclasses of <tt>DerivedType</tt>. Types can be
2710 named, but this is not a requirement. There exists exactly
2711 one instance of a given shape at any one time. This allows type equality to
2712 be performed with address equality of the Type Instance. That is, given two
2713 <tt>Type*</tt> values, the types are identical if the pointers are identical.
2717 <!-- _______________________________________________________________________ -->
2718 <div class="doc_subsubsection">
2719 <a name="m_Type">Important Public Methods</a>
2722 <div class="doc_text">
2725 <li><tt>bool isInteger() const</tt>: Returns true for any integer type.</li>
2727 <li><tt>bool isFloatingPoint()</tt>: Return true if this is one of the two
2728 floating point types.</li>
2730 <li><tt>bool isAbstract()</tt>: Return true if the type is abstract (contains
2731 an OpaqueType anywhere in its definition).</li>
2733 <li><tt>bool isSized()</tt>: Return true if the type has known size. Things
2734 that don't have a size are abstract types, labels and void.</li>
2739 <!-- _______________________________________________________________________ -->
2740 <div class="doc_subsubsection">
2741 <a name="derivedtypes">Important Derived Types</a>
2743 <div class="doc_text">
2745 <dt><tt>IntegerType</tt></dt>
2746 <dd>Subclass of DerivedType that represents integer types of any bit width.
2747 Any bit width between <tt>IntegerType::MIN_INT_BITS</tt> (1) and
2748 <tt>IntegerType::MAX_INT_BITS</tt> (~8 million) can be represented.
2750 <li><tt>static const IntegerType* get(unsigned NumBits)</tt>: get an integer
2751 type of a specific bit width.</li>
2752 <li><tt>unsigned getBitWidth() const</tt>: Get the bit width of an integer
2756 <dt><tt>SequentialType</tt></dt>
2757 <dd>This is subclassed by ArrayType and PointerType
2759 <li><tt>const Type * getElementType() const</tt>: Returns the type of each
2760 of the elements in the sequential type. </li>
2763 <dt><tt>ArrayType</tt></dt>
2764 <dd>This is a subclass of SequentialType and defines the interface for array
2767 <li><tt>unsigned getNumElements() const</tt>: Returns the number of
2768 elements in the array. </li>
2771 <dt><tt>PointerType</tt></dt>
2772 <dd>Subclass of SequentialType for pointer types.</dd>
2773 <dt><tt>VectorType</tt></dt>
2774 <dd>Subclass of SequentialType for vector types. A
2775 vector type is similar to an ArrayType but is distinguished because it is
2776 a first class type wherease ArrayType is not. Vector types are used for
2777 vector operations and are usually small vectors of of an integer or floating
2779 <dt><tt>StructType</tt></dt>
2780 <dd>Subclass of DerivedTypes for struct types.</dd>
2781 <dt><tt><a name="FunctionType">FunctionType</a></tt></dt>
2782 <dd>Subclass of DerivedTypes for function types.
2784 <li><tt>bool isVarArg() const</tt>: Returns true if its a vararg
2786 <li><tt> const Type * getReturnType() const</tt>: Returns the
2787 return type of the function.</li>
2788 <li><tt>const Type * getParamType (unsigned i)</tt>: Returns
2789 the type of the ith parameter.</li>
2790 <li><tt> const unsigned getNumParams() const</tt>: Returns the
2791 number of formal parameters.</li>
2794 <dt><tt>OpaqueType</tt></dt>
2795 <dd>Sublcass of DerivedType for abstract types. This class
2796 defines no content and is used as a placeholder for some other type. Note
2797 that OpaqueType is used (temporarily) during type resolution for forward
2798 references of types. Once the referenced type is resolved, the OpaqueType
2799 is replaced with the actual type. OpaqueType can also be used for data
2800 abstraction. At link time opaque types can be resolved to actual types
2801 of the same name.</dd>
2807 <!-- ======================================================================= -->
2808 <div class="doc_subsection">
2809 <a name="Module">The <tt>Module</tt> class</a>
2812 <div class="doc_text">
2815 href="/doxygen/Module_8h-source.html">llvm/Module.h</a>"</tt><br> doxygen info:
2816 <a href="/doxygen/classllvm_1_1Module.html">Module Class</a></p>
2818 <p>The <tt>Module</tt> class represents the top level structure present in LLVM
2819 programs. An LLVM module is effectively either a translation unit of the
2820 original program or a combination of several translation units merged by the
2821 linker. The <tt>Module</tt> class keeps track of a list of <a
2822 href="#Function"><tt>Function</tt></a>s, a list of <a
2823 href="#GlobalVariable"><tt>GlobalVariable</tt></a>s, and a <a
2824 href="#SymbolTable"><tt>SymbolTable</tt></a>. Additionally, it contains a few
2825 helpful member functions that try to make common operations easy.</p>
2829 <!-- _______________________________________________________________________ -->
2830 <div class="doc_subsubsection">
2831 <a name="m_Module">Important Public Members of the <tt>Module</tt> class</a>
2834 <div class="doc_text">
2837 <li><tt>Module::Module(std::string name = "")</tt></li>
2840 <p>Constructing a <a href="#Module">Module</a> is easy. You can optionally
2841 provide a name for it (probably based on the name of the translation unit).</p>
