1<!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.01//EN" 2 "http://www.w3.org/TR/html4/strict.dtd"> 3<html> 4<head> 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"> 8</head> 9<body> 10 11<h1> 12 LLVM Programmer's Manual 13</h1> 14 15<ol> 16 <li><a href="#introduction">Introduction</a></li> 17 <li><a href="#general">General Information</a> 18 <ul> 19 <li><a href="#stl">The C++ Standard Template Library</a></li> 20<!-- 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> 24 25--> 26 </ul> 27 </li> 28 <li><a href="#apis">Important and useful LLVM APIs</a> 29 <ul> 30 <li><a href="#isa">The <tt>isa<></tt>, <tt>cast<></tt> 31and <tt>dyn_cast<></tt> templates</a> </li> 32 <li><a href="#string_apis">Passing strings (the <tt>StringRef</tt> 33and <tt>Twine</tt> classes)</a> 34 <ul> 35 <li><a href="#StringRef">The <tt>StringRef</tt> class</a> </li> 36 <li><a href="#Twine">The <tt>Twine</tt> class</a> </li> 37 </ul> 38 </li> 39 <li><a href="#DEBUG">The <tt>DEBUG()</tt> macro and <tt>-debug</tt> 40option</a> 41 <ul> 42 <li><a href="#DEBUG_TYPE">Fine grained debug info with <tt>DEBUG_TYPE</tt> 43and the <tt>-debug-only</tt> option</a> </li> 44 </ul> 45 </li> 46 <li><a href="#Statistic">The <tt>Statistic</tt> class & <tt>-stats</tt> 47option</a></li> 48<!-- 49 <li>The <tt>InstVisitor</tt> template 50 <li>The general graph API 51--> 52 <li><a href="#ViewGraph">Viewing graphs while debugging code</a></li> 53 </ul> 54 </li> 55 <li><a href="#datastructure">Picking the Right Data Structure for a Task</a> 56 <ul> 57 <li><a href="#ds_sequential">Sequential Containers (std::vector, std::list, etc)</a> 58 <ul> 59 <li><a href="#dss_arrayref">llvm/ADT/ArrayRef.h</a></li> 60 <li><a href="#dss_fixedarrays">Fixed Size Arrays</a></li> 61 <li><a href="#dss_heaparrays">Heap Allocated Arrays</a></li> 62 <li><a href="#dss_tinyptrvector">"llvm/ADT/TinyPtrVector.h"</a></li> 63 <li><a href="#dss_smallvector">"llvm/ADT/SmallVector.h"</a></li> 64 <li><a href="#dss_vector"><vector></a></li> 65 <li><a href="#dss_deque"><deque></a></li> 66 <li><a href="#dss_list"><list></a></li> 67 <li><a href="#dss_ilist">llvm/ADT/ilist.h</a></li> 68 <li><a href="#dss_packedvector">llvm/ADT/PackedVector.h</a></li> 69 <li><a href="#dss_other">Other Sequential Container Options</a></li> 70 </ul></li> 71 <li><a href="#ds_string">String-like containers</a> 72 <ul> 73 <li><a href="#dss_stringref">llvm/ADT/StringRef.h</a></li> 74 <li><a href="#dss_twine">llvm/ADT/Twine.h</a></li> 75 <li><a href="#dss_smallstring">llvm/ADT/SmallString.h</a></li> 76 <li><a href="#dss_stdstring">std::string</a></li> 77 </ul></li> 78 <li><a href="#ds_set">Set-Like Containers (std::set, SmallSet, SetVector, etc)</a> 79 <ul> 80 <li><a href="#dss_sortedvectorset">A sorted 'vector'</a></li> 81 <li><a href="#dss_smallset">"llvm/ADT/SmallSet.h"</a></li> 82 <li><a href="#dss_smallptrset">"llvm/ADT/SmallPtrSet.h"</a></li> 83 <li><a href="#dss_denseset">"llvm/ADT/DenseSet.h"</a></li> 84 <li><a href="#dss_sparseset">"llvm/ADT/SparseSet.h"</a></li> 85 <li><a href="#dss_FoldingSet">"llvm/ADT/FoldingSet.h"</a></li> 86 <li><a href="#dss_set"><set></a></li> 87 <li><a href="#dss_setvector">"llvm/ADT/SetVector.h"</a></li> 88 <li><a href="#dss_uniquevector">"llvm/ADT/UniqueVector.h"</a></li> 89 <li><a href="#dss_immutableset">"llvm/ADT/ImmutableSet.h"</a></li> 90 <li><a href="#dss_otherset">Other Set-Like Container Options</a></li> 91 </ul></li> 92 <li><a href="#ds_map">Map-Like Containers (std::map, DenseMap, etc)</a> 93 <ul> 94 <li><a href="#dss_sortedvectormap">A sorted 'vector'</a></li> 95 <li><a href="#dss_stringmap">"llvm/ADT/StringMap.h"</a></li> 96 <li><a href="#dss_indexedmap">"llvm/ADT/IndexedMap.h"</a></li> 97 <li><a href="#dss_densemap">"llvm/ADT/DenseMap.h"</a></li> 98 <li><a href="#dss_valuemap">"llvm/ADT/ValueMap.h"</a></li> 99 <li><a href="#dss_intervalmap">"llvm/ADT/IntervalMap.h"</a></li> 100 <li><a href="#dss_map"><map></a></li> 101 <li><a href="#dss_inteqclasses">"llvm/ADT/IntEqClasses.h"</a></li> 102 <li><a href="#dss_immutablemap">"llvm/ADT/ImmutableMap.h"</a></li> 103 <li><a href="#dss_othermap">Other Map-Like Container Options</a></li> 104 </ul></li> 105 <li><a href="#ds_bit">BitVector-like containers</a> 106 <ul> 107 <li><a href="#dss_bitvector">A dense bitvector</a></li> 108 <li><a href="#dss_smallbitvector">A "small" dense bitvector</a></li> 109 <li><a href="#dss_sparsebitvector">A sparse bitvector</a></li> 110 </ul></li> 111 </ul> 112 </li> 113 <li><a href="#common">Helpful