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 and Exiting Multithreaded Mode
137 <li><a href="#shutdown">Ending execution with <tt>llvm_shutdown()</tt></a></li>
138 <li><a href="#managedstatic">Lazy initialization with <tt>ManagedStatic</tt></a></li>
142 <li><a href="#advanced">Advanced Topics</a>
144 <li><a href="#TypeResolve">LLVM Type Resolution</a>
146 <li><a href="#BuildRecType">Basic Recursive Type Construction</a></li>
147 <li><a href="#refineAbstractTypeTo">The <tt>refineAbstractTypeTo</tt> method</a></li>
148 <li><a href="#PATypeHolder">The PATypeHolder Class</a></li>
149 <li><a href="#AbstractTypeUser">The AbstractTypeUser Class</a></li>
152 <li><a href="#SymbolTable">The <tt>ValueSymbolTable</tt> and <tt>TypeSymbolTable</tt> classes</a></li>
153 <li><a href="#UserLayout">The <tt>User</tt> and owned <tt>Use</tt> classes' memory layout</a></li>
156 <li><a href="#coreclasses">The Core LLVM Class Hierarchy Reference</a>
158 <li><a href="#Type">The <tt>Type</tt> class</a> </li>
159 <li><a href="#Module">The <tt>Module</tt> class</a></li>
160 <li><a href="#Value">The <tt>Value</tt> class</a>
162 <li><a href="#User">The <tt>User</tt> class</a>
164 <li><a href="#Instruction">The <tt>Instruction</tt> class</a></li>
165 <li><a href="#Constant">The <tt>Constant</tt> class</a>
167 <li><a href="#GlobalValue">The <tt>GlobalValue</tt> class</a>
169 <li><a href="#Function">The <tt>Function</tt> class</a></li>
170 <li><a href="#GlobalVariable">The <tt>GlobalVariable</tt> class</a></li>
177 <li><a href="#BasicBlock">The <tt>BasicBlock</tt> class</a></li>
178 <li><a href="#Argument">The <tt>Argument</tt> class</a></li>
185 <div class="doc_author">
186 <p>Written by <a href="mailto:sabre@nondot.org">Chris Lattner</a>,
187 <a href="mailto:dhurjati@cs.uiuc.edu">Dinakar Dhurjati</a>,
188 <a href="mailto:ggreif@gmail.com">Gabor Greif</a>,
189 <a href="mailto:jstanley@cs.uiuc.edu">Joel Stanley</a>,
190 <a href="mailto:rspencer@x10sys.com">Reid Spencer</a> and
191 <a href="mailto:owen@apple.com">Owen Anderson</a></p>
194 <!-- *********************************************************************** -->
195 <div class="doc_section">
196 <a name="introduction">Introduction </a>
198 <!-- *********************************************************************** -->
200 <div class="doc_text">
202 <p>This document is meant to highlight some of the important classes and
203 interfaces available in the LLVM source-base. This manual is not
204 intended to explain what LLVM is, how it works, and what LLVM code looks
205 like. It assumes that you know the basics of LLVM and are interested
206 in writing transformations or otherwise analyzing or manipulating the
209 <p>This document should get you oriented so that you can find your
210 way in the continuously growing source code that makes up the LLVM
211 infrastructure. Note that this manual is not intended to serve as a
212 replacement for reading the source code, so if you think there should be
213 a method in one of these classes to do something, but it's not listed,
214 check the source. Links to the <a href="/doxygen/">doxygen</a> sources
215 are provided to make this as easy as possible.</p>
217 <p>The first section of this document describes general information that is
218 useful to know when working in the LLVM infrastructure, and the second describes
219 the Core LLVM classes. In the future this manual will be extended with
220 information describing how to use extension libraries, such as dominator
221 information, CFG traversal routines, and useful utilities like the <tt><a
222 href="/doxygen/InstVisitor_8h-source.html">InstVisitor</a></tt> template.</p>
226 <!-- *********************************************************************** -->
227 <div class="doc_section">
228 <a name="general">General Information</a>
230 <!-- *********************************************************************** -->
232 <div class="doc_text">
234 <p>This section contains general information that is useful if you are working
235 in the LLVM source-base, but that isn't specific to any particular API.</p>
239 <!-- ======================================================================= -->
240 <div class="doc_subsection">
241 <a name="stl">The C++ Standard Template Library</a>
244 <div class="doc_text">
246 <p>LLVM makes heavy use of the C++ Standard Template Library (STL),
247 perhaps much more than you are used to, or have seen before. Because of
248 this, you might want to do a little background reading in the
249 techniques used and capabilities of the library. There are many good
250 pages that discuss the STL, and several books on the subject that you
251 can get, so it will not be discussed in this document.</p>
253 <p>Here are some useful links:</p>
257 <li><a href="http://www.dinkumware.com/refxcpp.html">Dinkumware C++ Library
258 reference</a> - an excellent reference for the STL and other parts of the
259 standard C++ library.</li>
261 <li><a href="http://www.tempest-sw.com/cpp/">C++ In a Nutshell</a> - This is an
262 O'Reilly book in the making. It has a decent Standard Library
263 Reference that rivals Dinkumware's, and is unfortunately no longer free since the
264 book has been published.</li>
266 <li><a href="http://www.parashift.com/c++-faq-lite/">C++ Frequently Asked
269 <li><a href="http://www.sgi.com/tech/stl/">SGI's STL Programmer's Guide</a> -
271 href="http://www.sgi.com/tech/stl/stl_introduction.html">Introduction to the
274 <li><a href="http://www.research.att.com/%7Ebs/C++.html">Bjarne Stroustrup's C++
277 <li><a href="http://64.78.49.204/">
278 Bruce Eckel's Thinking in C++, 2nd ed. Volume 2 Revision 4.0 (even better, get
283 <p>You are also encouraged to take a look at the <a
284 href="CodingStandards.html">LLVM Coding Standards</a> guide which focuses on how
285 to write maintainable code more than where to put your curly braces.</p>
289 <!-- ======================================================================= -->
290 <div class="doc_subsection">
291 <a name="stl">Other useful references</a>
294 <div class="doc_text">
297 <li><a href="http://www.psc.edu/%7Esemke/cvs_branches.html">CVS
298 Branch and Tag Primer</a></li>
299 <li><a href="http://www.fortran-2000.com/ArnaudRecipes/sharedlib.html">Using
300 static and shared libraries across platforms</a></li>
305 <!-- *********************************************************************** -->
306 <div class="doc_section">
307 <a name="apis">Important and useful LLVM APIs</a>
309 <!-- *********************************************************************** -->
311 <div class="doc_text">
313 <p>Here we highlight some LLVM APIs that are generally useful and good to
314 know about when writing transformations.</p>
318 <!-- ======================================================================= -->
319 <div class="doc_subsection">
320 <a name="isa">The <tt>isa<></tt>, <tt>cast<></tt> and
321 <tt>dyn_cast<></tt> templates</a>
324 <div class="doc_text">
326 <p>The LLVM source-base makes extensive use of a custom form of RTTI.
327 These templates have many similarities to the C++ <tt>dynamic_cast<></tt>
328 operator, but they don't have some drawbacks (primarily stemming from
329 the fact that <tt>dynamic_cast<></tt> only works on classes that
330 have a v-table). Because they are used so often, you must know what they
331 do and how they work. All of these templates are defined in the <a
332 href="/doxygen/Casting_8h-source.html"><tt>llvm/Support/Casting.h</tt></a>
333 file (note that you very rarely have to include this file directly).</p>
336 <dt><tt>isa<></tt>: </dt>
338 <dd><p>The <tt>isa<></tt> operator works exactly like the Java
339 "<tt>instanceof</tt>" operator. It returns true or false depending on whether
340 a reference or pointer points to an instance of the specified class. This can
341 be very useful for constraint checking of various sorts (example below).</p>
344 <dt><tt>cast<></tt>: </dt>
346 <dd><p>The <tt>cast<></tt> operator is a "checked cast" operation. It
347 converts a pointer or reference from a base class to a derived class, causing
348 an assertion failure if it is not really an instance of the right type. This
349 should be used in cases where you have some information that makes you believe
350 that something is of the right type. An example of the <tt>isa<></tt>
351 and <tt>cast<></tt> template is:</p>
353 <div class="doc_code">
355 static bool isLoopInvariant(const <a href="#Value">Value</a> *V, const Loop *L) {
356 if (isa<<a href="#Constant">Constant</a>>(V) || isa<<a href="#Argument">Argument</a>>(V) || isa<<a href="#GlobalValue">GlobalValue</a>>(V))
359 // <i>Otherwise, it must be an instruction...</i>
360 return !L->contains(cast<<a href="#Instruction">Instruction</a>>(V)->getParent());
365 <p>Note that you should <b>not</b> use an <tt>isa<></tt> test followed
366 by a <tt>cast<></tt>, for that use the <tt>dyn_cast<></tt>
371 <dt><tt>dyn_cast<></tt>:</dt>
373 <dd><p>The <tt>dyn_cast<></tt> operator is a "checking cast" operation.
374 It checks to see if the operand is of the specified type, and if so, returns a
375 pointer to it (this operator does not work with references). If the operand is
376 not of the correct type, a null pointer is returned. Thus, this works very
377 much like the <tt>dynamic_cast<></tt> operator in C++, and should be
378 used in the same circumstances. Typically, the <tt>dyn_cast<></tt>
379 operator is used in an <tt>if</tt> statement or some other flow control
380 statement like this:</p>
382 <div class="doc_code">
384 if (<a href="#AllocationInst">AllocationInst</a> *AI = dyn_cast<<a href="#AllocationInst">AllocationInst</a>>(Val)) {
390 <p>This form of the <tt>if</tt> statement effectively combines together a call
391 to <tt>isa<></tt> and a call to <tt>cast<></tt> into one
392 statement, which is very convenient.</p>
394 <p>Note that the <tt>dyn_cast<></tt> operator, like C++'s
395 <tt>dynamic_cast<></tt> or Java's <tt>instanceof</tt> operator, can be
396 abused. In particular, you should not use big chained <tt>if/then/else</tt>
397 blocks to check for lots of different variants of classes. If you find
398 yourself wanting to do this, it is much cleaner and more efficient to use the
399 <tt>InstVisitor</tt> class to dispatch over the instruction type directly.</p>
403 <dt><tt>cast_or_null<></tt>: </dt>
405 <dd><p>The <tt>cast_or_null<></tt> operator works just like the
406 <tt>cast<></tt> operator, except that it allows for a null pointer as an
407 argument (which it then propagates). This can sometimes be useful, allowing
408 you to combine several null checks into one.</p></dd>
410 <dt><tt>dyn_cast_or_null<></tt>: </dt>
412 <dd><p>The <tt>dyn_cast_or_null<></tt> operator works just like the
413 <tt>dyn_cast<></tt> operator, except that it allows for a null pointer
414 as an argument (which it then propagates). This can sometimes be useful,
415 allowing you to combine several null checks into one.</p></dd>
419 <p>These five templates can be used with any classes, whether they have a
420 v-table or not. To add support for these templates, you simply need to add
421 <tt>classof</tt> static methods to the class you are interested casting
422 to. Describing this is currently outside the scope of this document, but there
423 are lots of examples in the LLVM source base.</p>
427 <!-- ======================================================================= -->
428 <div class="doc_subsection">
429 <a name="DEBUG">The <tt>DEBUG()</tt> macro and <tt>-debug</tt> option</a>
432 <div class="doc_text">
434 <p>Often when working on your pass you will put a bunch of debugging printouts
435 and other code into your pass. After you get it working, you want to remove
436 it, but you may need it again in the future (to work out new bugs that you run
439 <p> Naturally, because of this, you don't want to delete the debug printouts,
440 but you don't want them to always be noisy. A standard compromise is to comment
441 them out, allowing you to enable them if you need them in the future.</p>
443 <p>The "<tt><a href="/doxygen/Debug_8h-source.html">llvm/Support/Debug.h</a></tt>"
444 file provides a macro named <tt>DEBUG()</tt> that is a much nicer solution to
445 this problem. Basically, you can put arbitrary code into the argument of the
446 <tt>DEBUG</tt> macro, and it is only executed if '<tt>opt</tt>' (or any other
447 tool) is run with the '<tt>-debug</tt>' command line argument:</p>
449 <div class="doc_code">
451 DOUT << "I am here!\n";
455 <p>Then you can run your pass like this:</p>
457 <div class="doc_code">
459 $ opt < a.bc > /dev/null -mypass
460 <i><no output></i>
461 $ opt < a.bc > /dev/null -mypass -debug
466 <p>Using the <tt>DEBUG()</tt> macro instead of a home-brewed solution allows you
467 to not have to create "yet another" command line option for the debug output for
468 your pass. Note that <tt>DEBUG()</tt> macros are disabled for optimized builds,
469 so they do not cause a performance impact at all (for the same reason, they
470 should also not contain side-effects!).</p>
472 <p>One additional nice thing about the <tt>DEBUG()</tt> macro is that you can
473 enable or disable it directly in gdb. Just use "<tt>set DebugFlag=0</tt>" or
474 "<tt>set DebugFlag=1</tt>" from the gdb if the program is running. If the
475 program hasn't been started yet, you can always just run it with
480 <!-- _______________________________________________________________________ -->
481 <div class="doc_subsubsection">
482 <a name="DEBUG_TYPE">Fine grained debug info with <tt>DEBUG_TYPE</tt> and
483 the <tt>-debug-only</tt> option</a>
486 <div class="doc_text">
488 <p>Sometimes you may find yourself in a situation where enabling <tt>-debug</tt>
489 just turns on <b>too much</b> information (such as when working on the code
490 generator). If you want to enable debug information with more fine-grained
491 control, you define the <tt>DEBUG_TYPE</tt> macro and the <tt>-debug</tt> only
492 option as follows:</p>
494 <div class="doc_code">
496 DOUT << "No debug type\n";
498 #define DEBUG_TYPE "foo"
499 DOUT << "'foo' debug type\n";
501 #define DEBUG_TYPE "bar"
502 DOUT << "'bar' debug type\n";
504 #define DEBUG_TYPE ""
505 DOUT << "No debug type (2)\n";
509 <p>Then you can run your pass like this:</p>
511 <div class="doc_code">
513 $ opt < a.bc > /dev/null -mypass
514 <i><no output></i>
515 $ opt < a.bc > /dev/null -mypass -debug
520 $ opt < a.bc > /dev/null -mypass -debug-only=foo
522 $ opt < a.bc > /dev/null -mypass -debug-only=bar
527 <p>Of course, in practice, you should only set <tt>DEBUG_TYPE</tt> at the top of
528 a file, to specify the debug type for the entire module (if you do this before
529 you <tt>#include "llvm/Support/Debug.h"</tt>, you don't have to insert the ugly
530 <tt>#undef</tt>'s). Also, you should use names more meaningful than "foo" and
531 "bar", because there is no system in place to ensure that names do not
532 conflict. If two different modules use the same string, they will all be turned
533 on when the name is specified. This allows, for example, all debug information
534 for instruction scheduling to be enabled with <tt>-debug-type=InstrSched</tt>,
535 even if the source lives in multiple files.</p>
539 <!-- ======================================================================= -->
540 <div class="doc_subsection">
541 <a name="Statistic">The <tt>Statistic</tt> class & <tt>-stats</tt>
545 <div class="doc_text">
548 href="/doxygen/Statistic_8h-source.html">llvm/ADT/Statistic.h</a></tt>" file
549 provides a class named <tt>Statistic</tt> that is used as a unified way to
550 keep track of what the LLVM compiler is doing and how effective various
551 optimizations are. It is useful to see what optimizations are contributing to
552 making a particular program run faster.</p>
554 <p>Often you may run your pass on some big program, and you're interested to see
555 how many times it makes a certain transformation. Although you can do this with
556 hand inspection, or some ad-hoc method, this is a real pain and not very useful
557 for big programs. Using the <tt>Statistic</tt> class makes it very easy to
558 keep track of this information, and the calculated information is presented in a
559 uniform manner with the rest of the passes being executed.</p>
561 <p>There are many examples of <tt>Statistic</tt> uses, but the basics of using
562 it are as follows:</p>
565 <li><p>Define your statistic like this:</p>
567 <div class="doc_code">
569 #define <a href="#DEBUG_TYPE">DEBUG_TYPE</a> "mypassname" <i>// This goes before any #includes.</i>
570 STATISTIC(NumXForms, "The # of times I did stuff");
574 <p>The <tt>STATISTIC</tt> macro defines a static variable, whose name is
575 specified by the first argument. The pass name is taken from the DEBUG_TYPE
576 macro, and the description is taken from the second argument. The variable
577 defined ("NumXForms" in this case) acts like an unsigned integer.</p></li>
579 <li><p>Whenever you make a transformation, bump the counter:</p>
581 <div class="doc_code">
583 ++NumXForms; // <i>I did stuff!</i>
590 <p>That's all you have to do. To get '<tt>opt</tt>' to print out the
591 statistics gathered, use the '<tt>-stats</tt>' option:</p>
593 <div class="doc_code">
595 $ opt -stats -mypassname < program.bc > /dev/null
596 <i>... statistics output ...</i>
600 <p> When running <tt>opt</tt> on a C file from the SPEC benchmark
601 suite, it gives a report that looks like this:</p>
603 <div class="doc_code">
605 7646 bitcodewriter - Number of normal instructions
606 725 bitcodewriter - Number of oversized instructions
607 129996 bitcodewriter - Number of bitcode bytes written
608 2817 raise - Number of insts DCEd or constprop'd
609 3213 raise - Number of cast-of-self removed
610 5046 raise - Number of expression trees converted
611 75 raise - Number of other getelementptr's formed
612 138 raise - Number of load/store peepholes
613 42 deadtypeelim - Number of unused typenames removed from symtab
614 392 funcresolve - Number of varargs functions resolved
615 27 globaldce - Number of global variables removed
616 2 adce - Number of basic blocks removed
617 134 cee - Number of branches revectored
618 49 cee - Number of setcc instruction eliminated