2844 <li><tt>Module::iterator</tt> - Typedef for function list iterator<br>
2845 <tt>Module::const_iterator</tt> - Typedef for const_iterator.<br>
2847 <tt>begin()</tt>, <tt>end()</tt>
2848 <tt>size()</tt>, <tt>empty()</tt>
2850 <p>These are forwarding methods that make it easy to access the contents of
2851 a <tt>Module</tt> object's <a href="#Function"><tt>Function</tt></a>
2854 <li><tt>Module::FunctionListType &getFunctionList()</tt>
2856 <p> Returns the list of <a href="#Function"><tt>Function</tt></a>s. This is
2857 necessary to use when you need to update the list or perform a complex
2858 action that doesn't have a forwarding method.</p>
2860 <p><!-- Global Variable --></p></li>
2866 <li><tt>Module::global_iterator</tt> - Typedef for global variable list iterator<br>
2868 <tt>Module::const_global_iterator</tt> - Typedef for const_iterator.<br>
2870 <tt>global_begin()</tt>, <tt>global_end()</tt>
2871 <tt>global_size()</tt>, <tt>global_empty()</tt>
2873 <p> These are forwarding methods that make it easy to access the contents of
2874 a <tt>Module</tt> object's <a
2875 href="#GlobalVariable"><tt>GlobalVariable</tt></a> list.</p></li>
2877 <li><tt>Module::GlobalListType &getGlobalList()</tt>
2879 <p>Returns the list of <a
2880 href="#GlobalVariable"><tt>GlobalVariable</tt></a>s. This is necessary to
2881 use when you need to update the list or perform a complex action that
2882 doesn't have a forwarding method.</p>
2884 <p><!-- Symbol table stuff --> </p></li>
2890 <li><tt><a href="#SymbolTable">SymbolTable</a> *getSymbolTable()</tt>
2892 <p>Return a reference to the <a href="#SymbolTable"><tt>SymbolTable</tt></a>
2893 for this <tt>Module</tt>.</p>
2895 <p><!-- Convenience methods --></p></li>
2901 <li><tt><a href="#Function">Function</a> *getFunction(const std::string
2902 &Name, const <a href="#FunctionType">FunctionType</a> *Ty)</tt>
2904 <p>Look up the specified function in the <tt>Module</tt> <a
2905 href="#SymbolTable"><tt>SymbolTable</tt></a>. If it does not exist, return
2906 <tt>null</tt>.</p></li>
2908 <li><tt><a href="#Function">Function</a> *getOrInsertFunction(const
2909 std::string &Name, const <a href="#FunctionType">FunctionType</a> *T)</tt>
2911 <p>Look up the specified function in the <tt>Module</tt> <a
2912 href="#SymbolTable"><tt>SymbolTable</tt></a>. If it does not exist, add an
2913 external declaration for the function and return it.</p></li>
2915 <li><tt>std::string getTypeName(const <a href="#Type">Type</a> *Ty)</tt>
2917 <p>If there is at least one entry in the <a
2918 href="#SymbolTable"><tt>SymbolTable</tt></a> for the specified <a
2919 href="#Type"><tt>Type</tt></a>, return it. Otherwise return the empty
2922 <li><tt>bool addTypeName(const std::string &Name, const <a
2923 href="#Type">Type</a> *Ty)</tt>
2925 <p>Insert an entry in the <a href="#SymbolTable"><tt>SymbolTable</tt></a>
2926 mapping <tt>Name</tt> to <tt>Ty</tt>. If there is already an entry for this
2927 name, true is returned and the <a
2928 href="#SymbolTable"><tt>SymbolTable</tt></a> is not modified.</p></li>
2934 <!-- ======================================================================= -->
2935 <div class="doc_subsection">
2936 <a name="Value">The <tt>Value</tt> class</a>
2939 <div class="doc_text">
2941 <p><tt>#include "<a href="/doxygen/Value_8h-source.html">llvm/Value.h</a>"</tt>
2943 doxygen info: <a href="/doxygen/classllvm_1_1Value.html">Value Class</a></p>
2945 <p>The <tt>Value</tt> class is the most important class in the LLVM Source
2946 base. It represents a typed value that may be used (among other things) as an
2947 operand to an instruction. There are many different types of <tt>Value</tt>s,
2948 such as <a href="#Constant"><tt>Constant</tt></a>s,<a
2949 href="#Argument"><tt>Argument</tt></a>s. Even <a
2950 href="#Instruction"><tt>Instruction</tt></a>s and <a
2951 href="#Function"><tt>Function</tt></a>s are <tt>Value</tt>s.</p>
2953 <p>A particular <tt>Value</tt> may be used many times in the LLVM representation
2954 for a program. For example, an incoming argument to a function (represented
2955 with an instance of the <a href="#Argument">Argument</a> class) is "used" by
2956 every instruction in the function that references the argument. To keep track
2957 of this relationship, the <tt>Value</tt> class keeps a list of all of the <a
2958 href="#User"><tt>User</tt></a>s that is using it (the <a
2959 href="#User"><tt>User</tt></a> class is a base class for all nodes in the LLVM
2960 graph that can refer to <tt>Value</tt>s). This use list is how LLVM represents
2961 def-use information in the program, and is accessible through the <tt>use_</tt>*
2962 methods, shown below.</p>
2964 <p>Because LLVM is a typed representation, every LLVM <tt>Value</tt> is typed,
2965 and this <a href="#Type">Type</a> is available through the <tt>getType()</tt>
2966 method. In addition, all LLVM values can be named. The "name" of the
2967 <tt>Value</tt> is a symbolic string printed in the LLVM code:</p>
2969 <div class="doc_code">
2971 %<b>foo</b> = add i32 1, 2
2975 <p><a name="nameWarning">The name of this instruction is "foo".</a> <b>NOTE</b>