Hints for Common Operations</a> 114 <ul> 115 <li><a href="#inspection">Basic Inspection and Traversal Routines</a> 116 <ul> 117 <li><a href="#iterate_function">Iterating over the <tt>BasicBlock</tt>s 118in a <tt>Function</tt></a> </li> 119 <li><a href="#iterate_basicblock">Iterating over the <tt>Instruction</tt>s 120in a <tt>BasicBlock</tt></a> </li> 121 <li><a href="#iterate_institer">Iterating over the <tt>Instruction</tt>s 122in a <tt>Function</tt></a> </li> 123 <li><a href="#iterate_convert">Turning an iterator into a 124class pointer</a> </li> 125 <li><a href="#iterate_complex">Finding call sites: a more 126complex example</a> </li> 127 <li><a href="#calls_and_invokes">Treating calls and invokes 128the same way</a> </li> 129 <li><a href="#iterate_chains">Iterating over def-use & 130use-def chains</a> </li> 131 <li><a href="#iterate_preds">Iterating over predecessors & 132successors of blocks</a></li> 133 </ul> 134 </li> 135 <li><a href="#simplechanges">Making simple changes</a> 136 <ul> 137 <li><a href="#schanges_creating">Creating and inserting new 138 <tt>Instruction</tt>s</a> </li> 139 <li><a href="#schanges_deleting">Deleting <tt>Instruction</tt>s</a> </li> 140 <li><a href="#schanges_replacing">Replacing an <tt>Instruction</tt> 141with another <tt>Value</tt></a> </li> 142 <li><a href="#schanges_deletingGV">Deleting <tt>GlobalVariable</tt>s</a> </li> 143 </ul> 144 </li> 145 <li><a href="#create_types">How to Create Types</a></li> 146<!-- 147 <li>Working with the Control Flow Graph 148 <ul> 149 <li>Accessing predecessors and successors of a <tt>BasicBlock</tt> 150 <li> 151 <li> 152 </ul> 153--> 154 </ul> 155 </li> 156 157 <li><a href="#threading">Threads and LLVM</a> 158 <ul> 159 <li><a href="#startmultithreaded">Entering and Exiting Multithreaded Mode 160 </a></li> 161 <li><a href="#shutdown">Ending execution with <tt>llvm_shutdown()</tt></a></li> 162 <li><a href="#managedstatic">Lazy initialization with <tt>ManagedStatic</tt></a></li> 163 <li><a href="#llvmcontext">Achieving Isolation with <tt>LLVMContext</tt></a></li> 164 <li><a href="#jitthreading">Threads and the JIT</a></li> 165 </ul> 166 </li> 167 168 <li><a href="#advanced">Advanced Topics</a> 169 <ul> 170 171 <li><a href="#SymbolTable">The <tt>ValueSymbolTable</tt> class</a></li> 172 <li><a href="#UserLayout">The <tt>User</tt> and owned <tt>Use</tt> classes' memory layout</a></li> 173 </ul></li> 174 175 <li><a href="#coreclasses">The Core LLVM Class Hierarchy Reference</a> 176 <ul> 177 <li><a href="#Type">The <tt>Type</tt> class</a> </li> 178 <li><a href="#Module">The <tt>Module</tt> class</a></li> 179 <li><a href="#Value">The <tt>Value</tt> class</a> 180 <ul> 181 <li><a href="#User">The <tt>User</tt> class</a> 182 <ul> 183 <li><a href="#Instruction">The <tt>Instruction</tt> class</a></li> 184 <li><a href="#Constant">The <tt>Constant</tt> class</a> 185 <ul> 186 <li><a href="#GlobalValue">The <tt>GlobalValue</tt> class</a> 187 <ul> 188 <li><a href="#Function">The <tt>Function</tt> class</a></li> 189 <li><a href="#GlobalVariable">The <tt>GlobalVariable</tt> class</a></li> 190 </ul> 191 </li> 192 </ul> 193 </li> 194 </ul> 195 </li> 196 <li><a href="#BasicBlock">The <tt>BasicBlock</tt> class</a></li> 197 <li><a href="#Argument">The <tt>Argument</tt> class</a></li> 198 </ul> 199 </li> 200 </ul> 201 </li> 202</ol> 203 204<div class="doc_author"> 205 <p>Written by <a href="mailto:sabre@nondot.org">Chris Lattner</a>, 206 <a href="mailto:dhurjati@cs.uiuc.edu">Dinakar Dhurjati</a>, 207 <a href="mailto:ggreif@gmail.com">Gabor Greif</a>, 208 <a href="mailto:jstanley@cs.uiuc.edu">Joel Stanley</a>, 209 <a href="mailto:rspencer@x10sys.com">Reid Spencer</a> and 210 <a href="mailto:owen@apple.com">Owen Anderson</a></p> 211</div> 212 213<!-- *********************************************************************** --> 214<h2> 215 <a name="introduction">Introduction </a> 216</h2> 217<!-- *********************************************************************** --> 218 219<div> 220 221<p>This document is meant to highlight some of the important classes and 222interfaces available in the LLVM source-base. This manual is not 223intended to explain what LLVM is, how it works, and what LLVM code looks 224like. It assumes that you know the basics of LLVM and are interested 225in writing transformations or otherwise analyzing or manipulating the 226code.