619 532 gcse - Number of loads removed
620 2919 gcse - Number of instructions removed
621 86 indvars - Number of canonical indvars added
622 87 indvars - Number of aux indvars removed
623 25 instcombine - Number of dead inst eliminate
624 434 instcombine - Number of insts combined
625 248 licm - Number of load insts hoisted
626 1298 licm - Number of insts hoisted to a loop pre-header
627 3 licm - Number of insts hoisted to multiple loop preds (bad, no loop pre-header)
628 75 mem2reg - Number of alloca's promoted
629 1444 cfgsimplify - Number of blocks simplified
633 <p>Obviously, with so many optimizations, having a unified framework for this
634 stuff is very nice. Making your pass fit well into the framework makes it more
635 maintainable and useful.</p>
639 <!-- ======================================================================= -->
640 <div class="doc_subsection">
641 <a name="ViewGraph">Viewing graphs while debugging code</a>
644 <div class="doc_text">
646 <p>Several of the important data structures in LLVM are graphs: for example
647 CFGs made out of LLVM <a href="#BasicBlock">BasicBlock</a>s, CFGs made out of
648 LLVM <a href="CodeGenerator.html#machinebasicblock">MachineBasicBlock</a>s, and
649 <a href="CodeGenerator.html#selectiondag_intro">Instruction Selection
650 DAGs</a>. In many cases, while debugging various parts of the compiler, it is
651 nice to instantly visualize these graphs.</p>
653 <p>LLVM provides several callbacks that are available in a debug build to do
654 exactly that. If you call the <tt>Function::viewCFG()</tt> method, for example,
655 the current LLVM tool will pop up a window containing the CFG for the function
656 where each basic block is a node in the graph, and each node contains the
657 instructions in the block. Similarly, there also exists
658 <tt>Function::viewCFGOnly()</tt> (does not include the instructions), the
659 <tt>MachineFunction::viewCFG()</tt> and <tt>MachineFunction::viewCFGOnly()</tt>,
660 and the <tt>SelectionDAG::viewGraph()</tt> methods. Within GDB, for example,
661 you can usually use something like <tt>call DAG.viewGraph()</tt> to pop
662 up a window. Alternatively, you can sprinkle calls to these functions in your
663 code in places you want to debug.</p>
665 <p>Getting this to work requires a small amount of configuration. On Unix
666 systems with X11, install the <a href="http://www.graphviz.org">graphviz</a>
667 toolkit, and make sure 'dot' and 'gv' are in your path. If you are running on
668 Mac OS/X, download and install the Mac OS/X <a
669 href="http://www.pixelglow.com/graphviz/">Graphviz program</a>, and add
670 <tt>/Applications/Graphviz.app/Contents/MacOS/</tt> (or wherever you install
671 it) to your path. Once in your system and path are set up, rerun the LLVM
672 configure script and rebuild LLVM to enable this functionality.</p>
674 <p><tt>SelectionDAG</tt> has been extended to make it easier to locate
675 <i>interesting</i> nodes in large complex graphs. From gdb, if you
676 <tt>call DAG.setGraphColor(<i>node</i>, "<i>color</i>")</tt>, then the
677 next <tt>call DAG.viewGraph()</tt> would highlight the node in the
678 specified color (choices of colors can be found at <a
679 href="http://www.graphviz.org/doc/info/colors.html">colors</a>.) More
680 complex node attributes can be provided with <tt>call
681 DAG.setGraphAttrs(<i>node</i>, "<i>attributes</i>")</tt> (choices can be
682 found at <a href="http://www.graphviz.org/doc/info/attrs.html">Graph
683 Attributes</a>.) If you want to restart and clear all the current graph
684 attributes, then you can <tt>call DAG.clearGraphAttrs()</tt>. </p>
688 <!-- *********************************************************************** -->
689 <div class="doc_section">
690 <a name="datastructure">Picking the Right Data Structure for a Task</a>
692 <!-- *********************************************************************** -->
694 <div class="doc_text">
696 <p>LLVM has a plethora of data structures in the <tt>llvm/ADT/</tt> directory,
697 and we commonly use STL data structures. This section describes the trade-offs
698 you should consider when you pick one.</p>
701 The first step is a choose your own adventure: do you want a sequential
702 container, a set-like container, or a map-like container? The most important
703 thing when choosing a container is the algorithmic properties of how you plan to
704 access the container. Based on that, you should use:</p>
707 <li>a <a href="#ds_map">map-like</a> container if you need efficient look-up
708 of an value based on another value. Map-like containers also support
709 efficient queries for containment (whether a key is in the map). Map-like
710 containers generally do not support efficient reverse mapping (values to
711 keys). If you need that, use two maps. Some map-like containers also
712 support efficient iteration through the keys in sorted order. Map-like
713 containers are the most expensive sort, only use them if you need one of
714 these capabilities.</li>
716 <li>a <a href="#ds_set">set-like</a> container if you need to put a bunch of
717 stuff into a container that automatically eliminates duplicates. Some
718 set-like containers support efficient iteration through the elements in
719 sorted order. Set-like containers are more expensive than sequential
723 <li>a <a href="#ds_sequential">sequential</a> container provides
724 the most efficient way to add elements and keeps track of the order they are
725 added to the collection. They permit duplicates and support efficient
726 iteration, but do not support efficient look-up based on a key.
729 <li>a <a href="#ds_bit">bit</a> container provides an efficient way to store and
730 perform set operations on sets of numeric id's, while automatically
731 eliminating duplicates. Bit containers require a maximum of 1 bit for each
732 identifier you want to store.
737 Once the proper category of container is determined, you can fine tune the
738 memory use, constant factors, and cache behaviors of access by intelligently
739 picking a member of the category. Note that constant factors and cache behavior
740 can be a big deal. If you have a vector that usually only contains a few
741 elements (but could contain many), for example, it's much better to use
742 <a href="#dss_smallvector">SmallVector</a> than <a href="#dss_vector">vector</a>
743 . Doing so avoids (relatively) expensive malloc/free calls, which dwarf the
744 cost of adding the elements to the container. </p>
748 <!-- ======================================================================= -->
749 <div class="doc_subsection">
750 <a name="ds_sequential">Sequential Containers (std::vector, std::list, etc)</a>
753 <div class="doc_text">
754 There are a variety of sequential containers available for you, based on your
755 needs. Pick the first in this section that will do what you want.
758 <!-- _______________________________________________________________________ -->
759 <div class="doc_subsubsection">
760 <a name="dss_fixedarrays">Fixed Size Arrays</a>
763 <div class="doc_text">
764 <p>Fixed size arrays are very simple and very fast. They are good if you know
765 exactly how many elements you have, or you have a (low) upper bound on how many
769 <!-- _______________________________________________________________________ -->
770 <div class="doc_subsubsection">
771 <a name="dss_heaparrays">Heap Allocated Arrays</a>
774 <div class="doc_text">
775 <p>Heap allocated arrays (new[] + delete[]) are also simple. They are good if
776 the number of elements is variable, if you know how many elements you will need
777 before the array is allocated, and if the array is usually large (if not,
778 consider a <a href="#dss_smallvector">SmallVector</a>). The cost of a heap
779 allocated array is the cost of the new/delete (aka malloc/free). Also note that
780 if you are allocating an array of a type with a constructor, the constructor and
781 destructors will be run for every element in the array (re-sizable vectors only
782 construct those elements actually used).</p>
785 <!-- _______________________________________________________________________ -->
786 <div class="doc_subsubsection">
787 <a name="dss_smallvector">"llvm/ADT/SmallVector.h"</a>
790 <div class="doc_text">
791 <p><tt>SmallVector<Type, N></tt> is a simple class that looks and smells
792 just like <tt>vector<Type></tt>:
793 it supports efficient iteration, lays out elements in memory order (so you can
794 do pointer arithmetic between elements), supports efficient push_back/pop_back
795 operations, supports efficient random access to its elements, etc.</p>
797 <p>The advantage of SmallVector is that it allocates space for
798 some number of elements (N) <b>in the object itself</b>. Because of this, if
799 the SmallVector is dynamically smaller than N, no malloc is performed. This can
800 be a big win in cases where the malloc/free call is far more expensive than the
801 code that fiddles around with the elements.</p>
803 <p>This is good for vectors that are "usually small" (e.g. the number of
804 predecessors/successors of a block is usually less than 8). On the other hand,
805 this makes the size of the SmallVector itself large, so you don't want to
806 allocate lots of them (doing so will waste a lot of space). As such,
807 SmallVectors are most useful when on the stack.</p>
809 <p>SmallVector also provides a nice portable and efficient replacement for
814 <!-- _______________________________________________________________________ -->
815 <div class="doc_subsubsection">
816 <a name="dss_vector"><vector></a>
819 <div class="doc_text">
821 std::vector is well loved and respected. It is useful when SmallVector isn't:
822 when the size of the vector is often large (thus the small optimization will
823 rarely be a benefit) or if you will be allocating many instances of the vector
824 itself (which would waste space for elements that aren't in the container).
825 vector is also useful when interfacing with code that expects vectors :).
828 <p>One worthwhile note about std::vector: avoid code like this:</p>
830 <div class="doc_code">
833 std::vector<foo> V;
839 <p>Instead, write this as:</p>
841 <div class="doc_code">
843 std::vector<foo> V;
851 <p>Doing so will save (at least) one heap allocation and free per iteration of
856 <!-- _______________________________________________________________________ -->
857 <div class="doc_subsubsection">
858 <a name="dss_deque"><deque></a>
861 <div class="doc_text">
862 <p>std::deque is, in some senses, a generalized version of std::vector. Like
863 std::vector, it provides constant time random access and other similar
864 properties, but it also provides efficient access to the front of the list. It
865 does not guarantee continuity of elements within memory.</p>
867 <p>In exchange for this extra flexibility, std::deque has significantly higher
868 constant factor costs than std::vector. If possible, use std::vector or
869 something cheaper.</p>
872 <!-- _______________________________________________________________________ -->
873 <div class="doc_subsubsection">
874 <a name="dss_list"><list></a>
877 <div class="doc_text">
878 <p>std::list is an extremely inefficient class that is rarely useful.
879 It performs a heap allocation for every element inserted into it, thus having an
880 extremely high constant factor, particularly for small data types. std::list
881 also only supports bidirectional iteration, not random access iteration.</p>
883 <p>In exchange for this high cost, std::list supports efficient access to both
884 ends of the list (like std::deque, but unlike std::vector or SmallVector). In
885 addition, the iterator invalidation characteristics of std::list are stronger
886 than that of a vector class: inserting or removing an element into the list does
887 not invalidate iterator or pointers to other elements in the list.</p>
890 <!-- _______________________________________________________________________ -->
891 <div class="doc_subsubsection">
892 <a name="dss_ilist">llvm/ADT/ilist.h</a>
895 <div class="doc_text">
896 <p><tt>ilist<T></tt> implements an 'intrusive' doubly-linked list. It is
897 intrusive, because it requires the element to store and provide access to the
898 prev/next pointers for the list.</p>
900 <p><tt>ilist</tt> has the same drawbacks as <tt>std::list</tt>, and additionally
901 requires an <tt>ilist_traits</tt> implementation for the element type, but it
902 provides some novel characteristics. In particular, it can efficiently store
903 polymorphic objects, the traits class is informed when an element is inserted or
904 removed from the list, and <tt>ilist</tt>s are guaranteed to support a
905 constant-time splice operation.</p>
907 <p>These properties are exactly what we want for things like
908 <tt>Instruction</tt>s and basic blocks, which is why these are implemented with
911 Related classes of interest are explained in the following subsections:
913 <li><a href="#dss_ilist_traits">ilist_traits</a></li>
914 <li><a href="#dss_iplist">iplist</a></li>
915 <li><a href="#dss_ilist_node">llvm/ADT/ilist_node.h</a></li>
916 <li><a href="#dss_ilist_sentinel">Sentinels</a></li>
920 <!-- _______________________________________________________________________ -->
921 <div class="doc_subsubsection">
922 <a name="dss_ilist_traits">ilist_traits</a>
925 <div class="doc_text">
926 <p><tt>ilist_traits<T></tt> is <tt>ilist<T></tt>'s customization
927 mechanism. <tt>iplist<T></tt> (and consequently <tt>ilist<T></tt>)
928 publicly derive from this traits class.</p>
931 <!-- _______________________________________________________________________ -->
932 <div class="doc_subsubsection">
933 <a name="dss_iplist">iplist</a>
936 <div class="doc_text">
937 <p><tt>iplist<T></tt> is <tt>ilist<T></tt>'s base and as such
938 supports a slightly narrower interface. Notably, inserters from
939 <tt>T&</tt> are absent.</p>
941 <p><tt>ilist_traits<T></tt> is a public base of this class and can be
942 used for a wide variety of customizations.</p>
945 <!-- _______________________________________________________________________ -->
946 <div class="doc_subsubsection">
947 <a name="dss_ilist_node">llvm/ADT/ilist_node.h</a>
950 <div class="doc_text">
951 <p><tt>ilist_node<T></tt> implements a the forward and backward links
952 that are expected by the <tt>ilist<T></tt> (and analogous containers)
953 in the default manner.</p>
955 <p><tt>ilist_node<T></tt>s are meant to be embedded in the node type
956 <tt>T</tt>, usually <tt>T</tt> publicly derives from
957 <tt>ilist_node<T></tt>.</p>
960 <!-- _______________________________________________________________________ -->
961 <div class="doc_subsubsection">
962 <a name="dss_ilist_sentinel">Sentinels</a>
965 <div class="doc_text">
966 <p><tt>ilist</tt>s have another speciality that must be considered. To be a good
967 citizen in the C++ ecosystem, it needs to support the standard container
968 operations, such as <tt>begin</tt> and <tt>end</tt> iterators, etc. Also, the
969 <tt>operator--</tt> must work correctly on the <tt>end</tt> iterator in the
970 case of non-empty <tt>ilist</tt>s.</p>
972 <p>The only sensible solution to this problem is to allocate a so-called
973 <i>sentinel</i> along with the intrusive list, which serves as the <tt>end</tt>
974 iterator, providing the back-link to the last element. However conforming to the
975 C++ convention it is illegal to <tt>operator++</tt> beyond the sentinel and it
976 also must not be dereferenced.</p>
978 <p>These constraints allow for some implementation freedom to the <tt>ilist</tt>
979 how to allocate and store the sentinel. The corresponding policy is dictated
980 by <tt>ilist_traits<T></tt>. By default a <tt>T</tt> gets heap-allocated
981 whenever the need for a sentinel arises.</p>
983 <p>While the default policy is sufficient in most cases, it may break down when
984 <tt>T</tt> does not provide a default constructor. Also, in the case of many
985 instances of <tt>ilist</tt>s, the memory overhead of the associated sentinels
986 is wasted. To alleviate the situation with numerous and voluminous
987 <tt>T</tt>-sentinels, sometimes a trick is employed, leading to <i>ghostly
990 <p>Ghostly sentinels are obtained by specially-crafted <tt>ilist_traits<T></tt>
991 which superpose the sentinel with the <tt>ilist</tt> instance in memory. Pointer
992 arithmetic is used to obtain the sentinel, which is relative to the
993 <tt>ilist</tt>'s <tt>this</tt> pointer. The <tt>ilist</tt> is augmented by an
994 extra pointer, which serves as the back-link of the sentinel. This is the only
995 field in the ghostly sentinel which can be legally accessed.</p>
998 <!-- _______________________________________________________________________ -->
999 <div class="doc_subsubsection">
1000 <a name="dss_other">Other Sequential Container options</a>
1003 <div class="doc_text">
1004 <p>Other STL containers are available, such as std::string.</p>
1006 <p>There are also various STL adapter classes such as std::queue,
1007 std::priority_queue, std::stack, etc. These provide simplified access to an
1008 underlying container but don't affect the cost of the container itself.</p>
1013 <!-- ======================================================================= -->
1014 <div class="doc_subsection">
1015 <a name="ds_set">Set-Like Containers (std::set, SmallSet, SetVector, etc)</a>
1018 <div class="doc_text">
1020 <p>Set-like containers are useful when you need to canonicalize multiple values
1021 into a single representation. There are several different choices for how to do
1022 this, providing various trade-offs.</p>
1027 <!-- _______________________________________________________________________ -->
1028 <div class="doc_subsubsection">
1029 <a name="dss_sortedvectorset">A sorted 'vector'</a>
1032 <div class="doc_text">
1034 <p>If you intend to insert a lot of elements, then do a lot of queries, a
1035 great approach is to use a vector (or other sequential container) with
1036 std::sort+std::unique to remove duplicates. This approach works really well if
1037 your usage pattern has these two distinct phases (insert then query), and can be
1038 coupled with a good choice of <a href="#ds_sequential">sequential container</a>.