2976 that the name of any value may be missing (an empty string), so names should
2977 <b>ONLY</b> be used for debugging (making the source code easier to read,
2978 debugging printouts), they should not be used to keep track of values or map
2979 between them. For this purpose, use a <tt>std::map</tt> of pointers to the
2980 <tt>Value</tt> itself instead.</p>
2982 <p>One important aspect of LLVM is that there is no distinction between an SSA
2983 variable and the operation that produces it. Because of this, any reference to
2984 the value produced by an instruction (or the value available as an incoming
2985 argument, for example) is represented as a direct pointer to the instance of
2987 represents this value. Although this may take some getting used to, it
2988 simplifies the representation and makes it easier to manipulate.</p>
2992 <!-- _______________________________________________________________________ -->
2993 <div class="doc_subsubsection">
2994 <a name="m_Value">Important Public Members of the <tt>Value</tt> class</a>
2997 <div class="doc_text">
3000 <li><tt>Value::use_iterator</tt> - Typedef for iterator over the
3002 <tt>Value::use_const_iterator</tt> - Typedef for const_iterator over
3004 <tt>unsigned use_size()</tt> - Returns the number of users of the
3006 <tt>bool use_empty()</tt> - Returns true if there are no users.<br>
3007 <tt>use_iterator use_begin()</tt> - Get an iterator to the start of
3009 <tt>use_iterator use_end()</tt> - Get an iterator to the end of the
3011 <tt><a href="#User">User</a> *use_back()</tt> - Returns the last
3012 element in the list.
3013 <p> These methods are the interface to access the def-use
3014 information in LLVM. As with all other iterators in LLVM, the naming
3015 conventions follow the conventions defined by the <a href="#stl">STL</a>.</p>
3017 <li><tt><a href="#Type">Type</a> *getType() const</tt>
3018 <p>This method returns the Type of the Value.</p>
3020 <li><tt>bool hasName() const</tt><br>
3021 <tt>std::string getName() const</tt><br>
3022 <tt>void setName(const std::string &Name)</tt>
3023 <p> This family of methods is used to access and assign a name to a <tt>Value</tt>,
3024 be aware of the <a href="#nameWarning">precaution above</a>.</p>
3026 <li><tt>void replaceAllUsesWith(Value *V)</tt>
3028 <p>This method traverses the use list of a <tt>Value</tt> changing all <a
3029 href="#User"><tt>User</tt>s</a> of the current value to refer to
3030 "<tt>V</tt>" instead. For example, if you detect that an instruction always
3031 produces a constant value (for example through constant folding), you can
3032 replace all uses of the instruction with the constant like this:</p>
3034 <div class="doc_code">
3036 Inst->replaceAllUsesWith(ConstVal);
3044 <!-- ======================================================================= -->
3045 <div class="doc_subsection">
3046 <a name="User">The <tt>User</tt> class</a>
3049 <div class="doc_text">
3052 <tt>#include "<a href="/doxygen/User_8h-source.html">llvm/User.h</a>"</tt><br>
3053 doxygen info: <a href="/doxygen/classllvm_1_1User.html">User Class</a><br>
3054 Superclass: <a href="#Value"><tt>Value</tt></a></p>
3056 <p>The <tt>User</tt> class is the common base class of all LLVM nodes that may
3057 refer to <a href="#Value"><tt>Value</tt></a>s. It exposes a list of "Operands"
3058 that are all of the <a href="#Value"><tt>Value</tt></a>s that the User is
3059 referring to. The <tt>User</tt> class itself is a subclass of
3062 <p>The operands of a <tt>User</tt> point directly to the LLVM <a
3063 href="#Value"><tt>Value</tt></a> that it refers to. Because LLVM uses Static
3064 Single Assignment (SSA) form, there can only be one definition referred to,
3065 allowing this direct connection. This connection provides the use-def
3066 information in LLVM.</p>
3070 <!-- _______________________________________________________________________ -->
3071 <div class="doc_subsubsection">
3072 <a name="m_User">Important Public Members of the <tt>User</tt> class</a>
3075 <div class="doc_text">
3077 <p>The <tt>User</tt> class exposes the operand list in two ways: through
3078 an index access interface and through an iterator based interface.</p>
3081 <li><tt>Value *getOperand(unsigned i)</tt><br>
3082 <tt>unsigned getNumOperands()</tt>
3083 <p> These two methods expose the operands of the <tt>User</tt> in a
3084 convenient form for direct access.</p></li>
3086 <li><tt>User::op_iterator</tt> - Typedef for iterator over the operand
3088 <tt>op_iterator op_begin()</tt> - Get an iterator to the start of
3089 the operand list.<br>
3090 <tt>op_iterator op_end()</tt> - Get an iterator to the end of the
3092 <p> Together, these methods make up the iterator based interface to
3093 the operands of a <tt>User</tt>.</p></li>
3098 <!-- ======================================================================= -->
3099 <div class="doc_subsection">
3100 <a name="Instruction">The <tt>Instruction</tt> class</a>
3103 <div class="doc_text">
3105 <p><tt>#include "</tt><tt><a
3106 href="/doxygen/Instruction_8h-source.html">llvm/Instruction.h</a>"</tt><br>