</p> 227 228<p>This document should get you oriented so that you can find your 229way in the continuously growing source code that makes up the LLVM 230infrastructure. Note that this manual is not intended to serve as a 231replacement for reading the source code, so if you think there should be 232a method in one of these classes to do something, but it's not listed, 233check the source. Links to the <a href="/doxygen/">doxygen</a> sources 234are provided to make this as easy as possible.</p> 235 236<p>The first section of this document describes general information that is 237useful to know when working in the LLVM infrastructure, and the second describes 238the Core LLVM classes. In the future this manual will be extended with 239information describing how to use extension libraries, such as dominator 240information, CFG traversal routines, and useful utilities like the <tt><a 241href="/doxygen/InstVisitor_8h-source.html">InstVisitor</a></tt> template.</p> 242 243</div> 244 245<!-- *********************************************************************** --> 246<h2> 247 <a name="general">General Information</a> 248</h2> 249<!-- *********************************************************************** --> 250 251<div> 252 253<p>This section contains general information that is useful if you are working 254in the LLVM source-base, but that isn't specific to any particular API.</p> 255 256<!-- ======================================================================= --> 257<h3> 258 <a name="stl">The C++ Standard Template Library</a> 259</h3> 260 261<div> 262 263<p>LLVM makes heavy use of the C++ Standard Template Library (STL), 264perhaps much more than you are used to, or have seen before. Because of 265this, you might want to do a little background reading in the 266techniques used and capabilities of the library. There are many good 267pages that discuss the STL, and several books on the subject that you 268can get, so it will not be discussed in this document.</p> 269 270<p>Here are some useful links:</p> 271 272<ol> 273 274<li><a href="http://www.dinkumware.com/manuals/#Standard C++ Library">Dinkumware 275C++ Library reference</a> - an excellent reference for the STL and other parts 276of the standard C++ library.</li> 277 278<li><a href="http://www.tempest-sw.com/cpp/">C++ In a Nutshell</a> - This is an 279O'Reilly book in the making. It has a decent Standard Library 280Reference that rivals Dinkumware's, and is unfortunately no longer free since the 281book has been published.</li> 282 283<li><a href="http://www.parashift.com/c++-faq-lite/">C++ Frequently Asked 284Questions</a></li> 285 286<li><a href="http://www.sgi.com/tech/stl/">SGI's STL Programmer's Guide</a> - 287Contains a useful <a 288href="http://www.sgi.com/tech/stl/stl_introduction.html">Introduction to the 289STL</a>.</li> 290 291<li><a href="http://www.research.att.com/%7Ebs/C++.html">Bjarne Stroustrup's C++ 292Page</a></li> 293 294<li><a href="http://64.78.49.204/"> 295Bruce Eckel's Thinking in C++, 2nd ed. Volume 2 Revision 4.0 (even better, get 296the book).</a></li> 297 298</ol> 299 300<p>You are also encouraged to take a look at the <a 301href="CodingStandards.html">LLVM Coding Standards</a> guide which focuses on how 302to write maintainable code more than where to put your curly braces.</p> 303 304</div> 305 306<!-- ======================================================================= --> 307<h3> 308 <a name="stl">Other useful references</a> 309</h3> 310 311<div> 312 313<ol> 314<li><a href="http://www.fortran-2000.com/ArnaudRecipes/sharedlib.html">Using 315static and shared libraries across platforms</a></li> 316</ol> 317 318</div> 319 320</div> 321 322<!-- *********************************************************************** --> 323<h2> 324 <a name="apis">Important and useful LLVM APIs</a> 325</h2> 326<!-- *********************************************************************** --> 327 328<div> 329 330<p>Here we highlight some LLVM APIs that are generally useful and good to 331know about when writing transformations.</p> 332 333<!-- ======================================================================= --> 334<h3> 335 <a name="isa">The <tt>isa<></tt>, <tt>cast<></tt> and 336 <tt>dyn_cast<></tt> templates</a> 337</h3> 338 339<div> 340 341<p>The LLVM source-base makes extensive use of a custom form of RTTI. 342These templates have many similarities to the C++ <tt>dynamic_cast<></tt> 343operator, but they don't have some drawbacks (primarily stemming from 344the fact that <tt>dynamic_cast<></tt> only works on classes that 345have a v-table). Because they are used so often, you must know what they 346do and how they work. All of these templates are defined in the <a 347 href="/doxygen/Casting_8h-source.html"><tt>llvm/Support/Casting.h</tt></a> 348file (note that you very rarely have to include this file directly).