1042 This combination provides the several nice properties: the result data is
1043 contiguous in memory (good for cache locality), has few allocations, is easy to
1044 address (iterators in the final vector are just indices or pointers), and can be
1045 efficiently queried with a standard binary or radix search.</p>
1049 <!-- _______________________________________________________________________ -->
1050 <div class="doc_subsubsection">
1051 <a name="dss_smallset">"llvm/ADT/SmallSet.h"</a>
1054 <div class="doc_text">
1056 <p>If you have a set-like data structure that is usually small and whose elements
1057 are reasonably small, a <tt>SmallSet<Type, N></tt> is a good choice. This set
1058 has space for N elements in place (thus, if the set is dynamically smaller than
1059 N, no malloc traffic is required) and accesses them with a simple linear search.
1060 When the set grows beyond 'N' elements, it allocates a more expensive representation that
1061 guarantees efficient access (for most types, it falls back to std::set, but for
1062 pointers it uses something far better, <a
1063 href="#dss_smallptrset">SmallPtrSet</a>).</p>
1065 <p>The magic of this class is that it handles small sets extremely efficiently,
1066 but gracefully handles extremely large sets without loss of efficiency. The
1067 drawback is that the interface is quite small: it supports insertion, queries
1068 and erasing, but does not support iteration.</p>
1072 <!-- _______________________________________________________________________ -->
1073 <div class="doc_subsubsection">
1074 <a name="dss_smallptrset">"llvm/ADT/SmallPtrSet.h"</a>
1077 <div class="doc_text">
1079 <p>SmallPtrSet has all the advantages of SmallSet (and a SmallSet of pointers is
1080 transparently implemented with a SmallPtrSet), but also supports iterators. If
1081 more than 'N' insertions are performed, a single quadratically
1082 probed hash table is allocated and grows as needed, providing extremely
1083 efficient access (constant time insertion/deleting/queries with low constant
1084 factors) and is very stingy with malloc traffic.</p>
1086 <p>Note that, unlike std::set, the iterators of SmallPtrSet are invalidated
1087 whenever an insertion occurs. Also, the values visited by the iterators are not
1088 visited in sorted order.</p>
1092 <!-- _______________________________________________________________________ -->
1093 <div class="doc_subsubsection">
1094 <a name="dss_denseset">"llvm/ADT/DenseSet.h"</a>
1097 <div class="doc_text">
1100 DenseSet is a simple quadratically probed hash table. It excels at supporting
1101 small values: it uses a single allocation to hold all of the pairs that
1102 are currently inserted in the set. DenseSet is a great way to unique small
1103 values that are not simple pointers (use <a
1104 href="#dss_smallptrset">SmallPtrSet</a> for pointers). Note that DenseSet has
1105 the same requirements for the value type that <a
1106 href="#dss_densemap">DenseMap</a> has.
1111 <!-- _______________________________________________________________________ -->
1112 <div class="doc_subsubsection">
1113 <a name="dss_FoldingSet">"llvm/ADT/FoldingSet.h"</a>
1116 <div class="doc_text">
1119 FoldingSet is an aggregate class that is really good at uniquing
1120 expensive-to-create or polymorphic objects. It is a combination of a chained
1121 hash table with intrusive links (uniqued objects are required to inherit from
1122 FoldingSetNode) that uses <a href="#dss_smallvector">SmallVector</a> as part of
1125 <p>Consider a case where you want to implement a "getOrCreateFoo" method for
1126 a complex object (for example, a node in the code generator). The client has a
1127 description of *what* it wants to generate (it knows the opcode and all the
1128 operands), but we don't want to 'new' a node, then try inserting it into a set
1129 only to find out it already exists, at which point we would have to delete it
1130 and return the node that already exists.
1133 <p>To support this style of client, FoldingSet perform a query with a
1134 FoldingSetNodeID (which wraps SmallVector) that can be used to describe the
1135 element that we want to query for. The query either returns the element
1136 matching the ID or it returns an opaque ID that indicates where insertion should
1137 take place. Construction of the ID usually does not require heap traffic.</p>
1139 <p>Because FoldingSet uses intrusive links, it can support polymorphic objects
1140 in the set (for example, you can have SDNode instances mixed with LoadSDNodes).
1141 Because the elements are individually allocated, pointers to the elements are
1142 stable: inserting or removing elements does not invalidate any pointers to other
1148 <!-- _______________________________________________________________________ -->
1149 <div class="doc_subsubsection">
1150 <a name="dss_set"><set></a>
1153 <div class="doc_text">
1155 <p><tt>std::set</tt> is a reasonable all-around set class, which is decent at
1156 many things but great at nothing. std::set allocates memory for each element
1157 inserted (thus it is very malloc intensive) and typically stores three pointers
1158 per element in the set (thus adding a large amount of per-element space
1159 overhead). It offers guaranteed log(n) performance, which is not particularly
1160 fast from a complexity standpoint (particularly if the elements of the set are
1161 expensive to compare, like strings), and has extremely high constant factors for
1162 lookup, insertion and removal.</p>
1164 <p>The advantages of std::set are that its iterators are stable (deleting or
1165 inserting an element from the set does not affect iterators or pointers to other
1166 elements) and that iteration over the set is guaranteed to be in sorted order.
1167 If the elements in the set are large, then the relative overhead of the pointers
1168 and malloc traffic is not a big deal, but if the elements of the set are small,
1169 std::set is almost never a good choice.</p>
1173 <!-- _______________________________________________________________________ -->
1174 <div class="doc_subsubsection">
1175 <a name="dss_setvector">"llvm/ADT/SetVector.h"</a>
1178 <div class="doc_text">
1179 <p>LLVM's SetVector<Type> is an adapter class that combines your choice of
1180 a set-like container along with a <a href="#ds_sequential">Sequential
1181 Container</a>. The important property
1182 that this provides is efficient insertion with uniquing (duplicate elements are
1183 ignored) with iteration support. It implements this by inserting elements into
1184 both a set-like container and the sequential container, using the set-like
1185 container for uniquing and the sequential container for iteration.
1188 <p>The difference between SetVector and other sets is that the order of
1189 iteration is guaranteed to match the order of insertion into the SetVector.
1190 This property is really important for things like sets of pointers. Because
1191 pointer values are non-deterministic (e.g. vary across runs of the program on
1192 different machines), iterating over the pointers in the set will
1193 not be in a well-defined order.</p>
1196 The drawback of SetVector is that it requires twice as much space as a normal
1197 set and has the sum of constant factors from the set-like container and the
1198 sequential container that it uses. Use it *only* if you need to iterate over
1199 the elements in a deterministic order. SetVector is also expensive to delete
1200 elements out of (linear time), unless you use it's "pop_back" method, which is
1204 <p>SetVector is an adapter class that defaults to using std::vector and std::set
1205 for the underlying containers, so it is quite expensive. However,
1206 <tt>"llvm/ADT/SetVector.h"</tt> also provides a SmallSetVector class, which
1207 defaults to using a SmallVector and SmallSet of a specified size. If you use
1208 this, and if your sets are dynamically smaller than N, you will save a lot of
1213 <!-- _______________________________________________________________________ -->
1214 <div class="doc_subsubsection">
1215 <a name="dss_uniquevector">"llvm/ADT/UniqueVector.h"</a>
1218 <div class="doc_text">
1221 UniqueVector is similar to <a href="#dss_setvector">SetVector</a>, but it
1222 retains a unique ID for each element inserted into the set. It internally
1223 contains a map and a vector, and it assigns a unique ID for each value inserted
1226 <p>UniqueVector is very expensive: its cost is the sum of the cost of
1227 maintaining both the map and vector, it has high complexity, high constant
1228 factors, and produces a lot of malloc traffic. It should be avoided.</p>
1233 <!-- _______________________________________________________________________ -->
1234 <div class="doc_subsubsection">
1235 <a name="dss_otherset">Other Set-Like Container Options</a>
1238 <div class="doc_text">
1241 The STL provides several other options, such as std::multiset and the various
1242 "hash_set" like containers (whether from C++ TR1 or from the SGI library). We
1243 never use hash_set and unordered_set because they are generally very expensive
1244 (each insertion requires a malloc) and very non-portable.
1247 <p>std::multiset is useful if you're not interested in elimination of
1248 duplicates, but has all the drawbacks of std::set. A sorted vector (where you
1249 don't delete duplicate entries) or some other approach is almost always
1254 <!-- ======================================================================= -->
1255 <div class="doc_subsection">
1256 <a name="ds_map">Map-Like Containers (std::map, DenseMap, etc)</a>
1259 <div class="doc_text">
1260 Map-like containers are useful when you want to associate data to a key. As
1261 usual, there are a lot of different ways to do this. :)
1264 <!-- _______________________________________________________________________ -->
1265 <div class="doc_subsubsection">
1266 <a name="dss_sortedvectormap">A sorted 'vector'</a>
1269 <div class="doc_text">
1272 If your usage pattern follows a strict insert-then-query approach, you can
1273 trivially use the same approach as <a href="#dss_sortedvectorset">sorted vectors
1274 for set-like containers</a>. The only difference is that your query function
1275 (which uses std::lower_bound to get efficient log(n) lookup) should only compare
1276 the key, not both the key and value. This yields the same advantages as sorted
1281 <!-- _______________________________________________________________________ -->
1282 <div class="doc_subsubsection">
1283 <a name="dss_stringmap">"llvm/ADT/StringMap.h"</a>
1286 <div class="doc_text">
1289 Strings are commonly used as keys in maps, and they are difficult to support
1290 efficiently: they are variable length, inefficient to hash and compare when
1291 long, expensive to copy, etc. StringMap is a specialized container designed to
1292 cope with these issues. It supports mapping an arbitrary range of bytes to an
1293 arbitrary other object.</p>
1295 <p>The StringMap implementation uses a quadratically-probed hash table, where
1296 the buckets store a pointer to the heap allocated entries (and some other
1297 stuff). The entries in the map must be heap allocated because the strings are
1298 variable length. The string data (key) and the element object (value) are
1299 stored in the same allocation with the string data immediately after the element
1300 object. This container guarantees the "<tt>(char*)(&Value+1)</tt>" points
1301 to the key string for a value.</p>
1303 <p>The StringMap is very fast for several reasons: quadratic probing is very
1304 cache efficient for lookups, the hash value of strings in buckets is not
1305 recomputed when lookup up an element, StringMap rarely has to touch the
1306 memory for unrelated objects when looking up a value (even when hash collisions
1307 happen), hash table growth does not recompute the hash values for strings
1308 already in the table, and each pair in the map is store in a single allocation
1309 (the string data is stored in the same allocation as the Value of a pair).</p>
1311 <p>StringMap also provides query methods that take byte ranges, so it only ever
1312 copies a string if a value is inserted into the table.</p>
1315 <!-- _______________________________________________________________________ -->
1316 <div class="doc_subsubsection">
1317 <a name="dss_indexedmap">"llvm/ADT/IndexedMap.h"</a>
1320 <div class="doc_text">
1322 IndexedMap is a specialized container for mapping small dense integers (or
1323 values that can be mapped to small dense integers) to some other type. It is
1324 internally implemented as a vector with a mapping function that maps the keys to
1325 the dense integer range.
1329 This is useful for cases like virtual registers in the LLVM code generator: they
1330 have a dense mapping that is offset by a compile-time constant (the first
1331 virtual register ID).</p>
1335 <!-- _______________________________________________________________________ -->
1336 <div class="doc_subsubsection">
1337 <a name="dss_densemap">"llvm/ADT/DenseMap.h"</a>
1340 <div class="doc_text">
1343 DenseMap is a simple quadratically probed hash table. It excels at supporting
1344 small keys and values: it uses a single allocation to hold all of the pairs that
1345 are currently inserted in the map. DenseMap is a great way to map pointers to
1346 pointers, or map other small types to each other.
1350 There are several aspects of DenseMap that you should be aware of, however. The
1351 iterators in a densemap are invalidated whenever an insertion occurs, unlike
1352 map. Also, because DenseMap allocates space for a large number of key/value
1353 pairs (it starts with 64 by default), it will waste a lot of space if your keys
1354 or values are large. Finally, you must implement a partial specialization of
1355 DenseMapInfo for the key that you want, if it isn't already supported. This
1356 is required to tell DenseMap about two special marker values (which can never be
1357 inserted into the map) that it needs internally.</p>
1361 <!-- _______________________________________________________________________ -->
1362 <div class="doc_subsubsection">
1363 <a name="dss_map"><map></a>
1366 <div class="doc_text">
1369 std::map has similar characteristics to <a href="#dss_set">std::set</a>: it uses
1370 a single allocation per pair inserted into the map, it offers log(n) lookup with
1371 an extremely large constant factor, imposes a space penalty of 3 pointers per
1372 pair in the map, etc.</p>
1374 <p>std::map is most useful when your keys or values are very large, if you need
1375 to iterate over the collection in sorted order, or if you need stable iterators
1376 into the map (i.e. they don't get invalidated if an insertion or deletion of
1377 another element takes place).</p>
1381 <!-- _______________________________________________________________________ -->
1382 <div class="doc_subsubsection">
1383 <a name="dss_othermap">Other Map-Like Container Options</a>
1386 <div class="doc_text">
1389 The STL provides several other options, such as std::multimap and the various
1390 "hash_map" like containers (whether from C++ TR1 or from the SGI library). We
1391 never use hash_set and unordered_set because they are generally very expensive
1392 (each insertion requires a malloc) and very non-portable.</p>
1394 <p>std::multimap is useful if you want to map a key to multiple values, but has
1395 all the drawbacks of std::map. A sorted vector or some other approach is almost
1400 <!-- ======================================================================= -->
1401 <div class="doc_subsection">
1402 <a name="ds_bit">Bit storage containers (BitVector, SparseBitVector)</a>
1405 <div class="doc_text">
1406 <p>Unlike the other containers, there are only two bit storage containers, and
1407 choosing when to use each is relatively straightforward.</p>
1409 <p>One additional option is
1410 <tt>std::vector<bool></tt>: we discourage its use for two reasons 1) the
1411 implementation in many common compilers (e.g. commonly available versions of
1412 GCC) is extremely inefficient and 2) the C++ standards committee is likely to
1413 deprecate this container and/or change it significantly somehow. In any case,
1414 please don't use it.</p>
1417 <!-- _______________________________________________________________________ -->
1418 <div class="doc_subsubsection">
1419 <a name="dss_bitvector">BitVector</a>
1422 <div class="doc_text">
1423 <p> The BitVector container provides a fixed size set of bits for manipulation.
1424 It supports individual bit setting/testing, as well as set operations. The set
1425 operations take time O(size of bitvector), but operations are performed one word
1426 at a time, instead of one bit at a time. This makes the BitVector very fast for
1427 set operations compared to other containers. Use the BitVector when you expect
1428 the number of set bits to be high (IE a dense set).
1432 <!-- _______________________________________________________________________ -->
1433 <div class="doc_subsubsection">
1434 <a name="dss_sparsebitvector">SparseBitVector</a>
1437 <div class="doc_text">
1438 <p> The SparseBitVector container is much like BitVector, with one major
1439 difference: Only the bits that are set, are stored. This makes the
1440 SparseBitVector much more space efficient than BitVector when the set is sparse,
1441 as well as making set operations O(number of set bits) instead of O(size of
1442 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
1443 (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).