3107 doxygen info: <a href="/doxygen/classllvm_1_1Instruction.html">Instruction Class</a><br>
3108 Superclasses: <a href="#User"><tt>User</tt></a>, <a
3109 href="#Value"><tt>Value</tt></a></p>
3111 <p>The <tt>Instruction</tt> class is the common base class for all LLVM
3112 instructions. It provides only a few methods, but is a very commonly used
3113 class. The primary data tracked by the <tt>Instruction</tt> class itself is the
3114 opcode (instruction type) and the parent <a
3115 href="#BasicBlock"><tt>BasicBlock</tt></a> the <tt>Instruction</tt> is embedded
3116 into. To represent a specific type of instruction, one of many subclasses of
3117 <tt>Instruction</tt> are used.</p>
3119 <p> Because the <tt>Instruction</tt> class subclasses the <a
3120 href="#User"><tt>User</tt></a> class, its operands can be accessed in the same
3121 way as for other <a href="#User"><tt>User</tt></a>s (with the
3122 <tt>getOperand()</tt>/<tt>getNumOperands()</tt> and
3123 <tt>op_begin()</tt>/<tt>op_end()</tt> methods).</p> <p> An important file for
3124 the <tt>Instruction</tt> class is the <tt>llvm/Instruction.def</tt> file. This
3125 file contains some meta-data about the various different types of instructions
3126 in LLVM. It describes the enum values that are used as opcodes (for example
3127 <tt>Instruction::Add</tt> and <tt>Instruction::ICmp</tt>), as well as the
3128 concrete sub-classes of <tt>Instruction</tt> that implement the instruction (for
3129 example <tt><a href="#BinaryOperator">BinaryOperator</a></tt> and <tt><a
3130 href="#CmpInst">CmpInst</a></tt>). Unfortunately, the use of macros in
3131 this file confuses doxygen, so these enum values don't show up correctly in the
3132 <a href="/doxygen/classllvm_1_1Instruction.html">doxygen output</a>.</p>
3136 <!-- _______________________________________________________________________ -->
3137 <div class="doc_subsubsection">
3138 <a name="s_Instruction">Important Subclasses of the <tt>Instruction</tt>
3141 <div class="doc_text">
3143 <li><tt><a name="BinaryOperator">BinaryOperator</a></tt>
3144 <p>This subclasses represents all two operand instructions whose operands
3145 must be the same type, except for the comparison instructions.</p></li>
3146 <li><tt><a name="CastInst">CastInst</a></tt>
3147 <p>This subclass is the parent of the 12 casting instructions. It provides
3148 common operations on cast instructions.</p>
3149 <li><tt><a name="CmpInst">CmpInst</a></tt>
3150 <p>This subclass respresents the two comparison instructions,
3151 <a href="LangRef.html#i_icmp">ICmpInst</a> (integer opreands), and
3152 <a href="LangRef.html#i_fcmp">FCmpInst</a> (floating point operands).</p>
3153 <li><tt><a name="TerminatorInst">TerminatorInst</a></tt>
3154 <p>This subclass is the parent of all terminator instructions (those which
3155 can terminate a block).</p>
3159 <!-- _______________________________________________________________________ -->
3160 <div class="doc_subsubsection">
3161 <a name="m_Instruction">Important Public Members of the <tt>Instruction</tt>
3165 <div class="doc_text">
3168 <li><tt><a href="#BasicBlock">BasicBlock</a> *getParent()</tt>
3169 <p>Returns the <a href="#BasicBlock"><tt>BasicBlock</tt></a> that
3170 this <tt>Instruction</tt> is embedded into.</p></li>
3171 <li><tt>bool mayWriteToMemory()</tt>
3172 <p>Returns true if the instruction writes to memory, i.e. it is a
3173 <tt>call</tt>,<tt>free</tt>,<tt>invoke</tt>, or <tt>store</tt>.</p></li>
3174 <li><tt>unsigned getOpcode()</tt>
3175 <p>Returns the opcode for the <tt>Instruction</tt>.</p></li>
3176 <li><tt><a href="#Instruction">Instruction</a> *clone() const</tt>
3177 <p>Returns another instance of the specified instruction, identical
3178 in all ways to the original except that the instruction has no parent
3179 (ie it's not embedded into a <a href="#BasicBlock"><tt>BasicBlock</tt></a>),
3180 and it has no name</p></li>
3185 <!-- ======================================================================= -->
3186 <div class="doc_subsection">
3187 <a name="Constant">The <tt>Constant</tt> class and subclasses</a>
3190 <div class="doc_text">
3192 <p>Constant represents a base class for different types of constants. It
3193 is subclassed by ConstantInt, ConstantArray, etc. for representing
3194 the various types of Constants. <a href="#GlobalValue">GlobalValue</a> is also
3195 a subclass, which represents the address of a global variable or function.
3200 <!-- _______________________________________________________________________ -->
3201 <div class="doc_subsubsection">Important Subclasses of Constant </div>
3202 <div class="doc_text">
3204 <li>ConstantInt : This subclass of Constant represents an integer constant of
3207 <li><tt>const APInt& getValue() const</tt>: Returns the underlying
3208 value of this constant, an APInt value.</li>
3209 <li><tt>int64_t getSExtValue() const</tt>: Converts the underlying APInt
3210 value to an int64_t via sign extension. If the value (not the bit width)
3211 of the APInt is too large to fit in an int64_t, an assertion will result.