</p> 349 350<dl> 351 <dt><tt>isa<></tt>: </dt> 352 353 <dd><p>The <tt>isa<></tt> operator works exactly like the Java 354 "<tt>instanceof</tt>" operator. It returns true or false depending on whether 355 a reference or pointer points to an instance of the specified class. This can 356 be very useful for constraint checking of various sorts (example below).</p> 357 </dd> 358 359 <dt><tt>cast<></tt>: </dt> 360 361 <dd><p>The <tt>cast<></tt> operator is a "checked cast" operation. It 362 converts a pointer or reference from a base class to a derived class, causing 363 an assertion failure if it is not really an instance of the right type. This 364 should be used in cases where you have some information that makes you believe 365 that something is of the right type. An example of the <tt>isa<></tt> 366 and <tt>cast<></tt> template is:</p> 367 368<div class="doc_code"> 369<pre> 370static bool isLoopInvariant(const <a href="#Value">Value</a> *V, const Loop *L) { 371 if (isa<<a href="#Constant">Constant</a>>(V) || isa<<a href="#Argument">Argument</a>>(V) || isa<<a href="#GlobalValue">GlobalValue</a>>(V)) 372 return true; 373 374 // <i>Otherwise, it must be an instruction...</i> 375 return !L->contains(cast<<a href="#Instruction">Instruction</a>>(V)->getParent()); 376} 377</pre> 378</div> 379 380 <p>Note that you should <b>not</b> use an <tt>isa<></tt> test followed 381 by a <tt>cast<></tt>, for that use the <tt>dyn_cast<></tt> 382 operator.</p> 383 384 </dd> 385 386 <dt><tt>dyn_cast<></tt>:</dt> 387 388 <dd><p>The <tt>dyn_cast<></tt> operator is a "checking cast" operation. 389 It checks to see if the operand is of the specified type, and if so, returns a 390 pointer to it (this operator does not work with references). If the operand is 391 not of the correct type, a null pointer is returned. Thus, this works very 392 much like the <tt>dynamic_cast<></tt> operator in C++, and should be 393 used in the same circumstances. Typically, the <tt>dyn_cast<></tt> 394 operator is used in an <tt>if</tt> statement or some other flow control 395 statement like this:</p> 396 397<div class="doc_code"> 398<pre> 399if (<a href="#AllocationInst">AllocationInst</a> *AI = dyn_cast<<a href="#AllocationInst">AllocationInst</a>>(Val)) { 400 // <i>...</i> 401} 402</pre> 403</div> 404 405 <p>This form of the <tt>if</tt> statement effectively combines together a call 406 to <tt>isa<></tt> and a call to <tt>cast<></tt> into one 407 statement, which is very convenient.</p> 408 409 <p>Note that the <tt>dyn_cast<></tt> operator, like C++'s 410 <tt>dynamic_cast<></tt> or Java's <tt>instanceof</tt> operator, can be 411 abused. In particular, you should not use big chained <tt>if/then/else</tt> 412 blocks to check for lots of different variants of classes. If you find 413 yourself wanting to do this, it is much cleaner and more efficient to use the 414 <tt>InstVisitor</tt> class to dispatch over the instruction type directly.</p> 415 416 </dd> 417 418 <dt><tt>cast_or_null<></tt>: </dt> 419 420 <dd><p>The <tt>cast_or_null<></tt> operator works just like the 421 <tt>cast<></tt> operator, except that it allows for a null pointer as an 422 argument (which it then propagates). This can sometimes be useful, allowing 423 you to combine several null checks into one.</p></dd> 424 425 <dt><tt>dyn_cast_or_null<></tt>: </dt> 426 427 <dd><p>The <tt>dyn_cast_or_null<></tt> operator works just like the 428 <tt>dyn_cast<></tt> operator, except that it allows for a null pointer 429 as an argument (which it then propagates). This can sometimes be useful, 430 allowing you to combine several null checks into one.</p></dd> 431 432</dl> 433 434<p>These five templates can be used with any classes, whether they have a 435v-table or not. To add support for these templates, you simply need to add 436<tt>classof</tt> static methods to the class you are interested casting 437to. Describing this is currently outside the scope of this document, but there 438are lots of examples in the LLVM source base.</p> 439 440</div> 441 442 443<!