1447 <!-- *********************************************************************** -->
1448 <div class="doc_section">
1449 <a name="common">Helpful Hints for Common Operations</a>
1451 <!-- *********************************************************************** -->
1453 <div class="doc_text">
1455 <p>This section describes how to perform some very simple transformations of
1456 LLVM code. This is meant to give examples of common idioms used, showing the
1457 practical side of LLVM transformations. <p> Because this is a "how-to" section,
1458 you should also read about the main classes that you will be working with. The
1459 <a href="#coreclasses">Core LLVM Class Hierarchy Reference</a> contains details
1460 and descriptions of the main classes that you should know about.</p>
1464 <!-- NOTE: this section should be heavy on example code -->
1465 <!-- ======================================================================= -->
1466 <div class="doc_subsection">
1467 <a name="inspection">Basic Inspection and Traversal Routines</a>
1470 <div class="doc_text">
1472 <p>The LLVM compiler infrastructure have many different data structures that may
1473 be traversed. Following the example of the C++ standard template library, the
1474 techniques used to traverse these various data structures are all basically the
1475 same. For a enumerable sequence of values, the <tt>XXXbegin()</tt> function (or
1476 method) returns an iterator to the start of the sequence, the <tt>XXXend()</tt>
1477 function returns an iterator pointing to one past the last valid element of the
1478 sequence, and there is some <tt>XXXiterator</tt> data type that is common
1479 between the two operations.</p>
1481 <p>Because the pattern for iteration is common across many different aspects of
1482 the program representation, the standard template library algorithms may be used
1483 on them, and it is easier to remember how to iterate. First we show a few common
1484 examples of the data structures that need to be traversed. Other data
1485 structures are traversed in very similar ways.</p>
1489 <!-- _______________________________________________________________________ -->
1490 <div class="doc_subsubsection">
1491 <a name="iterate_function">Iterating over the </a><a
1492 href="#BasicBlock"><tt>BasicBlock</tt></a>s in a <a
1493 href="#Function"><tt>Function</tt></a>
1496 <div class="doc_text">
1498 <p>It's quite common to have a <tt>Function</tt> instance that you'd like to
1499 transform in some way; in particular, you'd like to manipulate its
1500 <tt>BasicBlock</tt>s. To facilitate this, you'll need to iterate over all of
1501 the <tt>BasicBlock</tt>s that constitute the <tt>Function</tt>. The following is
1502 an example that prints the name of a <tt>BasicBlock</tt> and the number of
1503 <tt>Instruction</tt>s it contains:</p>
1505 <div class="doc_code">
1507 // <i>func is a pointer to a Function instance</i>
1508 for (Function::iterator i = func->begin(), e = func->end(); i != e; ++i)
1509 // <i>Print out the name of the basic block if it has one, and then the</i>
1510 // <i>number of instructions that it contains</i>
1511 llvm::cerr << "Basic block (name=" << i->getName() << ") has "
1512 << i->size() << " instructions.\n";
1516 <p>Note that i can be used as if it were a pointer for the purposes of
1517 invoking member functions of the <tt>Instruction</tt> class. This is
1518 because the indirection operator is overloaded for the iterator
1519 classes. In the above code, the expression <tt>i->size()</tt> is
1520 exactly equivalent to <tt>(*i).size()</tt> just like you'd expect.</p>
1524 <!-- _______________________________________________________________________ -->
1525 <div class="doc_subsubsection">
1526 <a name="iterate_basicblock">Iterating over the </a><a
1527 href="#Instruction"><tt>Instruction</tt></a>s in a <a
1528 href="#BasicBlock"><tt>BasicBlock</tt></a>
1531 <div class="doc_text">
1533 <p>Just like when dealing with <tt>BasicBlock</tt>s in <tt>Function</tt>s, it's
1534 easy to iterate over the individual instructions that make up
1535 <tt>BasicBlock</tt>s. Here's a code snippet that prints out each instruction in
1536 a <tt>BasicBlock</tt>:</p>
1538 <div class="doc_code">
1540 // <i>blk is a pointer to a BasicBlock instance</i>
1541 for (BasicBlock::iterator i = blk->begin(), e = blk->end(); i != e; ++i)
1542 // <i>The next statement works since operator<<(ostream&,...)</i>
1543 // <i>is overloaded for Instruction&</i>
1544 llvm::cerr << *i << "\n";
1548 <p>However, this isn't really the best way to print out the contents of a
1549 <tt>BasicBlock</tt>! Since the ostream operators are overloaded for virtually
1550 anything you'll care about, you could have just invoked the print routine on the
1551 basic block itself: <tt>llvm::cerr << *blk << "\n";</tt>.</p>
1555 <!-- _______________________________________________________________________ -->
1556 <div class="doc_subsubsection">
1557 <a name="iterate_institer">Iterating over the </a><a
1558 href="#Instruction"><tt>Instruction</tt></a>s in a <a
1559 href="#Function"><tt>Function</tt></a>
1562 <div class="doc_text">
1564 <p>If you're finding that you commonly iterate over a <tt>Function</tt>'s
1565 <tt>BasicBlock</tt>s and then that <tt>BasicBlock</tt>'s <tt>Instruction</tt>s,
1566 <tt>InstIterator</tt> should be used instead. You'll need to include <a
1567 href="/doxygen/InstIterator_8h-source.html"><tt>llvm/Support/InstIterator.h</tt></a>,
1568 and then instantiate <tt>InstIterator</tt>s explicitly in your code. Here's a
1569 small example that shows how to dump all instructions in a function to the standard error stream:<p>
1571 <div class="doc_code">
1573 #include "<a href="/doxygen/InstIterator_8h-source.html">llvm/Support/InstIterator.h</a>"
1575 // <i>F is a pointer to a Function instance</i>
1576 for (inst_iterator I = inst_begin(F), E = inst_end(F); I != E; ++I)
1577 llvm::cerr << *I << "\n";
1581 <p>Easy, isn't it? You can also use <tt>InstIterator</tt>s to fill a
1582 work list with its initial contents. For example, if you wanted to
1583 initialize a work list to contain all instructions in a <tt>Function</tt>
1584 F, all you would need to do is something like:</p>
1586 <div class="doc_code">
1588 std::set<Instruction*> worklist;
1589 // or better yet, SmallPtrSet<Instruction*, 64> worklist;
1591 for (inst_iterator I = inst_begin(F), E = inst_end(F); I != E; ++I)
1592 worklist.insert(&*I);
1596 <p>The STL set <tt>worklist</tt> would now contain all instructions in the
1597 <tt>Function</tt> pointed to by F.</p>
1601 <!-- _______________________________________________________________________ -->
1602 <div class="doc_subsubsection">
1603 <a name="iterate_convert">Turning an iterator into a class pointer (and
1607 <div class="doc_text">
1609 <p>Sometimes, it'll be useful to grab a reference (or pointer) to a class
1610 instance when all you've got at hand is an iterator. Well, extracting
1611 a reference or a pointer from an iterator is very straight-forward.
1612 Assuming that <tt>i</tt> is a <tt>BasicBlock::iterator</tt> and <tt>j</tt>
1613 is a <tt>BasicBlock::const_iterator</tt>:</p>
1615 <div class="doc_code">
1617 Instruction& inst = *i; // <i>Grab reference to instruction reference</i>
1618 Instruction* pinst = &*i; // <i>Grab pointer to instruction reference</i>
1619 const Instruction& inst = *j;
1623 <p>However, the iterators you'll be working with in the LLVM framework are
1624 special: they will automatically convert to a ptr-to-instance type whenever they
1625 need to. Instead of dereferencing the iterator and then taking the address of
1626 the result, you can simply assign the iterator to the proper pointer type and
1627 you get the dereference and address-of operation as a result of the assignment
1628 (behind the scenes, this is a result of overloading casting mechanisms). Thus
1629 the last line of the last example,</p>
1631 <div class="doc_code">
1633 Instruction *pinst = &*i;
1637 <p>is semantically equivalent to</p>
1639 <div class="doc_code">
1641 Instruction *pinst = i;
1645 <p>It's also possible to turn a class pointer into the corresponding iterator,
1646 and this is a constant time operation (very efficient). The following code
1647 snippet illustrates use of the conversion constructors provided by LLVM
1648 iterators. By using these, you can explicitly grab the iterator of something
1649 without actually obtaining it via iteration over some structure:</p>
1651 <div class="doc_code">
1653 void printNextInstruction(Instruction* inst) {
1654 BasicBlock::iterator it(inst);
1655 ++it; // <i>After this line, it refers to the instruction after *inst</i>
1656 if (it != inst->getParent()->end()) llvm::cerr << *it << "\n";
1663 <!--_______________________________________________________________________-->
1664 <div class="doc_subsubsection">
1665 <a name="iterate_complex">Finding call sites: a slightly more complex
1669 <div class="doc_text">
1671 <p>Say that you're writing a FunctionPass and would like to count all the
1672 locations in the entire module (that is, across every <tt>Function</tt>) where a
1673 certain function (i.e., some <tt>Function</tt>*) is already in scope. As you'll
1674 learn later, you may want to use an <tt>InstVisitor</tt> to accomplish this in a
1675 much more straight-forward manner, but this example will allow us to explore how
1676 you'd do it if you didn't have <tt>InstVisitor</tt> around. In pseudo-code, this
1677 is what we want to do:</p>
1679 <div class="doc_code">
1681 initialize callCounter to zero
1682 for each Function f in the Module
1683 for each BasicBlock b in f
1684 for each Instruction i in b
1685 if (i is a CallInst and calls the given function)
1686 increment callCounter
1690 <p>And the actual code is (remember, because we're writing a
1691 <tt>FunctionPass</tt>, our <tt>FunctionPass</tt>-derived class simply has to
1692 override the <tt>runOnFunction</tt> method):</p>
1694 <div class="doc_code">
1696 Function* targetFunc = ...;
1698 class OurFunctionPass : public FunctionPass {
1700 OurFunctionPass(): callCounter(0) { }
1702 virtual runOnFunction(Function& F) {
1703 for (Function::iterator b = F.begin(), be = F.end(); b != be; ++b) {
1704 for (BasicBlock::iterator i = b->begin(), ie = b->end(); i != ie; ++i) {
1705 if (<a href="#CallInst">CallInst</a>* callInst = <a href="#isa">dyn_cast</a><<a
1706 href="#CallInst">CallInst</a>>(&*i)) {
1707 // <i>We know we've encountered a call instruction, so we</i>
1708 // <i>need to determine if it's a call to the</i>
1709 // <i>function pointed to by m_func or not.</i>
1710 if (callInst->getCalledFunction() == targetFunc)
1718 unsigned callCounter;
1725 <!--_______________________________________________________________________-->
1726 <div class="doc_subsubsection">
1727 <a name="calls_and_invokes">Treating calls and invokes the same way</a>
1730 <div class="doc_text">
1732 <p>You may have noticed that the previous example was a bit oversimplified in
1733 that it did not deal with call sites generated by 'invoke' instructions. In
1734 this, and in other situations, you may find that you want to treat
1735 <tt>CallInst</tt>s and <tt>InvokeInst</tt>s the same way, even though their
1736 most-specific common base class is <tt>Instruction</tt>, which includes lots of
1737 less closely-related things. For these cases, LLVM provides a handy wrapper
1739 href="http://llvm.org/doxygen/classllvm_1_1CallSite.html"><tt>CallSite</tt></a>.
1740 It is essentially a wrapper around an <tt>Instruction</tt> pointer, with some
1741 methods that provide functionality common to <tt>CallInst</tt>s and
1742 <tt>InvokeInst</tt>s.</p>
1744 <p>This class has "value semantics": it should be passed by value, not by
1745 reference and it should not be dynamically allocated or deallocated using
1746 <tt>operator new</tt> or <tt>operator delete</tt>. It is efficiently copyable,
1747 assignable and constructable, with costs equivalents to that of a bare pointer.
1748 If you look at its definition, it has only a single pointer member.</p>
1752 <!--_______________________________________________________________________-->
1753 <div class="doc_subsubsection">
1754 <a name="iterate_chains">Iterating over def-use & use-def chains</a>
1757 <div class="doc_text">
1759 <p>Frequently, we might have an instance of the <a
1760 href="/doxygen/classllvm_1_1Value.html">Value Class</a> and we want to
1761 determine which <tt>User</tt>s use the <tt>Value</tt>. The list of all
1762 <tt>User</tt>s of a particular <tt>Value</tt> is called a <i>def-use</i> chain.
1763 For example, let's say we have a <tt>Function*</tt> named <tt>F</tt> to a
1764 particular function <tt>foo</tt>. Finding all of the instructions that
1765 <i>use</i> <tt>foo</tt> is as simple as iterating over the <i>def-use</i> chain
1768 <div class="doc_code">
1772 for (Value::use_iterator i = F->use_begin(), e = F->use_end(); i != e; ++i)
1773 if (Instruction *Inst = dyn_cast<Instruction>(*i)) {
1774 llvm::cerr << "F is used in instruction:\n";
1775 llvm::cerr << *Inst << "\n";
1780 <p>Alternately, it's common to have an instance of the <a
1781 href="/doxygen/classllvm_1_1User.html">User Class</a> and need to know what
1782 <tt>Value</tt>s are used by it. The list of all <tt>Value</tt>s used by a
1783 <tt>User</tt> is known as a <i>use-def</i> chain. Instances of class
1784 <tt>Instruction</tt> are common <tt>User</tt>s, so we might want to iterate over
1785 all of the values that a particular instruction uses (that is, the operands of
1786 the particular <tt>Instruction</tt>):</p>
1788 <div class="doc_code">
1790 Instruction *pi = ...;
1792 for (User::op_iterator i = pi->op_begin(), e = pi->op_end(); i != e; ++i) {
1800 def-use chains ("finding all users of"): Value::use_begin/use_end
1801 use-def chains ("finding all values used"): User::op_begin/op_end [op=operand]
1806 <!--_______________________________________________________________________-->
1807 <div class="doc_subsubsection">
1808 <a name="iterate_preds">Iterating over predecessors &
1809 successors of blocks</a>
1812 <div class="doc_text">
1814 <p>Iterating over the predecessors and successors of a block is quite easy
1815 with the routines defined in <tt>"llvm/Support/CFG.h"</tt>. Just use code like
1816 this to iterate over all predecessors of BB:</p>
1818 <div class="doc_code">
1820 #include "llvm/Support/CFG.h"
1821 BasicBlock *BB = ...;
1823 for (pred_iterator PI = pred_begin(BB), E = pred_end(BB); PI != E; ++PI) {
1824 BasicBlock *Pred = *PI;
1830 <p>Similarly, to iterate over successors use
1831 succ_iterator/succ_begin/succ_end.</p>
1836 <!-- ======================================================================= -->
1837 <div class="doc_subsection">
1838 <a name="simplechanges">Making simple changes</a>
1841 <div class="doc_text">
1843 <p>There are some primitive transformation operations present in the LLVM
1844 infrastructure that are worth knowing about. When performing
1845 transformations, it's fairly common to manipulate the contents of basic
1846 blocks. This section describes some of the common methods for doing so
1847 and gives example code.</p>
1851 <!--_______________________________________________________________________-->
1852 <div class="doc_subsubsection">
1853 <a name="schanges_creating">Creating and inserting new
1854 <tt>Instruction</tt>s</a>
1857 <div class="doc_text">
1859 <p><i>Instantiating Instructions</i></p>
1861 <p>Creation of <tt>Instruction</tt>s is straight-forward: simply call the
1862 constructor for the kind of instruction to instantiate and provide the necessary
1863 parameters. For example, an <tt>AllocaInst</tt> only <i>requires</i> a
1864 (const-ptr-to) <tt>Type</tt>. Thus:</p>
1866 <div class="doc_code">
1868 AllocaInst* ai = new AllocaInst(Type::Int32Ty);
1872 <p>will create an <tt>AllocaInst</tt> instance that represents the allocation of
1873 one integer in the current stack frame, at run time. Each <tt>Instruction</tt>
1874 subclass is likely to have varying default parameters which change the semantics
1875 of the instruction, so refer to the <a
1876 href="/doxygen/classllvm_1_1Instruction.html">doxygen documentation for the subclass of
1877 Instruction</a> that you're interested in instantiating.</p>
1879 <p><i>Naming values</i></p>
1881 <p>It is very useful to name the values of instructions when you're able to, as
1882 this facilitates the debugging of your transformations. If you end up looking
1883 at generated LLVM machine code, you definitely want to have logical names
1884 associated with the results of instructions! By supplying a value for the
1885 <tt>Name</tt> (default) parameter of the <tt>Instruction</tt> constructor, you
1886 associate a logical name with the result of the instruction's execution at
1887 run time. For example, say that I'm writing a transformation that dynamically
1888 allocates space for an integer on the stack, and that integer is going to be
1889 used as some kind of index by some other code. To accomplish this, I place an
1890 <tt>AllocaInst</tt> at the first point in the first <tt>BasicBlock</tt> of some
1891 <tt>Function</tt>, and I'm intending to use it within the same
1892 <tt>Function</tt>. I might do:</p>
1894 <div class="doc_code">
1896 AllocaInst* pa = new AllocaInst(Type::Int32Ty, 0, "indexLoc");
1900 <p>where <tt>indexLoc</tt> is now the logical name of the instruction's
1901 execution value, which is a pointer to an integer on the run time stack.</p>
1903 <p><i>Inserting instructions</i></p>
1905 <p>There are essentially two ways to insert an <tt>Instruction</tt>
1906 into an existing sequence of instructions that form a <tt>BasicBlock</tt>:</p>
1909 <li>Insertion into an explicit instruction list
1911 <p>Given a <tt>BasicBlock* pb</tt>, an <tt>Instruction* pi</tt> within that
1912 <tt>BasicBlock</tt>, and a newly-created instruction we wish to insert
1913 before <tt>*pi</tt>, we do the following: </p>
1915 <div class="doc_code">
1917 BasicBlock *pb = ...;
1918 Instruction *pi = ...;
1919 Instruction *newInst = new Instruction(...);
1921 pb->getInstList().insert(pi, newInst); // <i>Inserts newInst before pi in pb</i>
1925 <p>Appending to the end of a <tt>BasicBlock</tt> is so common that
1926 the <tt>Instruction</tt> class and <tt>Instruction</tt>-derived
1927 classes provide constructors which take a pointer to a
1928 <tt>BasicBlock</tt> to be appended to. For example code that
1931 <div class="doc_code">
1933 BasicBlock *pb = ...;
1934 Instruction *newInst = new Instruction(...);
1936 pb->getInstList().push_back(newInst); // <i>Appends newInst to pb</i>
1942 <div class="doc_code">
1944 BasicBlock *pb = ...;
1945 Instruction *newInst = new Instruction(..., pb);
1949 <p>which is much cleaner, especially if you are creating
1950 long instruction streams.</p></li>
1952 <li>Insertion into an implicit instruction list
1954 <p><tt>Instruction</tt> instances that are already in <tt>BasicBlock</tt>s
1955 are implicitly associated with an existing instruction list: the instruction
1956 list of the enclosing basic block. Thus, we could have accomplished the same
1957 thing as the above code without being given a <tt>BasicBlock</tt> by doing:
1960 <div class="doc_code">
1962 Instruction *pi = ...;
1963 Instruction *newInst = new Instruction(...);
1965 pi->getParent()->getInstList().insert(pi, newInst);
1969 <p>In fact, this sequence of steps occurs so frequently that the
1970 <tt>Instruction</tt> class and <tt>Instruction</tt>-derived classes provide
1971 constructors which take (as a default parameter) a pointer to an
1972 <tt>Instruction</tt> which the newly-created <tt>Instruction</tt> should
1973 precede. That is, <tt>Instruction</tt> constructors are capable of
1974 inserting the newly-created instance into the <tt>BasicBlock</tt> of a
1975 provided instruction, immediately before that instruction. Using an
1976 <tt>Instruction</tt> constructor with a <tt>insertBefore</tt> (default)
1977 parameter, the above code becomes:</p>
1979 <div class="doc_code">
1981 Instruction* pi = ...;
1982 Instruction* newInst = new Instruction(..., pi);
1986 <p>which is much cleaner, especially if you're creating a lot of
1987 instructions and adding them to <tt>BasicBlock</tt>s.</p></li>
1992 <!--_______________________________________________________________________-->
1993 <div class="doc_subsubsection">
1994 <a name="schanges_deleting">Deleting <tt>Instruction</tt>s</a>
1997 <div class="doc_text">
1999 <p>Deleting an instruction from an existing sequence of instructions that form a
2000 <a href="#BasicBlock"><tt>BasicBlock</tt></a> is very straight-forward. First,
2001 you must have a pointer to the instruction that you wish to delete. Second, you
2002 need to obtain the pointer to that instruction's basic block. You use the
2003 pointer to the basic block to get its list of instructions and then use the
2004 erase function to remove your instruction. For example:</p>
2006 <div class="doc_code">
2008 <a href="#Instruction">Instruction</a> *I = .. ;
2009 I->eraseFromParent();
2015 <!--_______________________________________________________________________-->
2016 <div class="doc_subsubsection">
2017 <a name="schanges_replacing">Replacing an <tt>Instruction</tt> with another
2021 <div class="doc_text">
2023 <p><i>Replacing individual instructions</i></p>
2025 <p>Including "<a href="/doxygen/BasicBlockUtils_8h-source.html">llvm/Transforms/Utils/BasicBlockUtils.h</a>"
2026 permits use of two very useful replace functions: <tt>ReplaceInstWithValue</tt>
2027 and <tt>ReplaceInstWithInst</tt>.</p>
2029 <h4><a name="schanges_deleting">Deleting <tt>Instruction</tt>s</a></h4>
2032 <li><tt>ReplaceInstWithValue</tt>
2034 <p>This function replaces all uses of a given instruction with a value,
2035 and then removes the original instruction. The following example
2036 illustrates the replacement of the result of a particular
2037 <tt>AllocaInst</tt> that allocates memory for a single integer with a null
2038 pointer to an integer.</p>
2040 <div class="doc_code">
2042 AllocaInst* instToReplace = ...;
2043 BasicBlock::iterator ii(instToReplace);
2045 ReplaceInstWithValue(instToReplace->getParent()->getInstList(), ii,
2046 Constant::getNullValue(PointerType::getUnqual(Type::Int32Ty)));
2049 <li><tt>ReplaceInstWithInst</tt>
2051 <p>This function replaces a particular instruction with another
2052 instruction, inserting the new instruction into the basic block at the
2053 location where the old instruction was, and replacing any uses of the old
2054 instruction with the new instruction. The following example illustrates
2055 the replacement of one <tt>AllocaInst</tt> with another.</p>
2057 <div class="doc_code">
2059 AllocaInst* instToReplace = ...;
2060 BasicBlock::iterator ii(instToReplace);
2062 ReplaceInstWithInst(instToReplace->getParent()->getInstList(), ii,
2063 new AllocaInst(Type::Int32Ty, 0, "ptrToReplacedInt"));
2067 <p><i>Replacing multiple uses of <tt>User</tt>s and <tt>Value</tt>s</i></p>
2069 <p>You can use <tt>Value::replaceAllUsesWith</tt> and
2070 <tt>User::replaceUsesOfWith</tt> to change more than one use at a time. See the
2071 doxygen documentation for the <a href="/doxygen/classllvm_1_1Value.html">Value Class</a>
2072 and <a href="/doxygen/classllvm_1_1User.html">User Class</a>, respectively, for more
2075 <!-- Value::replaceAllUsesWith User::replaceUsesOfWith Point out:
2076 include/llvm/Transforms/Utils/ especially BasicBlockUtils.h with:
2077 ReplaceInstWithValue, ReplaceInstWithInst -->
2081 <!--_______________________________________________________________________-->
2082 <div class="doc_subsubsection">
2083 <a name="schanges_deletingGV">Deleting <tt>GlobalVariable</tt>s</a>
2086 <div class="doc_text">
2088 <p>Deleting a global variable from a module is just as easy as deleting an
2089 Instruction. First, you must have a pointer to the global variable that you wish
2090 to delete. You use this pointer to erase it from its parent, the module.