3212 For this reason, use of this method is discouraged.</li>
3213 <li><tt>uint64_t getZExtValue() const</tt>: Converts the underlying APInt
3214 value to a uint64_t via zero extension. IF the value (not the bit width)
3215 of the APInt is too large to fit in a uint64_t, an assertion will result.
3216 For this reason, use of this method is discouraged.</li>
3217 <li><tt>static ConstantInt* get(const APInt& Val)</tt>: Returns the
3218 ConstantInt object that represents the value provided by <tt>Val</tt>.
3219 The type is implied as the IntegerType that corresponds to the bit width
3220 of <tt>Val</tt>.</li>
3221 <li><tt>static ConstantInt* get(const Type *Ty, uint64_t Val)</tt>:
3222 Returns the ConstantInt object that represents the value provided by
3223 <tt>Val</tt> for integer type <tt>Ty</tt>.</li>
3226 <li>ConstantFP : This class represents a floating point constant.
3228 <li><tt>double getValue() const</tt>: Returns the underlying value of
3229 this constant. </li>
3232 <li>ConstantArray : This represents a constant array.
3234 <li><tt>const std::vector<Use> &getValues() const</tt>: Returns
3235 a vector of component constants that makeup this array. </li>
3238 <li>ConstantStruct : This represents a constant struct.
3240 <li><tt>const std::vector<Use> &getValues() const</tt>: Returns
3241 a vector of component constants that makeup this array. </li>
3244 <li>GlobalValue : This represents either a global variable or a function. In
3245 either case, the value is a constant fixed address (after linking).
3251 <!-- ======================================================================= -->
3252 <div class="doc_subsection">
3253 <a name="GlobalValue">The <tt>GlobalValue</tt> class</a>
3256 <div class="doc_text">
3259 href="/doxygen/GlobalValue_8h-source.html">llvm/GlobalValue.h</a>"</tt><br>
3260 doxygen info: <a href="/doxygen/classllvm_1_1GlobalValue.html">GlobalValue
3262 Superclasses: <a href="#Constant"><tt>Constant</tt></a>,
3263 <a href="#User"><tt>User</tt></a>, <a href="#Value"><tt>Value</tt></a></p>
3265 <p>Global values (<a href="#GlobalVariable"><tt>GlobalVariable</tt></a>s or <a
3266 href="#Function"><tt>Function</tt></a>s) are the only LLVM values that are
3267 visible in the bodies of all <a href="#Function"><tt>Function</tt></a>s.
3268 Because they are visible at global scope, they are also subject to linking with
3269 other globals defined in different translation units. To control the linking
3270 process, <tt>GlobalValue</tt>s know their linkage rules. Specifically,
3271 <tt>GlobalValue</tt>s know whether they have internal or external linkage, as
3272 defined by the <tt>LinkageTypes</tt> enumeration.</p>
3274 <p>If a <tt>GlobalValue</tt> has internal linkage (equivalent to being
3275 <tt>static</tt> in C), it is not visible to code outside the current translation
3276 unit, and does not participate in linking. If it has external linkage, it is
3277 visible to external code, and does participate in linking. In addition to
3278 linkage information, <tt>GlobalValue</tt>s keep track of which <a
3279 href="#Module"><tt>Module</tt></a> they are currently part of.</p>
3281 <p>Because <tt>GlobalValue</tt>s are memory objects, they are always referred to
3282 by their <b>address</b>. As such, the <a href="#Type"><tt>Type</tt></a> of a
3283 global is always a pointer to its contents. It is important to remember this
3284 when using the <tt>GetElementPtrInst</tt> instruction because this pointer must
3285 be dereferenced first. For example, if you have a <tt>GlobalVariable</tt> (a
3286 subclass of <tt>GlobalValue)</tt> that is an array of 24 ints, type <tt>[24 x
3287 i32]</tt>, then the <tt>GlobalVariable</tt> is a pointer to that array. Although
3288 the address of the first element of this array and the value of the
3289 <tt>GlobalVariable</tt> are the same, they have different types. The
3290 <tt>GlobalVariable</tt>'s type is <tt>[24 x i32]</tt>. The first element's type
3291 is <tt>i32.</tt> Because of this, accessing a global value requires you to
3292 dereference the pointer with <tt>GetElementPtrInst</tt> first, then its elements
3293 can be accessed. This is explained in the <a href="LangRef.html#globalvars">LLVM
3294 Language Reference Manual</a>.</p>
3298 <!-- _______________________________________________________________________ -->
3299 <div class="doc_subsubsection">
3300 <a name="m_GlobalValue">Important Public Members of the <tt>GlobalValue</tt>
3304 <div class="doc_text">
3307 <li><tt>bool hasInternalLinkage() const</tt><br>
3308 <tt>bool hasExternalLinkage() const</tt><br>
3309 <tt>void setInternalLinkage(bool HasInternalLinkage)</tt>
3310 <p> These methods manipulate the linkage characteristics of the <tt>GlobalValue</tt>.</p>
3313 <li><tt><a href="#Module">Module</a> *getParent()</tt>
3314 <p> This returns the <a href="#Module"><tt>Module</tt></a> that the
3315 GlobalValue is currently embedded into.</p></li>
3320 <!-- ======================================================================= -->
3321 <div class="doc_subsection">
3322 <a name="Function">The <tt>Function</tt> class</a>
3325 <div class="doc_text">