-- ======================================================================= --> 444<h3> 445 <a name="string_apis">Passing strings (the <tt>StringRef</tt> 446and <tt>Twine</tt> classes)</a> 447</h3> 448 449<div> 450 451<p>Although LLVM generally does not do much string manipulation, we do have 452several important APIs which take strings. Two important examples are the 453Value class -- which has names for instructions, functions, etc. -- and the 454StringMap class which is used extensively in LLVM and Clang.</p> 455 456<p>These are generic classes, and they need to be able to accept strings which 457may have embedded null characters. Therefore, they cannot simply take 458a <tt>const char *</tt>, and taking a <tt>const std::string&</tt> requires 459clients to perform a heap allocation which is usually unnecessary. Instead, 460many LLVM APIs use a <tt>StringRef</tt> or a <tt>const Twine&</tt> for 461passing strings efficiently.</p> 462 463<!-- _______________________________________________________________________ --> 464<h4> 465 <a name="StringRef">The <tt>StringRef</tt> class</a> 466</h4> 467 468<div> 469 470<p>The <tt>StringRef</tt> data type represents a reference to a constant string 471(a character array and a length) and supports the common operations available 472on <tt>std:string</tt>, but does not require heap allocation.</p> 473 474<p>It can be implicitly constructed using a C style null-terminated string, 475an <tt>std::string</tt>, or explicitly with a character pointer and length. 476For example, the <tt>StringRef</tt> find function is declared as:</p> 477 478<pre class="doc_code"> 479 iterator find(StringRef Key); 480</pre> 481 482<p>and clients can call it using any one of:</p> 483 484<pre class="doc_code"> 485 Map.find("foo"); <i>// Lookup "foo"</i> 486 Map.find(std::string("bar")); <i>// Lookup "bar"</i> 487 Map.find(StringRef("\0baz", 4)); <i>// Lookup "\0baz"</i> 488</pre> 489 490<p>Similarly, APIs which need to return a string may return a <tt>StringRef</tt> 491instance, which can be used directly or converted to an <tt>std::string</tt> 492using the <tt>str</tt> member function. See 493"<tt><a href="/doxygen/classllvm_1_1StringRef_8h-source.html">llvm/ADT/StringRef.h</a></tt>" 494for more information.</p> 495 496<p>You should rarely use the <tt>StringRef</tt> class directly, because it contains 497pointers to external memory it is not generally safe to store an instance of the 498class (unless you know that the external storage will not be freed). StringRef is 499small and pervasive enough in LLVM that it should always be passed by value.</p> 500 501</div> 502 503<!-- _______________________________________________________________________ --> 504<h4> 505 <a name="Twine">The <tt>Twine</tt> class</a> 506</h4> 507 508<div> 509 510<p>The <tt>Twine</tt> class is an efficient way for APIs to accept concatenated 511strings. For example, a common LLVM paradigm is to name one instruction based on 512the name of another instruction with a suffix, for example:</p> 513 514<div class="doc_code"> 515<pre> 516 New = CmpInst::Create(<i>...</i>, SO->getName() + ".cmp"); 517</pre> 518</div> 519 520<p>The <tt>Twine</tt> class is effectively a 521lightweight <a href="http://en.wikipedia.org/wiki/Rope_(computer_science)">rope</a> 522which points to temporary (stack allocated) objects. Twines can be implicitly 523constructed as the result of the plus operator applied to strings (i.e., a C 524strings, an <tt>std::string</tt>, or a <tt>StringRef</tt>). The twine delays the 525actual concatenation of strings until it is actually required, at which point 526it can be efficiently rendered directly into a character array. This avoids 527unnecessary heap allocation involved in constructing the temporary results of 528string concatenation. See 529"<tt><a href="/doxygen/classllvm_1_1Twine_8h-source.html">llvm/ADT/Twine.h</a></tt>" 530for more information.</p> 531 532<p>As with a <tt>StringRef</tt>, <tt>Twine</tt> objects point to external memory 533and should almost never be stored or mentioned directly. They are intended 534solely for use when defining a function which should be able to efficiently 535accept concatenated strings.</p> 536 537</div> 538 539</div> 540 541<!-- ======================================================================= --> 542<h3> 543 <a name="DEBUG">The <tt>DEBUG()</tt> macro and <tt>-debug</tt> option</a> 544</h3> 545 546<div> 547 548<p>Often when working on your pass you will put a bunch of debugging printouts 549and other code into your pass. After you get it working, you want to remove 550it, but you may need it again in the future (to work out new bugs that you run 551across).