2093 <div class="doc_code">
2095 <a href="#GlobalVariable">GlobalVariable</a> *GV = .. ;
2097 GV->eraseFromParent();
2103 <!-- ======================================================================= -->
2104 <div class="doc_subsection">
2105 <a name="create_types">How to Create Types</a>
2108 <div class="doc_text">
2110 <p>In generating IR, you may need some complex types. If you know these types
2111 statically, you can use <tt>TypeBuilder<...>::get()</tt>, defined
2112 in <tt>llvm/Support/TypeBuilder.h</tt>, to retrieve them. <tt>TypeBuilder</tt>
2113 has two forms depending on whether you're building types for cross-compilation
2114 or native library use. <tt>TypeBuilder<T, true></tt> requires
2115 that <tt>T</tt> be independent of the host environment, meaning that it's built
2117 the <a href="/doxygen/namespacellvm_1_1types.html"><tt>llvm::types</tt></a>
2118 namespace and pointers, functions, arrays, etc. built of
2119 those. <tt>TypeBuilder<T, false></tt> additionally allows native C types
2120 whose size may depend on the host compiler. For example,</p>
2122 <div class="doc_code">
2124 FunctionType *ft = TypeBuilder<types::i<8>(types::i<32>*), true>::get();
2128 <p>is easier to read and write than the equivalent</p>
2130 <div class="doc_code">
2132 std::vector<const Type*> params;
2133 params.push_back(PointerType::getUnqual(Type::Int32Ty));
2134 FunctionType *ft = FunctionType::get(Type::Int8Ty, params, false);
2138 <p>See the <a href="/doxygen/TypeBuilder_8h-source.html#l00001">class
2139 comment</a> for more details.</p>
2143 <!-- *********************************************************************** -->
2144 <div class="doc_section">
2145 <a name="threading">Threads and LLVM</a>
2147 <!-- *********************************************************************** -->
2149 <div class="doc_text">
2151 This section describes the interaction of the LLVM APIs with multithreading,
2152 both on the part of client applications, and in the JIT, in the hosted
2157 Note that LLVM's support for multithreading is still relatively young. Up
2158 through version 2.5, the execution of threaded hosted applications was
2159 supported, but not threaded client access to the APIs. While this use case is
2160 now supported, clients <em>must</em> adhere to the guidelines specified below to
2161 ensure proper operation in multithreaded mode.
2165 Note that, on Unix-like platforms, LLVM requires the presence of GCC's atomic
2166 intrinsics in order to support threaded operation. If you need a
2167 multhreading-capable LLVM on a platform without a suitably modern system
2168 compiler, consider compiling LLVM and LLVM-GCC in single-threaded mode, and
2169 using the resultant compiler to build a copy of LLVM with multithreading
2174 <!-- ======================================================================= -->
2175 <div class="doc_subsection">
2176 <a name="startmultithreaded">Entering and Exiting Multithreaded Mode</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 must intialize
2184 certain data structures necessary to provide guards around its internals. To do
2185 so, the client program must invoke <tt>llvm_start_multithreaded()</tt> before
2186 making any concurrent LLVM API calls. To subsequently tear down these
2187 structures, use the <tt>llvm_stop_multithreaded()</tt> call. You can also use
2188 the <tt>llvm_is_multithreaded()</tt> call to check the status of multithreaded
2193 Note that both of these calls must be made <em>in isolation</em>. That is to
2194 say that no other LLVM API calls may be executing at any time during the
2195 execution of <tt>llvm_start_multithreaded()</tt> or <tt>llvm_stop_multithreaded
2196 </tt>. It's is the client's responsibility to enforce this isolation.
2200 The return value of <tt>llvm_start_multithreaded()</tt> indicates the success or
2201 failure of the initialization. Failure typically indicates that your copy of
2202 LLVM was built without multithreading support, typically because GCC atomic
2203 intrinsics were not found in your system compiler. In this case, the LLVM API
2204 will not be safe for concurrent calls. However, it <em>will</em> be safe for
2205 hosting threaded applications in the JIT, though care must be taken to ensure
2206 that side exits and the like do not accidentally result in concurrent LLVM API
2211 <!-- ======================================================================= -->
2212 <div class="doc_subsection">
2213 <a name="shutdown">Ending Execution with <tt>llvm_shutdown()</tt></a>
2216 <div class="doc_text">
2218 When you are done using the LLVM APIs, you should call <tt>llvm_shutdown()</tt>
2219 to deallocate memory used for internal structures. This will also invoke
2220 <tt>llvm_stop_multithreaded()</tt> if LLVM is operating in multithreaded mode.
2221 As such, <tt>llvm_shutdown()</tt> requires the same isolation guarantees as
2222 <tt>llvm_stop_multithreaded()</tt>.
2226 Note that, if you use scope-based shutdown, you can use the
2227 <tt>llvm_shutdown_obj</tt> class, which calls <tt>llvm_shutdown()</tt> in its
2231 <!-- ======================================================================= -->
2232 <div class="doc_subsection">
2233 <a name="managedstatic">Lazy Initialization with <tt>ManagedStatic</tt></a>
2236 <div class="doc_text">
2238 <tt>ManagedStatic</tt> is a utility class in LLVM used to implement static
2239 initialization of static resources, such as the global type tables. Before the
2240 invocation of <tt>llvm_shutdown()</tt>, it implements a simple lazy
2241 initialization scheme. Once <tt>llvm_start_multithreaded()</tt> returns,
2242 however, it uses double-checked locking to implement thread-safe lazy
2247 Note that, because no other threads are allowed to issue LLVM API calls before
2248 <tt>llvm_start_multithreaded()</tt> returns, it is possible to have
2249 <tt>ManagedStatic</tt>s of <tt>llvm::sys::Mutex</tt>s.
2253 The <tt>llvm_acquire_global_lock()</tt> and <tt>llvm_release_global_lock</tt>
2254 APIs provide access to the global lock used to implement the double-checked
2255 locking for lazy initialization. These should only be used internally to LLVM,
2256 and only if you know what you're doing!
2260 <!-- *********************************************************************** -->
2261 <div class="doc_section">
2262 <a name="advanced">Advanced Topics</a>
2264 <!-- *********************************************************************** -->
2266 <div class="doc_text">
2268 This section describes some of the advanced or obscure API's that most clients
2269 do not need to be aware of. These API's tend manage the inner workings of the
2270 LLVM system, and only need to be accessed in unusual circumstances.
2274 <!-- ======================================================================= -->
2275 <div class="doc_subsection">
2276 <a name="TypeResolve">LLVM Type Resolution</a>
2279 <div class="doc_text">
2282 The LLVM type system has a very simple goal: allow clients to compare types for
2283 structural equality with a simple pointer comparison (aka a shallow compare).
2284 This goal makes clients much simpler and faster, and is used throughout the LLVM
2289 Unfortunately achieving this goal is not a simple matter. In particular,
2290 recursive types and late resolution of opaque types makes the situation very
2291 difficult to handle. Fortunately, for the most part, our implementation makes
2292 most clients able to be completely unaware of the nasty internal details. The
2293 primary case where clients are exposed to the inner workings of it are when
2294 building a recursive type. In addition to this case, the LLVM bitcode reader,
2295 assembly parser, and linker also have to be aware of the inner workings of this
2300 For our purposes below, we need three concepts. First, an "Opaque Type" is
2301 exactly as defined in the <a href="LangRef.html#t_opaque">language
2302 reference</a>. Second an "Abstract Type" is any type which includes an
2303 opaque type as part of its type graph (for example "<tt>{ opaque, i32 }</tt>").
2304 Third, a concrete type is a type that is not an abstract type (e.g. "<tt>{ i32,
2310 <!-- ______________________________________________________________________ -->
2311 <div class="doc_subsubsection">
2312 <a name="BuildRecType">Basic Recursive Type Construction</a>
2315 <div class="doc_text">
2318 Because the most common question is "how do I build a recursive type with LLVM",
2319 we answer it now and explain it as we go. Here we include enough to cause this
2320 to be emitted to an output .ll file:
2323 <div class="doc_code">
2325 %mylist = type { %mylist*, i32 }
2330 To build this, use the following LLVM APIs:
2333 <div class="doc_code">
2335 // <i>Create the initial outer struct</i>
2336 <a href="#PATypeHolder">PATypeHolder</a> StructTy = OpaqueType::get();
2337 std::vector<const Type*> Elts;
2338 Elts.push_back(PointerType::getUnqual(StructTy));
2339 Elts.push_back(Type::Int32Ty);
2340 StructType *NewSTy = StructType::get(Elts);
2342 // <i>At this point, NewSTy = "{ opaque*, i32 }". Tell VMCore that</i>
2343 // <i>the struct and the opaque type are actually the same.</i>
2344 cast<OpaqueType>(StructTy.get())-><a href="#refineAbstractTypeTo">refineAbstractTypeTo</a>(NewSTy);
2346 // <i>NewSTy is potentially invalidated, but StructTy (a <a href="#PATypeHolder">PATypeHolder</a>) is</i>
2347 // <i>kept up-to-date</i>
2348 NewSTy = cast<StructType>(StructTy.get());
2350 // <i>Add a name for the type to the module symbol table (optional)</i>
2351 MyModule->addTypeName("mylist", NewSTy);
2356 This code shows the basic approach used to build recursive types: build a
2357 non-recursive type using 'opaque', then use type unification to close the cycle.
2358 The type unification step is performed by the <tt><a
2359 href="#refineAbstractTypeTo">refineAbstractTypeTo</a></tt> method, which is
2360 described next. After that, we describe the <a
2361 href="#PATypeHolder">PATypeHolder class</a>.
2366 <!-- ______________________________________________________________________ -->
2367 <div class="doc_subsubsection">
2368 <a name="refineAbstractTypeTo">The <tt>refineAbstractTypeTo</tt> method</a>
2371 <div class="doc_text">
2373 The <tt>refineAbstractTypeTo</tt> method starts the type unification process.
2374 While this method is actually a member of the DerivedType class, it is most
2375 often used on OpaqueType instances. Type unification is actually a recursive
2376 process. After unification, types can become structurally isomorphic to
2377 existing types, and all duplicates are deleted (to preserve pointer equality).
2381 In the example above, the OpaqueType object is definitely deleted.
2382 Additionally, if there is an "{ \2*, i32}" type already created in the system,
2383 the pointer and struct type created are <b>also</b> deleted. Obviously whenever
2384 a type is deleted, any "Type*" pointers in the program are invalidated. As
2385 such, it is safest to avoid having <i>any</i> "Type*" pointers to abstract types
2386 live across a call to <tt>refineAbstractTypeTo</tt> (note that non-abstract
2387 types can never move or be deleted). To deal with this, the <a
2388 href="#PATypeHolder">PATypeHolder</a> class is used to maintain a stable
2389 reference to a possibly refined type, and the <a
2390 href="#AbstractTypeUser">AbstractTypeUser</a> class is used to update more
2391 complex datastructures.
2396 <!-- ______________________________________________________________________ -->
2397 <div class="doc_subsubsection">
2398 <a name="PATypeHolder">The PATypeHolder Class</a>
2401 <div class="doc_text">
2403 PATypeHolder is a form of a "smart pointer" for Type objects. When VMCore
2404 happily goes about nuking types that become isomorphic to existing types, it
2405 automatically updates all PATypeHolder objects to point to the new type. In the
2406 example above, this allows the code to maintain a pointer to the resultant
2407 resolved recursive type, even though the Type*'s are potentially invalidated.
2411 PATypeHolder is an extremely light-weight object that uses a lazy union-find
2412 implementation to update pointers. For example the pointer from a Value to its
2413 Type is maintained by PATypeHolder objects.
2418 <!-- ______________________________________________________________________ -->
2419 <div class="doc_subsubsection">
2420 <a name="AbstractTypeUser">The AbstractTypeUser Class</a>
2423 <div class="doc_text">
2426 Some data structures need more to perform more complex updates when types get
2427 resolved. To support this, a class can derive from the AbstractTypeUser class.
2429 allows it to get callbacks when certain types are resolved. To register to get
2430 callbacks for a particular type, the DerivedType::{add/remove}AbstractTypeUser
2431 methods can be called on a type. Note that these methods only work for <i>
2432 abstract</i> types. Concrete types (those that do not include any opaque
2433 objects) can never be refined.
2438 <!-- ======================================================================= -->
2439 <div class="doc_subsection">
2440 <a name="SymbolTable">The <tt>ValueSymbolTable</tt> and
2441 <tt>TypeSymbolTable</tt> classes</a>
2444 <div class="doc_text">
2445 <p>The <tt><a href="http://llvm.org/doxygen/classllvm_1_1ValueSymbolTable.html">
2446 ValueSymbolTable</a></tt> class provides a symbol table that the <a
2447 href="#Function"><tt>Function</tt></a> and <a href="#Module">
2448 <tt>Module</tt></a> classes use for naming value definitions. The symbol table
2449 can provide a name for any <a href="#Value"><tt>Value</tt></a>.