3328 href="/doxygen/Function_8h-source.html">llvm/Function.h</a>"</tt><br> doxygen
3329 info: <a href="/doxygen/classllvm_1_1Function.html">Function Class</a><br>
3330 Superclasses: <a href="#GlobalValue"><tt>GlobalValue</tt></a>,
3331 <a href="#Constant"><tt>Constant</tt></a>,
3332 <a href="#User"><tt>User</tt></a>,
3333 <a href="#Value"><tt>Value</tt></a></p>
3335 <p>The <tt>Function</tt> class represents a single procedure in LLVM. It is
3336 actually one of the more complex classes in the LLVM heirarchy because it must
3337 keep track of a large amount of data. The <tt>Function</tt> class keeps track
3338 of a list of <a href="#BasicBlock"><tt>BasicBlock</tt></a>s, a list of formal
3339 <a href="#Argument"><tt>Argument</tt></a>s, and a
3340 <a href="#SymbolTable"><tt>SymbolTable</tt></a>.</p>
3342 <p>The list of <a href="#BasicBlock"><tt>BasicBlock</tt></a>s is the most
3343 commonly used part of <tt>Function</tt> objects. The list imposes an implicit
3344 ordering of the blocks in the function, which indicate how the code will be
3345 layed out by the backend. Additionally, the first <a
3346 href="#BasicBlock"><tt>BasicBlock</tt></a> is the implicit entry node for the
3347 <tt>Function</tt>. It is not legal in LLVM to explicitly branch to this initial
3348 block. There are no implicit exit nodes, and in fact there may be multiple exit
3349 nodes from a single <tt>Function</tt>. If the <a
3350 href="#BasicBlock"><tt>BasicBlock</tt></a> list is empty, this indicates that
3351 the <tt>Function</tt> is actually a function declaration: the actual body of the
3352 function hasn't been linked in yet.</p>
3354 <p>In addition to a list of <a href="#BasicBlock"><tt>BasicBlock</tt></a>s, the
3355 <tt>Function</tt> class also keeps track of the list of formal <a
3356 href="#Argument"><tt>Argument</tt></a>s that the function receives. This
3357 container manages the lifetime of the <a href="#Argument"><tt>Argument</tt></a>
3358 nodes, just like the <a href="#BasicBlock"><tt>BasicBlock</tt></a> list does for
3359 the <a href="#BasicBlock"><tt>BasicBlock</tt></a>s.</p>
3361 <p>The <a href="#SymbolTable"><tt>SymbolTable</tt></a> is a very rarely used
3362 LLVM feature that is only used when you have to look up a value by name. Aside
3363 from that, the <a href="#SymbolTable"><tt>SymbolTable</tt></a> is used
3364 internally to make sure that there are not conflicts between the names of <a
3365 href="#Instruction"><tt>Instruction</tt></a>s, <a
3366 href="#BasicBlock"><tt>BasicBlock</tt></a>s, or <a
3367 href="#Argument"><tt>Argument</tt></a>s in the function body.</p>
3369 <p>Note that <tt>Function</tt> is a <a href="#GlobalValue">GlobalValue</a>
3370 and therefore also a <a href="#Constant">Constant</a>. The value of the function
3371 is its address (after linking) which is guaranteed to be constant.</p>
3374 <!-- _______________________________________________________________________ -->
3375 <div class="doc_subsubsection">
3376 <a name="m_Function">Important Public Members of the <tt>Function</tt>
3380 <div class="doc_text">
3383 <li><tt>Function(const </tt><tt><a href="#FunctionType">FunctionType</a>
3384 *Ty, LinkageTypes Linkage, const std::string &N = "", Module* Parent = 0)</tt>
3386 <p>Constructor used when you need to create new <tt>Function</tt>s to add
3387 the the program. The constructor must specify the type of the function to
3388 create and what type of linkage the function should have. The <a
3389 href="#FunctionType"><tt>FunctionType</tt></a> argument
3390 specifies the formal arguments and return value for the function. The same
3391 <a href="#FunctionType"><tt>FunctionType</tt></a> value can be used to
3392 create multiple functions. The <tt>Parent</tt> argument specifies the Module
3393 in which the function is defined. If this argument is provided, the function
3394 will automatically be inserted into that module's list of
3397 <li><tt>bool isDeclaration()</tt>
3399 <p>Return whether or not the <tt>Function</tt> has a body defined. If the
3400 function is "external", it does not have a body, and thus must be resolved
3401 by linking with a function defined in a different translation unit.</p></li>
3403 <li><tt>Function::iterator</tt> - Typedef for basic block list iterator<br>
3404 <tt>Function::const_iterator</tt> - Typedef for const_iterator.<br>
3406 <tt>begin()</tt>, <tt>end()</tt>
3407 <tt>size()</tt>, <tt>empty()</tt>
3409 <p>These are forwarding methods that make it easy to access the contents of
3410 a <tt>Function</tt> object's <a href="#BasicBlock"><tt>BasicBlock</tt></a>
3413 <li><tt>Function::BasicBlockListType &getBasicBlockList()</tt>
3415 <p>Returns the list of <a href="#BasicBlock"><tt>BasicBlock</tt></a>s. This
3416 is necessary to use when you need to update the list or perform a complex
3417 action that doesn't have a forwarding method.</p></li>
3419 <li><tt>Function::arg_iterator</tt> - Typedef for the argument list
3421 <tt>Function::const_arg_iterator</tt> - Typedef for const_iterator.<br>
3423 <tt>arg_begin()</tt>, <tt>arg_end()</tt>
3424 <tt>arg_size()</tt>, <tt>arg_empty()</tt>