</p> 552 553<p> Naturally, because of this, you don't want to delete the debug printouts, 554but you don't want them to always be noisy. A standard compromise is to comment 555them out, allowing you to enable them if you need them in the future.</p> 556 557<p>The "<tt><a href="/doxygen/Debug_8h-source.html">llvm/Support/Debug.h</a></tt>" 558file provides a macro named <tt>DEBUG()</tt> that is a much nicer solution to 559this problem. Basically, you can put arbitrary code into the argument of the 560<tt>DEBUG</tt> macro, and it is only executed if '<tt>opt</tt>' (or any other 561tool) is run with the '<tt>-debug</tt>' command line argument:</p> 562 563<div class="doc_code"> 564<pre> 565DEBUG(errs() << "I am here!\n"); 566</pre> 567</div> 568 569<p>Then you can run your pass like this:</p> 570 571<div class="doc_code"> 572<pre> 573$ opt < a.bc > /dev/null -mypass 574<i><no output></i> 575$ opt < a.bc > /dev/null -mypass -debug 576I am here! 577</pre> 578</div> 579 580<p>Using the <tt>DEBUG()</tt> macro instead of a home-brewed solution allows you 581to not have to create "yet another" command line option for the debug output for 582your pass. Note that <tt>DEBUG()</tt> macros are disabled for optimized builds, 583so they do not cause a performance impact at all (for the same reason, they 584should also not contain side-effects!).</p> 585 586<p>One additional nice thing about the <tt>DEBUG()</tt> macro is that you can 587enable or disable it directly in gdb. Just use "<tt>set DebugFlag=0</tt>" or 588"<tt>set DebugFlag=1</tt>" from the gdb if the program is running. If the 589program hasn't been started yet, you can always just run it with 590<tt>-debug</tt>.</p> 591 592<!-- _______________________________________________________________________ --> 593<h4> 594 <a name="DEBUG_TYPE">Fine grained debug info with <tt>DEBUG_TYPE</tt> and 595 the <tt>-debug-only</tt> option</a> 596</h4> 597 598<div> 599 600<p>Sometimes you may find yourself in a situation where enabling <tt>-debug</tt> 601just turns on <b>too much</b> information (such as when working on the code 602generator). If you want to enable debug information with more fine-grained 603control, you define the <tt>DEBUG_TYPE</tt> macro and the <tt>-debug</tt> only 604option as follows:</p> 605 606<div class="doc_code"> 607<pre> 608#undef DEBUG_TYPE 609DEBUG(errs() << "No debug type\n"); 610#define DEBUG_TYPE "foo" 611DEBUG(errs() << "'foo' debug type\n"); 612#undef DEBUG_TYPE 613#define DEBUG_TYPE "bar" 614DEBUG(errs() << "'bar' debug type\n")); 615#undef DEBUG_TYPE 616#define DEBUG_TYPE "" 617DEBUG(errs() << "No debug type (2)\n"); 618</pre> 619</div> 620 621<p>Then you can run your pass like this:</p> 622 623<div class="doc_code"> 624<pre> 625$ opt < a.bc > /dev/null -mypass 626<i><no output></i> 627$ opt < a.bc > /dev/null -mypass -debug 628No debug type 629'foo' debug type 630'bar' debug type 631No debug type (2) 632$ opt < a.bc > /dev/null -mypass -debug-only=foo 633'foo' debug type 634$ opt < a.bc > /dev/null -mypass -debug-only=bar 635'bar' debug type 636</pre> 637</div> 638 639<p>Of course, in practice, you should only set <tt>DEBUG_TYPE</tt> at the top of 640a file, to specify the debug type for the entire module (if you do this before 641you <tt>#include "llvm/Support/Debug.h"</tt>, you don't have to insert the ugly 642<tt>#undef</tt>'s). Also, you should use names more meaningful than "foo" and 643"bar", because there is no system in place to ensure that names do not 644conflict. If two different modules use the same string, they will all be turned 645on when the name is specified. This allows, for example, all debug information 646for instruction scheduling to be enabled with <tt>-debug-type=InstrSched</tt>, 647even if the source lives in multiple files.</p> 648 649<p>The <tt>DEBUG_WITH_TYPE</tt> macro is also available for situations where you 650would like to set <tt>DEBUG_TYPE</tt>, but only for one specific <tt>DEBUG</tt> 651statement. It takes an additional first parameter, which is the type to use. For 652example, the preceding example could be written as:</p> 653 654 655<div class="doc_code"> 656<pre> 657DEBUG_WITH_TYPE("", errs() << "No debug type\n"); 658DEBUG_WITH_TYPE("foo", errs() << "'foo' debug type\n"); 659DEBUG_WITH_TYPE("bar", errs() << "'bar' debug type\n")); 660DEBUG_WITH_TYPE("", errs() << "No debug type (2)\n"); 661</pre> 662</div> 663 664</div> 665 666</div> 667 668<!