2450 The <tt><a href="http://llvm.org/doxygen/classllvm_1_1TypeSymbolTable.html">
2451 TypeSymbolTable</a></tt> class is used by the <tt>Module</tt> class to store
2452 names for types.</p>
2454 <p>Note that the <tt>SymbolTable</tt> class should not be directly accessed
2455 by most clients. It should only be used when iteration over the symbol table
2456 names themselves are required, which is very special purpose. Note that not
2458 <tt><a href="#Value">Value</a></tt>s have names, and those without names (i.e. they have
2459 an empty name) do not exist in the symbol table.
2462 <p>These symbol tables support iteration over the values/types in the symbol
2463 table with <tt>begin/end/iterator</tt> and supports querying to see if a
2464 specific name is in the symbol table (with <tt>lookup</tt>). The
2465 <tt>ValueSymbolTable</tt> class exposes no public mutator methods, instead,
2466 simply call <tt>setName</tt> on a value, which will autoinsert it into the
2467 appropriate symbol table. For types, use the Module::addTypeName method to
2468 insert entries into the symbol table.</p>
2474 <!-- ======================================================================= -->
2475 <div class="doc_subsection">
2476 <a name="UserLayout">The <tt>User</tt> and owned <tt>Use</tt> classes' memory layout</a>
2479 <div class="doc_text">
2480 <p>The <tt><a href="http://llvm.org/doxygen/classllvm_1_1User.html">
2481 User</a></tt> class provides a basis for expressing the ownership of <tt>User</tt>
2482 towards other <tt><a href="http://llvm.org/doxygen/classllvm_1_1Value.html">
2483 Value</a></tt>s. The <tt><a href="http://llvm.org/doxygen/classllvm_1_1Use.html">
2484 Use</a></tt> helper class is employed to do the bookkeeping and to facilitate <i>O(1)</i>
2485 addition and removal.</p>
2487 <!-- ______________________________________________________________________ -->
2488 <div class="doc_subsubsection">
2489 <a name="Use2User">Interaction and relationship between <tt>User</tt> and <tt>Use</tt> objects</a>
2492 <div class="doc_text">
2494 A subclass of <tt>User</tt> can choose between incorporating its <tt>Use</tt> objects
2495 or refer to them out-of-line by means of a pointer. A mixed variant
2496 (some <tt>Use</tt>s inline others hung off) is impractical and breaks the invariant
2497 that the <tt>Use</tt> objects belonging to the same <tt>User</tt> form a contiguous array.
2502 We have 2 different layouts in the <tt>User</tt> (sub)classes:
2505 The <tt>Use</tt> object(s) are inside (resp. at fixed offset) of the <tt>User</tt>
2506 object and there are a fixed number of them.</p>
2509 The <tt>Use</tt> object(s) are referenced by a pointer to an
2510 array from the <tt>User</tt> object and there may be a variable
2514 As of v2.4 each layout still possesses a direct pointer to the
2515 start of the array of <tt>Use</tt>s. Though not mandatory for layout a),
2516 we stick to this redundancy for the sake of simplicity.
2517 The <tt>User</tt> object also stores the number of <tt>Use</tt> objects it
2518 has. (Theoretically this information can also be calculated
2519 given the scheme presented below.)</p>
2521 Special forms of allocation operators (<tt>operator new</tt>)
2522 enforce the following memory layouts:</p>
2525 <li><p>Layout a) is modelled by prepending the <tt>User</tt> object by the <tt>Use[]</tt> array.</p>
2528 ...---.---.---.---.-------...
2529 | P | P | P | P | User
2530 '''---'---'---'---'-------'''
2533 <li><p>Layout b) is modelled by pointing at the <tt>Use[]</tt> array.</p>
2545 <i>(In the above figures '<tt>P</tt>' stands for the <tt>Use**</tt> that
2546 is stored in each <tt>Use</tt> object in the member <tt>Use::Prev</tt>)</i>
2548 <!-- ______________________________________________________________________ -->
2549 <div class="doc_subsubsection">
2550 <a name="Waymarking">The waymarking algorithm</a>
2553 <div class="doc_text">
2555 Since the <tt>Use</tt> objects are deprived of the direct (back)pointer to
2556 their <tt>User</tt> objects, there must be a fast and exact method to
2557 recover it. This is accomplished by the following scheme:</p>
2560 A bit-encoding in the 2 LSBits (least significant bits) of the <tt>Use::Prev</tt> allows to find the
2561 start of the <tt>User</tt> object:
2563 <li><tt>00</tt> —> binary digit 0</li>
2564 <li><tt>01</tt> —> binary digit 1</li>
2565 <li><tt>10</tt> —> stop and calculate (<tt>s</tt>)</li>
2566 <li><tt>11</tt> —> full stop (<tt>S</tt>)</li>
2569 Given a <tt>Use*</tt>, all we have to do is to walk till we get
2570 a stop and we either have a <tt>User</tt> immediately behind or
2571 we have to walk to the next stop picking up digits
2572 and calculating the offset:</p>
2574 .---.---.---.---.---.---.---.---.---.---.---.---.---.---.---.---.----------------
2575 | 1 | s | 1 | 0 | 1 | 0 | s | 1 | 1 | 0 | s | 1 | 1 | s | 1 | S | User (or User*)
2576 '---'---'---'---'---'---'---'---'---'---'---'---'---'---'---'---'----------------
2577 |+15 |+10 |+6 |+3 |+1
2580 | | |______________________>
2581 | |______________________________________>
2582 |__________________________________________________________>
2585 Only the significant number of bits need to be stored between the
2586 stops, so that the <i>worst case is 20 memory accesses</i> when there are
2587 1000 <tt>Use</tt> objects associated with a <tt>User</tt>.</p>
2589 <!-- ______________________________________________________________________ -->
2590 <div class="doc_subsubsection">
2591 <a name="ReferenceImpl">Reference implementation</a>
2594 <div class="doc_text">
2596 The following literate Haskell fragment demonstrates the concept:</p>
2599 <div class="doc_code">
2601 > import Test.QuickCheck
2603 > digits :: Int -> [Char] -> [Char]
2604 > digits 0 acc = '0' : acc
2605 > digits 1 acc = '1' : acc
2606 > digits n acc = digits (n `div` 2) $ digits (n `mod` 2) acc
2608 > dist :: Int -> [Char] -> [Char]
2611 > dist 1 acc = let r = dist 0 acc in 's' : digits (length r) r
2612 > dist n acc = dist (n - 1) $ dist 1 acc
2614 > takeLast n ss = reverse $ take n $ reverse ss
2616 > test = takeLast 40 $ dist 20 []
2621 Printing <test> gives: <tt>"1s100000s11010s10100s1111s1010s110s11s1S"</tt></p>
2623 The reverse algorithm computes the length of the string just by examining
2624 a certain prefix:</p>
2626 <div class="doc_code">
2628 > pref :: [Char] -> Int
2630 > pref ('s':'1':rest) = decode 2 1 rest
2631 > pref (_:rest) = 1 + pref rest
2633 > decode walk acc ('0':rest) = decode (walk + 1) (acc * 2) rest
2634 > decode walk acc ('1':rest) = decode (walk + 1) (acc * 2 + 1) rest
2635 > decode walk acc _ = walk + acc
2640 Now, as expected, printing <pref test> gives <tt>40</tt>.</p>
2642 We can <i>quickCheck</i> this with following property:</p>
2644 <div class="doc_code">
2646 > testcase = dist 2000 []
2647 > testcaseLength = length testcase
2649 > identityProp n = n > 0 && n <= testcaseLength ==> length arr == pref arr
2650 > where arr = takeLast n testcase
2655 As expected <quickCheck identityProp> gives:</p>
2658 *Main> quickCheck identityProp
2659 OK, passed 100 tests.
2662 Let's be a bit more exhaustive:</p>
2664 <div class="doc_code">
2667 > deepCheck p = check (defaultConfig { configMaxTest = 500 }) p
2672 And here is the result of <deepCheck identityProp>:</p>
2675 *Main> deepCheck identityProp
2676 OK, passed 500 tests.
2679 <!-- ______________________________________________________________________ -->
2680 <div class="doc_subsubsection">
2681 <a name="Tagging">Tagging considerations</a>
2685 To maintain the invariant that the 2 LSBits of each <tt>Use**</tt> in <tt>Use</tt>
2686 never change after being set up, setters of <tt>Use::Prev</tt> must re-tag the
2687 new <tt>Use**</tt> on every modification. Accordingly getters must strip the
2690 For layout b) instead of the <tt>User</tt> we find a pointer (<tt>User*</tt> with LSBit set).
2691 Following this pointer brings us to the <tt>User</tt>. A portable trick ensures
2692 that the first bytes of <tt>User</tt> (if interpreted as a pointer) never has
2693 the LSBit set. (Portability is relying on the fact that all known compilers place the
2694 <tt>vptr</tt> in the first word of the instances.)</p>
2698 <!-- *********************************************************************** -->
2699 <div class="doc_section">
2700 <a name="coreclasses">The Core LLVM Class Hierarchy Reference </a>
2702 <!-- *********************************************************************** -->
2704 <div class="doc_text">
2705 <p><tt>#include "<a href="/doxygen/Type_8h-source.html">llvm/Type.h</a>"</tt>
2706 <br>doxygen info: <a href="/doxygen/classllvm_1_1Type.html">Type Class</a></p>
2708 <p>The Core LLVM classes are the primary means of representing the program
2709 being inspected or transformed. The core LLVM classes are defined in
2710 header files in the <tt>include/llvm/</tt> directory, and implemented in
2711 the <tt>lib/VMCore</tt> directory.</p>
2715 <!-- ======================================================================= -->
2716 <div class="doc_subsection">
2717 <a name="Type">The <tt>Type</tt> class and Derived Types</a>
2720 <div class="doc_text">
2722 <p><tt>Type</tt> is a superclass of all type classes. Every <tt>Value</tt> has
2723 a <tt>Type</tt>. <tt>Type</tt> cannot be instantiated directly but only
2724 through its subclasses. Certain primitive types (<tt>VoidType</tt>,
2725 <tt>LabelType</tt>, <tt>FloatType</tt> and <tt>DoubleType</tt>) have hidden
2726 subclasses. They are hidden because they offer no useful functionality beyond
2727 what the <tt>Type</tt> class offers except to distinguish themselves from
2728 other subclasses of <tt>Type</tt>.</p>
2729 <p>All other types are subclasses of <tt>DerivedType</tt>. Types can be
2730 named, but this is not a requirement. There exists exactly
2731 one instance of a given shape at any one time. This allows type equality to
2732 be performed with address equality of the Type Instance. That is, given two
2733 <tt>Type*</tt> values, the types are identical if the pointers are identical.
2737 <!-- _______________________________________________________________________ -->
2738 <div class="doc_subsubsection">
2739 <a name="m_Type">Important Public Methods</a>
2742 <div class="doc_text">
2745 <li><tt>bool isInteger() const</tt>: Returns true for any integer type.</li>
2747 <li><tt>bool isFloatingPoint()</tt>: Return true if this is one of the two
2748 floating point types.</li>
2750 <li><tt>bool isAbstract()</tt>: Return true if the type is abstract (contains
2751 an OpaqueType anywhere in its definition).</li>
2753 <li><tt>bool isSized()</tt>: Return true if the type has known size. Things
2754 that don't have a size are abstract types, labels and void.</li>
2759 <!-- _______________________________________________________________________ -->
2760 <div class="doc_subsubsection">
2761 <a name="derivedtypes">Important Derived Types</a>
2763 <div class="doc_text">
2765 <dt><tt>IntegerType</tt></dt>
2766 <dd>Subclass of DerivedType that represents integer types of any bit width.
2767 Any bit width between <tt>IntegerType::MIN_INT_BITS</tt> (1) and
2768 <tt>IntegerType::MAX_INT_BITS</tt> (~8 million) can be represented.
2770 <li><tt>static const IntegerType* get(unsigned NumBits)</tt>: get an integer
2771 type of a specific bit width.</li>
2772 <li><tt>unsigned getBitWidth() const</tt>: Get the bit width of an integer
2776 <dt><tt>SequentialType</tt></dt>
2777 <dd>This is subclassed by ArrayType and PointerType
2779 <li><tt>const Type * getElementType() const</tt>: Returns the type of each
2780 of the elements in the sequential type. </li>
2783 <dt><tt>ArrayType</tt></dt>
2784 <dd>This is a subclass of SequentialType and defines the interface for array
2787 <li><tt>unsigned getNumElements() const</tt>: Returns the number of
2788 elements in the array. </li>
2791 <dt><tt>PointerType</tt></dt>
2792 <dd>Subclass of SequentialType for pointer types.</dd>
2793 <dt><tt>VectorType</tt></dt>
2794 <dd>Subclass of SequentialType for vector types. A
2795 vector type is similar to an ArrayType but is distinguished because it is
2796 a first class type wherease ArrayType is not. Vector types are used for
2797 vector operations and are usually small vectors of of an integer or floating
2799 <dt><tt>StructType</tt></dt>
2800 <dd>Subclass of DerivedTypes for struct types.</dd>
2801 <dt><tt><a name="FunctionType">FunctionType</a></tt></dt>
2802 <dd>Subclass of DerivedTypes for function types.