3426 <p>These are forwarding methods that make it easy to access the contents of
3427 a <tt>Function</tt> object's <a href="#Argument"><tt>Argument</tt></a>
3430 <li><tt>Function::ArgumentListType &getArgumentList()</tt>
3432 <p>Returns the list of <a href="#Argument"><tt>Argument</tt></a>s. This is
3433 necessary to use when you need to update the list or perform a complex
3434 action that doesn't have a forwarding method.</p></li>
3436 <li><tt><a href="#BasicBlock">BasicBlock</a> &getEntryBlock()</tt>
3438 <p>Returns the entry <a href="#BasicBlock"><tt>BasicBlock</tt></a> for the
3439 function. Because the entry block for the function is always the first
3440 block, this returns the first block of the <tt>Function</tt>.</p></li>
3442 <li><tt><a href="#Type">Type</a> *getReturnType()</tt><br>
3443 <tt><a href="#FunctionType">FunctionType</a> *getFunctionType()</tt>
3445 <p>This traverses the <a href="#Type"><tt>Type</tt></a> of the
3446 <tt>Function</tt> and returns the return type of the function, or the <a
3447 href="#FunctionType"><tt>FunctionType</tt></a> of the actual
3450 <li><tt><a href="#SymbolTable">SymbolTable</a> *getSymbolTable()</tt>
3452 <p> Return a pointer to the <a href="#SymbolTable"><tt>SymbolTable</tt></a>
3453 for this <tt>Function</tt>.</p></li>
3458 <!-- ======================================================================= -->
3459 <div class="doc_subsection">
3460 <a name="GlobalVariable">The <tt>GlobalVariable</tt> class</a>
3463 <div class="doc_text">
3466 href="/doxygen/GlobalVariable_8h-source.html">llvm/GlobalVariable.h</a>"</tt>
3468 doxygen info: <a href="/doxygen/classllvm_1_1GlobalVariable.html">GlobalVariable
3470 Superclasses: <a href="#GlobalValue"><tt>GlobalValue</tt></a>,
3471 <a href="#Constant"><tt>Constant</tt></a>,
3472 <a href="#User"><tt>User</tt></a>,
3473 <a href="#Value"><tt>Value</tt></a></p>
3475 <p>Global variables are represented with the (suprise suprise)
3476 <tt>GlobalVariable</tt> class. Like functions, <tt>GlobalVariable</tt>s are also
3477 subclasses of <a href="#GlobalValue"><tt>GlobalValue</tt></a>, and as such are
3478 always referenced by their address (global values must live in memory, so their
3479 "name" refers to their constant address). See
3480 <a href="#GlobalValue"><tt>GlobalValue</tt></a> for more on this. Global
3481 variables may have an initial value (which must be a
3482 <a href="#Constant"><tt>Constant</tt></a>), and if they have an initializer,
3483 they may be marked as "constant" themselves (indicating that their contents
3484 never change at runtime).</p>
3487 <!-- _______________________________________________________________________ -->
3488 <div class="doc_subsubsection">
3489 <a name="m_GlobalVariable">Important Public Members of the
3490 <tt>GlobalVariable</tt> class</a>
3493 <div class="doc_text">
3496 <li><tt>GlobalVariable(const </tt><tt><a href="#Type">Type</a> *Ty, bool
3497 isConstant, LinkageTypes& Linkage, <a href="#Constant">Constant</a>
3498 *Initializer = 0, const std::string &Name = "", Module* Parent = 0)</tt>
3500 <p>Create a new global variable of the specified type. If
3501 <tt>isConstant</tt> is true then the global variable will be marked as
3502 unchanging for the program. The Linkage parameter specifies the type of
3503 linkage (internal, external, weak, linkonce, appending) for the variable.
3504 If the linkage is InternalLinkage, WeakAnyLinkage, WeakODRLinkage,
3505 LinkOnceAnyLinkage or LinkOnceODRLinkage, then the resultant
3506 global variable will have internal linkage. AppendingLinkage concatenates
3507 together all instances (in different translation units) of the variable
3508 into a single variable but is only applicable to arrays. See
3509 the <a href="LangRef.html#modulestructure">LLVM Language Reference</a> for
3510 further details on linkage types. Optionally an initializer, a name, and the
3511 module to put the variable into may be specified for the global variable as
3514 <li><tt>bool isConstant() const</tt>
3516 <p>Returns true if this is a global variable that is known not to
3517 be modified at runtime.</p></li>
3519 <li><tt>bool hasInitializer()</tt>
3521 <p>Returns true if this <tt>GlobalVariable</tt> has an intializer.</p></li>
3523 <li><tt><a href="#Constant">Constant</a> *getInitializer()</tt>
3525 <p>Returns the intial value for a <tt>GlobalVariable</tt>. It is not legal
3526 to call this method if there is no initializer.</p></li>
3532 <!-- ======================================================================= -->
3533 <div class="doc_subsection">
3534 <a name="BasicBlock">The <tt>BasicBlock</tt> class</a>
3537 <div class="doc_text">
3540 href="/doxygen/BasicBlock_8h-source.html">llvm/BasicBlock.h</a>"</tt><br>
3541 doxygen info: <a href="/doxygen/structllvm_1_1BasicBlock.html">BasicBlock
3543 Superclass: <a href="#Value"><tt>Value</tt></a></p>
3545 <p>This class represents a single entry multiple exit section of the code,
3546 commonly known as a basic block by the compiler community. The
3547 <tt>BasicBlock</tt> class maintains a list of <a
3548 href="#Instruction"><tt>Instruction</tt></a>s, which form the body of the block.