-- ======================================================================= --> 669<h3> 670 <a name="Statistic">The <tt>Statistic</tt> class & <tt>-stats</tt> 671 option</a> 672</h3> 673 674<div> 675 676<p>The "<tt><a 677href="/doxygen/Statistic_8h-source.html">llvm/ADT/Statistic.h</a></tt>" file 678provides a class named <tt>Statistic</tt> that is used as a unified way to 679keep track of what the LLVM compiler is doing and how effective various 680optimizations are. It is useful to see what optimizations are contributing to 681making a particular program run faster.</p> 682 683<p>Often you may run your pass on some big program, and you're interested to see 684how many times it makes a certain transformation. Although you can do this with 685hand inspection, or some ad-hoc method, this is a real pain and not very useful 686for big programs. Using the <tt>Statistic</tt> class makes it very easy to 687keep track of this information, and the calculated information is presented in a 688uniform manner with the rest of the passes being executed.</p> 689 690<p>There are many examples of <tt>Statistic</tt> uses, but the basics of using 691it are as follows:</p> 692 693<ol> 694 <li><p>Define your statistic like this:</p> 695 696<div class="doc_code"> 697<pre> 698#define <a href="#DEBUG_TYPE">DEBUG_TYPE</a> "mypassname" <i>// This goes before any #includes.</i> 699STATISTIC(NumXForms, "The # of times I did stuff"); 700</pre> 701</div> 702 703 <p>The <tt>STATISTIC</tt> macro defines a static variable, whose name is 704 specified by the first argument. The pass name is taken from the DEBUG_TYPE 705 macro, and the description is taken from the second argument. The variable 706 defined ("NumXForms" in this case) acts like an unsigned integer.</p></li> 707 708 <li><p>Whenever you make a transformation, bump the counter:</p> 709 710<div class="doc_code"> 711<pre> 712++NumXForms; // <i>I did stuff!</i> 713</pre> 714</div> 715 716 </li> 717 </ol> 718 719 <p>That's all you have to do. To get '<tt>opt</tt>' to print out the 720 statistics gathered, use the '<tt>-stats</tt>' option:</p> 721 722<div class="doc_code"> 723<pre> 724$ opt -stats -mypassname < program.bc > /dev/null 725<i>... statistics output ...</i> 726</pre> 727</div> 728 729 <p> When running <tt>opt</tt> on a C file from the SPEC benchmark 730suite, it gives a report that looks like this:</p> 731 732<div class="doc_code"> 733<pre> 734 7646 bitcodewriter - Number of normal instructions 735 725 bitcodewriter - Number of oversized instructions 736 129996 bitcodewriter - Number of bitcode bytes written 737 2817 raise - Number of insts DCEd or constprop'd 738 3213 raise - Number of cast-of-self removed 739 5046 raise - Number of expression trees converted 740 75 raise - Number of other getelementptr's formed 741 138 raise - Number of load/store peepholes 742 42 deadtypeelim - Number of unused typenames removed from symtab 743 392 funcresolve - Number of varargs functions resolved 744 27 globaldce - Number of global variables removed 745 2 adce - Number of basic blocks removed 746 134 cee - Number of branches revectored 747 49 cee - Number of setcc instruction eliminated 748 532 gcse - Number of loads removed 749 2919 gcse - Number of instructions removed 750 86 indvars - Number of canonical indvars added 751 87 indvars - Number of aux indvars removed 752 25 instcombine - Number of dead inst eliminate 753 434 instcombine - Number of insts combined 754 248 licm - Number of load insts hoisted 755 1298 licm - Number of insts hoisted to a loop pre-header 756 3 licm - Number of insts hoisted to multiple loop preds (bad, no loop pre-header) 757 75 mem2reg - Number of alloca's promoted 758 1444 cfgsimplify - Number of blocks simplified 759</pre> 760</div> 761 762<p>Obviously, with so many optimizations, having a unified framework for this 763stuff is very nice. Making your pass fit well into the framework makes it more 764maintainable and useful.</p> 765 766</div> 767 768<!-- ======================================================================= --> 769<h3> 770 <a name="ViewGraph">Viewing graphs while debugging code</a> 771</h3> 772 773<div> 774 775<p>Several of the important data structures in LLVM are graphs: for example 776CFGs made out of LLVM <a href="#BasicBlock">BasicBlock</a>s, CFGs made out of 777LLVM <a href="CodeGenerator.html#machinebasicblock">MachineBasicBlock</a>s, and 778<a href="CodeGenerator.html#selectiondag_intro">Instruction Selection 779DAGs</a>. In many cases, while debugging various parts of the compiler, it is 780nice to instantly visualize these graphs.