2804 <li><tt>bool isVarArg() const</tt>: Returns true if its a vararg
2806 <li><tt> const Type * getReturnType() const</tt>: Returns the
2807 return type of the function.</li>
2808 <li><tt>const Type * getParamType (unsigned i)</tt>: Returns
2809 the type of the ith parameter.</li>
2810 <li><tt> const unsigned getNumParams() const</tt>: Returns the
2811 number of formal parameters.</li>
2814 <dt><tt>OpaqueType</tt></dt>
2815 <dd>Sublcass of DerivedType for abstract types. This class
2816 defines no content and is used as a placeholder for some other type. Note
2817 that OpaqueType is used (temporarily) during type resolution for forward
2818 references of types. Once the referenced type is resolved, the OpaqueType
2819 is replaced with the actual type. OpaqueType can also be used for data
2820 abstraction. At link time opaque types can be resolved to actual types
2821 of the same name.</dd>
2827 <!-- ======================================================================= -->
2828 <div class="doc_subsection">
2829 <a name="Module">The <tt>Module</tt> class</a>
2832 <div class="doc_text">
2835 href="/doxygen/Module_8h-source.html">llvm/Module.h</a>"</tt><br> doxygen info:
2836 <a href="/doxygen/classllvm_1_1Module.html">Module Class</a></p>
2838 <p>The <tt>Module</tt> class represents the top level structure present in LLVM
2839 programs. An LLVM module is effectively either a translation unit of the
2840 original program or a combination of several translation units merged by the
2841 linker. The <tt>Module</tt> class keeps track of a list of <a
2842 href="#Function"><tt>Function</tt></a>s, a list of <a
2843 href="#GlobalVariable"><tt>GlobalVariable</tt></a>s, and a <a
2844 href="#SymbolTable"><tt>SymbolTable</tt></a>. Additionally, it contains a few
2845 helpful member functions that try to make common operations easy.</p>
2849 <!-- _______________________________________________________________________ -->
2850 <div class="doc_subsubsection">
2851 <a name="m_Module">Important Public Members of the <tt>Module</tt> class</a>
2854 <div class="doc_text">
2857 <li><tt>Module::Module(std::string name = "")</tt></li>
2860 <p>Constructing a <a href="#Module">Module</a> is easy. You can optionally
2861 provide a name for it (probably based on the name of the translation unit).</p>
2864 <li><tt>Module::iterator</tt> - Typedef for function list iterator<br>
2865 <tt>Module::const_iterator</tt> - Typedef for const_iterator.<br>
2867 <tt>begin()</tt>, <tt>end()</tt>
2868 <tt>size()</tt>, <tt>empty()</tt>
2870 <p>These are forwarding methods that make it easy to access the contents of
2871 a <tt>Module</tt> object's <a href="#Function"><tt>Function</tt></a>
2874 <li><tt>Module::FunctionListType &getFunctionList()</tt>
2876 <p> Returns the list of <a href="#Function"><tt>Function</tt></a>s. This is
2877 necessary to use when you need to update the list or perform a complex
2878 action that doesn't have a forwarding method.</p>
2880 <p><!-- Global Variable --></p></li>
2886 <li><tt>Module::global_iterator</tt> - Typedef for global variable list iterator<br>
2888 <tt>Module::const_global_iterator</tt> - Typedef for const_iterator.<br>
2890 <tt>global_begin()</tt>, <tt>global_end()</tt>
2891 <tt>global_size()</tt>, <tt>global_empty()</tt>
2893 <p> These are forwarding methods that make it easy to access the contents of
2894 a <tt>Module</tt> object's <a
2895 href="#GlobalVariable"><tt>GlobalVariable</tt></a> list.</p></li>
2897 <li><tt>Module::GlobalListType &getGlobalList()</tt>
2899 <p>Returns the list of <a
2900 href="#GlobalVariable"><tt>GlobalVariable</tt></a>s. This is necessary to
2901 use when you need to update the list or perform a complex action that
2902 doesn't have a forwarding method.</p>
2904 <p><!-- Symbol table stuff --> </p></li>
2910 <li><tt><a href="#SymbolTable">SymbolTable</a> *getSymbolTable()</tt>
2912 <p>Return a reference to the <a href="#SymbolTable"><tt>SymbolTable</tt></a>
2913 for this <tt>Module</tt>.</p>
2915 <p><!-- Convenience methods --></p></li>
2921 <li><tt><a href="#Function">Function</a> *getFunction(const std::string
2922 &Name, const <a href="#FunctionType">FunctionType</a> *Ty)</tt>
2924 <p>Look up the specified function in the <tt>Module</tt> <a
2925 href="#SymbolTable"><tt>SymbolTable</tt></a>. If it does not exist, return
2926 <tt>null</tt>.</p></li>
2928 <li><tt><a href="#Function">Function</a> *getOrInsertFunction(const
2929 std::string &Name, const <a href="#FunctionType">FunctionType</a> *T)</tt>
2931 <p>Look up the specified function in the <tt>Module</tt> <a
2932 href="#SymbolTable"><tt>SymbolTable</tt></a>. If it does not exist, add an
2933 external declaration for the function and return it.</p></li>
2935 <li><tt>std::string getTypeName(const <a href="#Type">Type</a> *Ty)</tt>
2937 <p>If there is at least one entry in the <a
2938 href="#SymbolTable"><tt>SymbolTable</tt></a> for the specified <a
2939 href="#Type"><tt>Type</tt></a>, return it. Otherwise return the empty
2942 <li><tt>bool addTypeName(const std::string &Name, const <a
2943 href="#Type">Type</a> *Ty)</tt>
2945 <p>Insert an entry in the <a href="#SymbolTable"><tt>SymbolTable</tt></a>
2946 mapping <tt>Name</tt> to <tt>Ty</tt>. If there is already an entry for this
2947 name, true is returned and the <a
2948 href="#SymbolTable"><tt>SymbolTable</tt></a> is not modified.</p></li>
2954 <!-- ======================================================================= -->
2955 <div class="doc_subsection">
2956 <a name="Value">The <tt>Value</tt> class</a>
2959 <div class="doc_text">
2961 <p><tt>#include "<a href="/doxygen/Value_8h-source.html">llvm/Value.h</a>"</tt>
2963 doxygen info: <a href="/doxygen/classllvm_1_1Value.html">Value Class</a></p>
2965 <p>The <tt>Value</tt> class is the most important class in the LLVM Source
2966 base. It represents a typed value that may be used (among other things) as an
2967 operand to an instruction. There are many different types of <tt>Value</tt>s,
2968 such as <a href="#Constant"><tt>Constant</tt></a>s,<a
2969 href="#Argument"><tt>Argument</tt></a>s. Even <a
2970 href="#Instruction"><tt>Instruction</tt></a>s and <a
2971 href="#Function"><tt>Function</tt></a>s are <tt>Value</tt>s.</p>
2973 <p>A particular <tt>Value</tt> may be used many times in the LLVM representation
2974 for a program. For example, an incoming argument to a function (represented
2975 with an instance of the <a href="#Argument">Argument</a> class) is "used" by
2976 every instruction in the function that references the argument. To keep track
2977 of this relationship, the <tt>Value</tt> class keeps a list of all of the <a
2978 href="#User"><tt>User</tt></a>s that is using it (the <a
2979 href="#User"><tt>User</tt></a> class is a base class for all nodes in the LLVM
2980 graph that can refer to <tt>Value</tt>s). This use list is how LLVM represents
2981 def-use information in the program, and is accessible through the <tt>use_</tt>*
2982 methods, shown below.</p>
2984 <p>Because LLVM is a typed representation, every LLVM <tt>Value</tt> is typed,
2985 and this <a href="#Type">Type</a> is available through the <tt>getType()</tt>
2986 method. In addition, all LLVM values can be named. The "name" of the
2987 <tt>Value</tt> is a symbolic string printed in the LLVM code:</p>
2989 <div class="doc_code">
2991 %<b>foo</b> = add i32 1, 2
2995 <p><a name="nameWarning">The name of this instruction is "foo".</a> <b>NOTE</b>
2996 that the name of any value may be missing (an empty string), so names should
2997 <b>ONLY</b> be used for debugging (making the source code easier to read,
2998 debugging printouts), they should not be used to keep track of values or map
2999 between them. For this purpose, use a <tt>std::map</tt> of pointers to the
3000 <tt>Value</tt> itself instead.</p>
3002 <p>One important aspect of LLVM is that there is no distinction between an SSA
3003 variable and the operation that produces it. Because of this, any reference to
3004 the value produced by an instruction (or the value available as an incoming
3005 argument, for example) is represented as a direct pointer to the instance of
3007 represents this value. Although this may take some getting used to, it
3008 simplifies the representation and makes it easier to manipulate.</p>
3012 <!-- _______________________________________________________________________ -->
3013 <div class="doc_subsubsection">
3014 <a name="m_Value">Important Public Members of the <tt>Value</tt> class</a>
3017 <div class="doc_text">
3020 <li><tt>Value::use_iterator</tt> - Typedef for iterator over the
3022 <tt>Value::use_const_iterator</tt> - Typedef for const_iterator over
3024 <tt>unsigned use_size()</tt> - Returns the number of users of the
3026 <tt>bool use_empty()</tt> - Returns true if there are no users.<br>
3027 <tt>use_iterator use_begin()</tt> - Get an iterator to the start of
3029 <tt>use_iterator use_end()</tt> - Get an iterator to the end of the
3031 <tt><a href="#User">User</a> *use_back()</tt> - Returns the last
3032 element in the list.
3033 <p> These methods are the interface to access the def-use
3034 information in LLVM. As with all other iterators in LLVM, the naming
3035 conventions follow the conventions defined by the <a href="#stl">STL</a>.</p>
3037 <li><tt><a href="#Type">Type</a> *getType() const</tt>
3038 <p>This method returns the Type of the Value.</p>
3040 <li><tt>bool hasName() const</tt><br>
3041 <tt>std::string getName() const</tt><br>
3042 <tt>void setName(const std::string &Name)</tt>
3043 <p> This family of methods is used to access and assign a name to a <tt>Value</tt>,
3044 be aware of the <a href="#nameWarning">precaution above</a>.</p>
3046 <li><tt>void replaceAllUsesWith(Value *V)</tt>
3048 <p>This method traverses the use list of a <tt>Value</tt> changing all <a
3049 href="#User"><tt>User</tt>s</a> of the current value to refer to
3050 "<tt>V</tt>" instead. For example, if you detect that an instruction always
3051 produces a constant value (for example through constant folding), you can
3052 replace all uses of the instruction with the constant like this:</p>
3054 <div class="doc_code">
3056 Inst->replaceAllUsesWith(ConstVal);
3064 <!-- ======================================================================= -->
3065 <div class="doc_subsection">
3066 <a name="User">The <tt>User</tt> class</a>
3069 <div class="doc_text">
3072 <tt>#include "<a href="/doxygen/User_8h-source.html">llvm/User.h</a>"</tt><br>
3073 doxygen info: <a href="/doxygen/classllvm_1_1User.html">User Class</a><br>
3074 Superclass: <a href="#Value"><tt>Value</tt></a></p>
3076 <p>The <tt>User</tt> class is the common base class of all LLVM nodes that may
3077 refer to <a href="#Value"><tt>Value</tt></a>s. It exposes a list of "Operands"
3078 that are all of the <a href="#Value"><tt>Value</tt></a>s that the User is
3079 referring to. The <tt>User</tt> class itself is a subclass of
3082 <p>The operands of a <tt>User</tt> point directly to the LLVM <a
3083 href="#Value"><tt>Value</tt></a> that it refers to. Because LLVM uses Static
3084 Single Assignment (SSA) form, there can only be one definition referred to,
3085 allowing this direct connection. This connection provides the use-def
3086 information in LLVM.</p>
3090 <!-- _______________________________________________________________________ -->
3091 <div class="doc_subsubsection">
3092 <a name="m_User">Important Public Members of the <tt>User</tt> class</a>
3095 <div class="doc_text">
3097 <p>The <tt>User</tt> class exposes the operand list in two ways: through
3098 an index access interface and through an iterator based interface.</p>
3101 <li><tt>Value *getOperand(unsigned i)</tt><br>
3102 <tt>unsigned getNumOperands()</tt>
3103 <p> These two methods expose the operands of the <tt>User</tt> in a
3104 convenient form for direct access.</p></li>
3106 <li><tt>User::op_iterator</tt> - Typedef for iterator over the operand
3108 <tt>op_iterator op_begin()</tt> - Get an iterator to the start of
3109 the operand list.<br>
3110 <tt>op_iterator op_end()</tt> - Get an iterator to the end of the
3112 <p> Together, these methods make up the iterator based interface to
3113 the operands of a <tt>User</tt>.</p></li>
3118 <!-- ======================================================================= -->
3119 <div class="doc_subsection">
3120 <a name="Instruction">The <tt>Instruction</tt> class</a>
3123 <div class="doc_text">
3125 <p><tt>#include "</tt><tt><a
3126 href="/doxygen/Instruction_8h-source.html">llvm/Instruction.h</a>"</tt><br>
3127 doxygen info: <a href="/doxygen/classllvm_1_1Instruction.html">Instruction Class</a><br>
3128 Superclasses: <a href="#User"><tt>User</tt></a>, <a
3129 href="#Value"><tt>Value</tt></a></p>
3131 <p>The <tt>Instruction</tt> class is the common base class for all LLVM
3132 instructions. It provides only a few methods, but is a very commonly used
3133 class. The primary data tracked by the <tt>Instruction</tt> class itself is the
3134 opcode (instruction type) and the parent <a
3135 href="#BasicBlock"><tt>BasicBlock</tt></a> the <tt>Instruction</tt> is embedded
3136 into. To represent a specific type of instruction, one of many subclasses of
3137 <tt>Instruction</tt> are used.</p>
3139 <p> Because the <tt>Instruction</tt> class subclasses the <a
3140 href="#User"><tt>User</tt></a> class, its operands can be accessed in the same
3141 way as for other <a href="#User"><tt>User</tt></a>s (with the
3142 <tt>getOperand()</tt>/<tt>getNumOperands()</tt> and
3143 <tt>op_begin()</tt>/<tt>op_end()</tt> methods).</p> <p> An important file for
3144 the <tt>Instruction</tt> class is the <tt>llvm/Instruction.def</tt> file. This
3145 file contains some meta-data about the various different types of instructions
3146 in LLVM. It describes the enum values that are used as opcodes (for example
3147 <tt>Instruction::Add</tt> and <tt>Instruction::ICmp</tt>), as well as the
3148 concrete sub-classes of <tt>Instruction</tt> that implement the instruction (for
3149 example <tt><a href="#BinaryOperator">BinaryOperator</a></tt> and <tt><a
3150 href="#CmpInst">CmpInst</a></tt>). Unfortunately, the use of macros in
3151 this file confuses doxygen, so these enum values don't show up correctly in the
3152 <a href="/doxygen/classllvm_1_1Instruction.html">doxygen output</a>.</p>
3156 <!-- _______________________________________________________________________ -->
3157 <div class="doc_subsubsection">
3158 <a name="s_Instruction">Important Subclasses of the <tt>Instruction</tt>
3161 <div class="doc_text">
3163 <li><tt><a name="BinaryOperator">BinaryOperator</a></tt>
3164 <p>This subclasses represents all two operand instructions whose operands
3165 must be the same type, except for the comparison instructions.</p></li>
3166 <li><tt><a name="CastInst">CastInst</a></tt>
3167 <p>This subclass is the parent of the 12 casting instructions. It provides
3168 common operations on cast instructions.</p>
3169 <li><tt><a name="CmpInst">CmpInst</a></tt>
3170 <p>This subclass respresents the two comparison instructions,
3171 <a href="LangRef.html#i_icmp">ICmpInst</a> (integer opreands), and
3172 <a href="LangRef.html#i_fcmp">FCmpInst</a> (floating point operands).</p>
3173 <li><tt><a name="TerminatorInst">TerminatorInst</a></tt>
3174 <p>This subclass is the parent of all terminator instructions (those which
3175 can terminate a block).</p>
3179 <!-- _______________________________________________________________________ -->
3180 <div class="doc_subsubsection">
3181 <a name="m_Instruction">Important Public Members of the <tt>Instruction</tt>
3185 <div class="doc_text">
3188 <li><tt><a href="#BasicBlock">BasicBlock</a> *getParent()</tt>
3189 <p>Returns the <a href="#BasicBlock"><tt>BasicBlock</tt></a> that
3190 this <tt>Instruction</tt> is embedded into.</p></li>
3191 <li><tt>bool mayWriteToMemory()</tt>
3192 <p>Returns true if the instruction writes to memory, i.e. it is a
3193 <tt>call</tt>,<tt>free</tt>,<tt>invoke</tt>, or <tt>store</tt>.</p></li>
3194 <li><tt>unsigned getOpcode()</tt>
3195 <p>Returns the opcode for the <tt>Instruction</tt>.</p></li>
3196 <li><tt><a href="#Instruction">Instruction</a> *clone() const</tt>
3197 <p>Returns another instance of the specified instruction, identical
3198 in all ways to the original except that the instruction has no parent
3199 (ie it's not embedded into a <a href="#BasicBlock"><tt>BasicBlock</tt></a>),
3200 and it has no name</p></li>
3205 <!-- ======================================================================= -->
3206 <div class="doc_subsection">
3207 <a name="Constant">The <tt>Constant</tt> class and subclasses</a>
3210 <div class="doc_text">
3212 <p>Constant represents a base class for different types of constants. It
3213 is subclassed by ConstantInt, ConstantArray, etc. for representing
3214 the various types of Constants. <a href="#GlobalValue">GlobalValue</a> is also
3215 a subclass, which represents the address of a global variable or function.
3220 <!-- _______________________________________________________________________ -->
3221 <div class="doc_subsubsection">Important Subclasses of Constant </div>
3222 <div class="doc_text">
3224 <li>ConstantInt : This subclass of Constant represents an integer constant of
3227 <li><tt>const APInt& getValue() const</tt>: Returns the underlying
3228 value of this constant, an APInt value.</li>
3229 <li><tt>int64_t getSExtValue() const</tt>: Converts the underlying APInt
3230 value to an int64_t via sign extension. If the value (not the bit width)
3231 of the APInt is too large to fit in an int64_t, an assertion will result.
3232 For this reason, use of this method is discouraged.</li>
3233 <li><tt>uint64_t getZExtValue() const</tt>: Converts the underlying APInt
3234 value to a uint64_t via zero extension. IF the value (not the bit width)
3235 of the APInt is too large to fit in a uint64_t, an assertion will result.
3236 For this reason, use of this method is discouraged.</li>
3237 <li><tt>static ConstantInt* get(const APInt& Val)</tt>: Returns the
3238 ConstantInt object that represents the value provided by <tt>Val</tt>.
3239 The type is implied as the IntegerType that corresponds to the bit width
3240 of <tt>Val</tt>.</li>
3241 <li><tt>static ConstantInt* get(const Type *Ty, uint64_t Val)</tt>:
3242 Returns the ConstantInt object that represents the value provided by
3243 <tt>Val</tt> for integer type <tt>Ty</tt>.</li>
3246 <li>ConstantFP : This class represents a floating point constant.
3248 <li><tt>double getValue() const</tt>: Returns the underlying value of
3249 this constant. </li>
3252 <li>ConstantArray : This represents a constant array.
3254 <li><tt>const std::vector<Use> &getValues() const</tt>: Returns
3255 a vector of component constants that makeup this array. </li>
3258 <li>ConstantStruct : This represents a constant struct.
3260 <li><tt>const std::vector<Use> &getValues() const</tt>: Returns
3261 a vector of component constants that makeup this array. </li>
3264 <li>GlobalValue : This represents either a global variable or a function. In
3265 either case, the value is a constant fixed address (after linking).
3271 <!-- ======================================================================= -->
3272 <div class="doc_subsection">
3273 <a name="GlobalValue">The <tt>GlobalValue</tt> class</a>
3276 <div class="doc_text">
3279 href="/doxygen/GlobalValue_8h-source.html">llvm/GlobalValue.h</a>"</tt><br>
3280 doxygen info: <a href="/doxygen/classllvm_1_1GlobalValue.html">GlobalValue
3282 Superclasses: <a href="#Constant"><tt>Constant</tt></a>,
3283 <a href="#User"><tt>User</tt></a>, <a href="#Value"><tt>Value</tt></a></p>
3285 <p>Global values (<a href="#GlobalVariable"><tt>GlobalVariable</tt></a>s or <a
3286 href="#Function"><tt>Function</tt></a>s) are the only LLVM values that are
3287 visible in the bodies of all <a href="#Function"><tt>Function</tt></a>s.