3549 Matching the language definition, the last element of this list of instructions
3550 is always a terminator instruction (a subclass of the <a
3551 href="#TerminatorInst"><tt>TerminatorInst</tt></a> class).</p>
3553 <p>In addition to tracking the list of instructions that make up the block, the
3554 <tt>BasicBlock</tt> class also keeps track of the <a
3555 href="#Function"><tt>Function</tt></a> that it is embedded into.</p>
3557 <p>Note that <tt>BasicBlock</tt>s themselves are <a
3558 href="#Value"><tt>Value</tt></a>s, because they are referenced by instructions
3559 like branches and can go in the switch tables. <tt>BasicBlock</tt>s have type
3564 <!-- _______________________________________________________________________ -->
3565 <div class="doc_subsubsection">
3566 <a name="m_BasicBlock">Important Public Members of the <tt>BasicBlock</tt>
3570 <div class="doc_text">
3573 <li><tt>BasicBlock(const std::string &Name = "", </tt><tt><a
3574 href="#Function">Function</a> *Parent = 0)</tt>
3576 <p>The <tt>BasicBlock</tt> constructor is used to create new basic blocks for
3577 insertion into a function. The constructor optionally takes a name for the new
3578 block, and a <a href="#Function"><tt>Function</tt></a> to insert it into. If
3579 the <tt>Parent</tt> parameter is specified, the new <tt>BasicBlock</tt> is
3580 automatically inserted at the end of the specified <a
3581 href="#Function"><tt>Function</tt></a>, if not specified, the BasicBlock must be
3582 manually inserted into the <a href="#Function"><tt>Function</tt></a>.</p></li>
3584 <li><tt>BasicBlock::iterator</tt> - Typedef for instruction list iterator<br>
3585 <tt>BasicBlock::const_iterator</tt> - Typedef for const_iterator.<br>
3586 <tt>begin()</tt>, <tt>end()</tt>, <tt>front()</tt>, <tt>back()</tt>,
3587 <tt>size()</tt>, <tt>empty()</tt>
3588 STL-style functions for accessing the instruction list.
3590 <p>These methods and typedefs are forwarding functions that have the same
3591 semantics as the standard library methods of the same names. These methods
3592 expose the underlying instruction list of a basic block in a way that is easy to
3593 manipulate. To get the full complement of container operations (including
3594 operations to update the list), you must use the <tt>getInstList()</tt>
3597 <li><tt>BasicBlock::InstListType &getInstList()</tt>
3599 <p>This method is used to get access to the underlying container that actually
3600 holds the Instructions. This method must be used when there isn't a forwarding
3601 function in the <tt>BasicBlock</tt> class for the operation that you would like
3602 to perform. Because there are no forwarding functions for "updating"
3603 operations, you need to use this if you want to update the contents of a
3604 <tt>BasicBlock</tt>.</p></li>
3606 <li><tt><a href="#Function">Function</a> *getParent()</tt>
3608 <p> Returns a pointer to <a href="#Function"><tt>Function</tt></a> the block is
3609 embedded into, or a null pointer if it is homeless.</p></li>
3611 <li><tt><a href="#TerminatorInst">TerminatorInst</a> *getTerminator()</tt>
3613 <p> Returns a pointer to the terminator instruction that appears at the end of
3614 the <tt>BasicBlock</tt>. If there is no terminator instruction, or if the last
3615 instruction in the block is not a terminator, then a null pointer is
3623 <!-- ======================================================================= -->
3624 <div class="doc_subsection">
3625 <a name="Argument">The <tt>Argument</tt> class</a>
3628 <div class="doc_text">
3630 <p>This subclass of Value defines the interface for incoming formal
3631 arguments to a function. A Function maintains a list of its formal
3632 arguments. An argument has a pointer to the parent Function.</p>
3636 <!-- *********************************************************************** -->
3639 <a href="http://jigsaw.w3.org/css-validator/check/referer"><img
3640 src="http://jigsaw.w3.org/css-validator/images/vcss-blue" alt="Valid CSS"></a>
3641 <a href="http://validator.w3.org/check/referer"><img
3642 src="http://www.w3.org/Icons/valid-html401" alt="Valid HTML 4.01 Strict"></a>
3644 <a href="mailto:dhurjati@cs.uiuc.edu">Dinakar Dhurjati</a> and
3645 <a href="mailto:sabre@nondot.org">Chris Lattner</a><br>
3646 <a href="http://llvm.org">The LLVM Compiler Infrastructure</a><br>
3647 Last modified: $Date$