3288 Because they are visible at global scope, they are also subject to linking with
3289 other globals defined in different translation units. To control the linking
3290 process, <tt>GlobalValue</tt>s know their linkage rules. Specifically,
3291 <tt>GlobalValue</tt>s know whether they have internal or external linkage, as
3292 defined by the <tt>LinkageTypes</tt> enumeration.</p>
3294 <p>If a <tt>GlobalValue</tt> has internal linkage (equivalent to being
3295 <tt>static</tt> in C), it is not visible to code outside the current translation
3296 unit, and does not participate in linking. If it has external linkage, it is
3297 visible to external code, and does participate in linking. In addition to
3298 linkage information, <tt>GlobalValue</tt>s keep track of which <a
3299 href="#Module"><tt>Module</tt></a> they are currently part of.</p>
3301 <p>Because <tt>GlobalValue</tt>s are memory objects, they are always referred to
3302 by their <b>address</b>. As such, the <a href="#Type"><tt>Type</tt></a> of a
3303 global is always a pointer to its contents. It is important to remember this
3304 when using the <tt>GetElementPtrInst</tt> instruction because this pointer must
3305 be dereferenced first. For example, if you have a <tt>GlobalVariable</tt> (a
3306 subclass of <tt>GlobalValue)</tt> that is an array of 24 ints, type <tt>[24 x
3307 i32]</tt>, then the <tt>GlobalVariable</tt> is a pointer to that array. Although
3308 the address of the first element of this array and the value of the
3309 <tt>GlobalVariable</tt> are the same, they have different types. The
3310 <tt>GlobalVariable</tt>'s type is <tt>[24 x i32]</tt>. The first element's type
3311 is <tt>i32.</tt> Because of this, accessing a global value requires you to
3312 dereference the pointer with <tt>GetElementPtrInst</tt> first, then its elements
3313 can be accessed. This is explained in the <a href="LangRef.html#globalvars">LLVM
3314 Language Reference Manual</a>.</p>
3318 <!-- _______________________________________________________________________ -->
3319 <div class="doc_subsubsection">
3320 <a name="m_GlobalValue">Important Public Members of the <tt>GlobalValue</tt>
3324 <div class="doc_text">
3327 <li><tt>bool hasInternalLinkage() const</tt><br>
3328 <tt>bool hasExternalLinkage() const</tt><br>
3329 <tt>void setInternalLinkage(bool HasInternalLinkage)</tt>
3330 <p> These methods manipulate the linkage characteristics of the <tt>GlobalValue</tt>.</p>
3333 <li><tt><a href="#Module">Module</a> *getParent()</tt>
3334 <p> This returns the <a href="#Module"><tt>Module</tt></a> that the
3335 GlobalValue is currently embedded into.</p></li>
3340 <!-- ======================================================================= -->
3341 <div class="doc_subsection">
3342 <a name="Function">The <tt>Function</tt> class</a>
3345 <div class="doc_text">
3348 href="/doxygen/Function_8h-source.html">llvm/Function.h</a>"</tt><br> doxygen
3349 info: <a href="/doxygen/classllvm_1_1Function.html">Function Class</a><br>
3350 Superclasses: <a href="#GlobalValue"><tt>GlobalValue</tt></a>,
3351 <a href="#Constant"><tt>Constant</tt></a>,
3352 <a href="#User"><tt>User</tt></a>,
3353 <a href="#Value"><tt>Value</tt></a></p>
3355 <p>The <tt>Function</tt> class represents a single procedure in LLVM. It is
3356 actually one of the more complex classes in the LLVM heirarchy because it must
3357 keep track of a large amount of data. The <tt>Function</tt> class keeps track
3358 of a list of <a href="#BasicBlock"><tt>BasicBlock</tt></a>s, a list of formal
3359 <a href="#Argument"><tt>Argument</tt></a>s, and a
3360 <a href="#SymbolTable"><tt>SymbolTable</tt></a>.</p>
3362 <p>The list of <a href="#BasicBlock"><tt>BasicBlock</tt></a>s is the most
3363 commonly used part of <tt>Function</tt> objects. The list imposes an implicit
3364 ordering of the blocks in the function, which indicate how the code will be
3365 layed out by the backend. Additionally, the first <a
3366 href="#BasicBlock"><tt>BasicBlock</tt></a> is the implicit entry node for the
3367 <tt>Function</tt>. It is not legal in LLVM to explicitly branch to this initial
3368 block. There are no implicit exit nodes, and in fact there may be multiple exit
3369 nodes from a single <tt>Function</tt>. If the <a
3370 href="#BasicBlock"><tt>BasicBlock</tt></a> list is empty, this indicates that
3371 the <tt>Function</tt> is actually a function declaration: the actual body of the
3372 function hasn't been linked in yet.</p>
3374 <p>In addition to a list of <a href="#BasicBlock"><tt>BasicBlock</tt></a>s, the
3375 <tt>Function</tt> class also keeps track of the list of formal <a
3376 href="#Argument"><tt>Argument</tt></a>s that the function receives. This
3377 container manages the lifetime of the <a href="#Argument"><tt>Argument</tt></a>
3378 nodes, just like the <a href="#BasicBlock"><tt>BasicBlock</tt></a> list does for
3379 the <a href="#BasicBlock"><tt>BasicBlock</tt></a>s.</p>
3381 <p>The <a href="#SymbolTable"><tt>SymbolTable</tt></a> is a very rarely used
3382 LLVM feature that is only used when you have to look up a value by name. Aside
3383 from that, the <a href="#SymbolTable"><tt>SymbolTable</tt></a> is used
3384 internally to make sure that there are not conflicts between the names of <a
3385 href="#Instruction"><tt>Instruction</tt></a>s, <a
3386 href="#BasicBlock"><tt>BasicBlock</tt></a>s, or <a
3387 href="#Argument"><tt>Argument</tt></a>s in the function body.</p>
3389 <p>Note that <tt>Function</tt> is a <a href="#GlobalValue">GlobalValue</a>
3390 and therefore also a <a href="#Constant">Constant</a>. The value of the function
3391 is its address (after linking) which is guaranteed to be constant.</p>
3394 <!-- _______________________________________________________________________ -->
3395 <div class="doc_subsubsection">
3396 <a name="m_Function">Important Public Members of the <tt>Function</tt>
3400 <div class="doc_text">
3403 <li><tt>Function(const </tt><tt><a href="#FunctionType">FunctionType</a>
3404 *Ty, LinkageTypes Linkage, const std::string &N = "", Module* Parent = 0)</tt>
3406 <p>Constructor used when you need to create new <tt>Function</tt>s to add
3407 the the program. The constructor must specify the type of the function to
3408 create and what type of linkage the function should have. The <a
3409 href="#FunctionType"><tt>FunctionType</tt></a> argument
3410 specifies the formal arguments and return value for the function. The same
3411 <a href="#FunctionType"><tt>FunctionType</tt></a> value can be used to
3412 create multiple functions. The <tt>Parent</tt> argument specifies the Module
3413 in which the function is defined. If this argument is provided, the function
3414 will automatically be inserted into that module's list of
3417 <li><tt>bool isDeclaration()</tt>
3419 <p>Return whether or not the <tt>Function</tt> has a body defined. If the
3420 function is "external", it does not have a body, and thus must be resolved
3421 by linking with a function defined in a different translation unit.</p></li>
3423 <li><tt>Function::iterator</tt> - Typedef for basic block list iterator<br>
3424 <tt>Function::const_iterator</tt> - Typedef for const_iterator.<br>
3426 <tt>begin()</tt>, <tt>end()</tt>
3427 <tt>size()</tt>, <tt>empty()</tt>
3429 <p>These are forwarding methods that make it easy to access the contents of
3430 a <tt>Function</tt> object's <a href="#BasicBlock"><tt>BasicBlock</tt></a>
3433 <li><tt>Function::BasicBlockListType &getBasicBlockList()</tt>
3435 <p>Returns the list of <a href="#BasicBlock"><tt>BasicBlock</tt></a>s. This
3436 is necessary to use when you need to update the list or perform a complex
3437 action that doesn't have a forwarding method.</p></li>
3439 <li><tt>Function::arg_iterator</tt> - Typedef for the argument list
3441 <tt>Function::const_arg_iterator</tt> - Typedef for const_iterator.<br>
3443 <tt>arg_begin()</tt>, <tt>arg_end()</tt>
3444 <tt>arg_size()</tt>, <tt>arg_empty()</tt>
3446 <p>These are forwarding methods that make it easy to access the contents of
3447 a <tt>Function</tt> object's <a href="#Argument"><tt>Argument</tt></a>
3450 <li><tt>Function::ArgumentListType &getArgumentList()</tt>
3452 <p>Returns the list of <a href="#Argument"><tt>Argument</tt></a>s. This is
3453 necessary to use when you need to update the list or perform a complex
3454 action that doesn't have a forwarding method.</p></li>
3456 <li><tt><a href="#BasicBlock">BasicBlock</a> &getEntryBlock()</tt>
3458 <p>Returns the entry <a href="#BasicBlock"><tt>BasicBlock</tt></a> for the
3459 function. Because the entry block for the function is always the first
3460 block, this returns the first block of the <tt>Function</tt>.</p></li>
3462 <li><tt><a href="#Type">Type</a> *getReturnType()</tt><br>
3463 <tt><a href="#FunctionType">FunctionType</a> *getFunctionType()</tt>
3465 <p>This traverses the <a href="#Type"><tt>Type</tt></a> of the
3466 <tt>Function</tt> and returns the return type of the function, or the <a
3467 href="#FunctionType"><tt>FunctionType</tt></a> of the actual
3470 <li><tt><a href="#SymbolTable">SymbolTable</a> *getSymbolTable()</tt>
3472 <p> Return a pointer to the <a href="#SymbolTable"><tt>SymbolTable</tt></a>
3473 for this <tt>Function</tt>.</p></li>
3478 <!-- ======================================================================= -->
3479 <div class="doc_subsection">
3480 <a name="GlobalVariable">The <tt>GlobalVariable</tt> class</a>
3483 <div class="doc_text">
3486 href="/doxygen/GlobalVariable_8h-source.html">llvm/GlobalVariable.h</a>"</tt>
3488 doxygen info: <a href="/doxygen/classllvm_1_1GlobalVariable.html">GlobalVariable
3490 Superclasses: <a href="#GlobalValue"><tt>GlobalValue</tt></a>,
3491 <a href="#Constant"><tt>Constant</tt></a>,
3492 <a href="#User"><tt>User</tt></a>,
3493 <a href="#Value"><tt>Value</tt></a></p>
3495 <p>Global variables are represented with the (suprise suprise)
3496 <tt>GlobalVariable</tt> class. Like functions, <tt>GlobalVariable</tt>s are also
3497 subclasses of <a href="#GlobalValue"><tt>GlobalValue</tt></a>, and as such are
3498 always referenced by their address (global values must live in memory, so their
3499 "name" refers to their constant address). See
3500 <a href="#GlobalValue"><tt>GlobalValue</tt></a> for more on this. Global
3501 variables may have an initial value (which must be a
3502 <a href="#Constant"><tt>Constant</tt></a>), and if they have an initializer,
3503 they may be marked as "constant" themselves (indicating that their contents
3504 never change at runtime).</p>
3507 <!-- _______________________________________________________________________ -->
3508 <div class="doc_subsubsection">
3509 <a name="m_GlobalVariable">Important Public Members of the
3510 <tt>GlobalVariable</tt> class</a>
3513 <div class="doc_text">
3516 <li><tt>GlobalVariable(const </tt><tt><a href="#Type">Type</a> *Ty, bool
3517 isConstant, LinkageTypes& Linkage, <a href="#Constant">Constant</a>
3518 *Initializer = 0, const std::string &Name = "", Module* Parent = 0)</tt>
3520 <p>Create a new global variable of the specified type. If
3521 <tt>isConstant</tt> is true then the global variable will be marked as
3522 unchanging for the program. The Linkage parameter specifies the type of
3523 linkage (internal, external, weak, linkonce, appending) for the variable.
3524 If the linkage is InternalLinkage, WeakAnyLinkage, WeakODRLinkage,
3525 LinkOnceAnyLinkage or LinkOnceODRLinkage, then the resultant
3526 global variable will have internal linkage. AppendingLinkage concatenates
3527 together all instances (in different translation units) of the variable
3528 into a single variable but is only applicable to arrays. See
3529 the <a href="LangRef.html#modulestructure">LLVM Language Reference</a> for
3530 further details on linkage types. Optionally an initializer, a name, and the
3531 module to put the variable into may be specified for the global variable as
3534 <li><tt>bool isConstant() const</tt>
3536 <p>Returns true if this is a global variable that is known not to
3537 be modified at runtime.</p></li>
3539 <li><tt>bool hasInitializer()</tt>
3541 <p>Returns true if this <tt>GlobalVariable</tt> has an intializer.</p></li>
3543 <li><tt><a href="#Constant">Constant</a> *getInitializer()</tt>
3545 <p>Returns the intial value for a <tt>GlobalVariable</tt>. It is not legal
3546 to call this method if there is no initializer.</p></li>
3552 <!-- ======================================================================= -->
3553 <div class="doc_subsection">
3554 <a name="BasicBlock">The <tt>BasicBlock</tt> class</a>
3557 <div class="doc_text">
3560 href="/doxygen/BasicBlock_8h-source.html">llvm/BasicBlock.h</a>"</tt><br>
3561 doxygen info: <a href="/doxygen/structllvm_1_1BasicBlock.html">BasicBlock
3563 Superclass: <a href="#Value"><tt>Value</tt></a></p>
3565 <p>This class represents a single entry multiple exit section of the code,
3566 commonly known as a basic block by the compiler community. The
3567 <tt>BasicBlock</tt> class maintains a list of <a
3568 href="#Instruction"><tt>Instruction</tt></a>s, which form the body of the block.
3569 Matching the language definition, the last element of this list of instructions
3570 is always a terminator instruction (a subclass of the <a
3571 href="#TerminatorInst"><tt>TerminatorInst</tt></a> class).</p>
3573 <p>In addition to tracking the list of instructions that make up the block, the
3574 <tt>BasicBlock</tt> class also keeps track of the <a
3575 href="#Function"><tt>Function</tt></a> that it is embedded into.</p>
3577 <p>Note that <tt>BasicBlock</tt>s themselves are <a
3578 href="#Value"><tt>Value</tt></a>s, because they are referenced by instructions
3579 like branches and can go in the switch tables. <tt>BasicBlock</tt>s have type
3584 <!-- _______________________________________________________________________ -->
3585 <div class="doc_subsubsection">
3586 <a name="m_BasicBlock">Important Public Members of the <tt>BasicBlock</tt>
3590 <div class="doc_text">
3593 <li><tt>BasicBlock(const std::string &Name = "", </tt><tt><a
3594 href="#Function">Function</a> *Parent = 0)</tt>
3596 <p>The <tt>BasicBlock</tt> constructor is used to create new basic blocks for
3597 insertion into a function. The constructor optionally takes a name for the new
3598 block, and a <a href="#Function"><tt>Function</tt></a> to insert it into. If
3599 the <tt>Parent</tt> parameter is specified, the new <tt>BasicBlock</tt> is
3600 automatically inserted at the end of the specified <a
3601 href="#Function"><tt>Function</tt></a>, if not specified, the BasicBlock must be
3602 manually inserted into the <a href="#Function"><tt>Function</tt></a>.</p></li>
3604 <li><tt>BasicBlock::iterator</tt> - Typedef for instruction list iterator<br>
3605 <tt>BasicBlock::const_iterator</tt> - Typedef for const_iterator.<br>
3606 <tt>begin()</tt>, <tt>end()</tt>, <tt>front()</tt>, <tt>back()</tt>,
3607 <tt>size()</tt>, <tt>empty()</tt>
3608 STL-style functions for accessing the instruction list.
3610 <p>These methods and typedefs are forwarding functions that have the same
3611 semantics as the standard library methods of the same names. These methods
3612 expose the underlying instruction list of a basic block in a way that is easy to
3613 manipulate. To get the full complement of container operations (including
3614 operations to update the list), you must use the <tt>getInstList()</tt>
3617 <li><tt>BasicBlock::InstListType &getInstList()</tt>
3619 <p>This method is used to get access to the underlying container that actually
3620 holds the Instructions. This method must be used when there isn't a forwarding
3621 function in the <tt>BasicBlock</tt> class for the operation that you would like
3622 to perform. Because there are no forwarding functions for "updating"
3623 operations, you need to use this if you want to update the contents of a
3624 <tt>BasicBlock</tt>.</p></li>
3626 <li><tt><a href="#Function">Function</a> *getParent()</tt>
3628 <p> Returns a pointer to <a href="#Function"><tt>Function</tt></a> the block is
3629 embedded into, or a null pointer if it is homeless.</p></li>
3631 <li><tt><a href="#TerminatorInst">TerminatorInst</a> *getTerminator()</tt>
3633 <p> Returns a pointer to the terminator instruction that appears at the end of
3634 the <tt>BasicBlock</tt>. If there is no terminator instruction, or if the last
3635 instruction in the block is not a terminator, then a null pointer is
3643 <!-- ======================================================================= -->
3644 <div class="doc_subsection">
3645 <a name="Argument">The <tt>Argument</tt> class</a>
3648 <div class="doc_text">
3650 <p>This subclass of Value defines the interface for incoming formal
3651 arguments to a function. A Function maintains a list of its formal
3652 arguments. An argument has a pointer to the parent Function.</p>
3656 <!-- *********************************************************************** -->
3659 <a href="http://jigsaw.w3.org/css-validator/check/referer"><img
3660 src="http://jigsaw.w3.org/css-validator/images/vcss-blue" alt="Valid CSS"></a>
3661 <a href="http://validator.w3.org/check/referer"><img
3662 src="http://www.w3.org/Icons/valid-html401" alt="Valid HTML 4.01 Strict"></a>
3664 <a href="mailto:dhurjati@cs.uiuc.edu">Dinakar Dhurjati</a> and
3665 <a href="mailto:sabre@nondot.org">Chris Lattner</a><br>
3666 <a href="http://llvm.org">The LLVM Compiler Infrastructure</a><br>
3667 Last modified: $Date$