</p> 781 782<p>LLVM provides several callbacks that are available in a debug build to do 783exactly that. If you call the <tt>Function::viewCFG()</tt> method, for example, 784the current LLVM tool will pop up a window containing the CFG for the function 785where each basic block is a node in the graph, and each node contains the 786instructions in the block. Similarly, there also exists 787<tt>Function::viewCFGOnly()</tt> (does not include the instructions), the 788<tt>MachineFunction::viewCFG()</tt> and <tt>MachineFunction::viewCFGOnly()</tt>, 789and the <tt>SelectionDAG::viewGraph()</tt> methods. Within GDB, for example, 790you can usually use something like <tt>call DAG.viewGraph()</tt> to pop 791up a window. Alternatively, you can sprinkle calls to these functions in your 792code in places you want to debug.</p> 793 794<p>Getting this to work requires a small amount of configuration. On Unix 795systems with X11, install the <a href="http://www.graphviz.org">graphviz</a> 796toolkit, and make sure 'dot' and 'gv' are in your path. If you are running on 797Mac OS/X, download and install the Mac OS/X <a 798href="http://www.pixelglow.com/graphviz/">Graphviz program</a>, and add 799<tt>/Applications/Graphviz.app/Contents/MacOS/</tt> (or wherever you install 800it) to your path. Once in your system and path are set up, rerun the LLVM 801configure script and rebuild LLVM to enable this functionality.</p> 802 803<p><tt>SelectionDAG</tt> has been extended to make it easier to locate 804<i>interesting</i> nodes in large complex graphs. From gdb, if you 805<tt>call DAG.setGraphColor(<i>node</i>, "<i>color</i>")</tt>, then the 806next <tt>call DAG.viewGraph()</tt> would highlight the node in the 807specified color (choices of colors can be found at <a 808href="http://www.graphviz.org/doc/info/colors.html">colors</a>.) More 809complex node attributes can be provided with <tt>call 810DAG.setGraphAttrs(<i>node</i>, "<i>attributes</i>")</tt> (choices can be 811found at <a href="http://www.graphviz.org/doc/info/attrs.html">Graph 812Attributes</a>.) If you want to restart and clear all the current graph 813attributes, then you can <tt>call DAG.clearGraphAttrs()</tt>. </p> 814 815<p>Note that graph visualization features are compiled out of Release builds 816to reduce file size. This means that you need a Debug+Asserts or 817Release+Asserts build to use these features.</p> 818 819</div> 820 821</div> 822 823<!-- *********************************************************************** --> 824<h2> 825 <a name="datastructure">Picking the Right Data Structure for a Task</a> 826</h2> 827<!-- *********************************************************************** --> 828 829<div> 830 831<p>LLVM has a plethora of data structures in the <tt>llvm/ADT/</tt> directory, 832 and we commonly use STL data structures. This section describes the trade-offs 833 you should consider when you pick one.</p> 834 835<p> 836The first step is a choose your own adventure: do you want a sequential 837container, a set-like container, or a map-like container? The most important 838thing when choosing a container is the algorithmic properties of how you plan to 839access the container. Based on that, you should use:</p> 840 841<ul> 842<li>a <a href="#ds_map">map-like</a> container if you need efficient look-up 843 of an value based on another value. Map-like containers also support 844 efficient queries for containment (whether a key is in the map). Map-like 845 containers generally do not support efficient reverse mapping (values to 846 keys). If you need that, use two maps. Some map-like containers also 847 support efficient iteration through the keys in sorted order. Map-like 848 containers are the most expensive sort, only use them if you need one of 849 these capabilities.</li> 850 851<li>a <a href="#ds_set">set-like</a> container if you need to put a bunch of 852 stuff into a container that automatically eliminates duplicates. Some 853 set-like containers support efficient iteration through the elements in 854 sorted order. Set-like containers are more expensive than sequential 855 containers. 856</li> 857 858<li>a <a href="#ds_sequential">sequential</a> container provides 859 the most efficient way to add elements and keeps track of the order they are 860 added to the collection. They permit duplicates and support efficient 861 iteration, but do not support efficient look-up based on a key. 862</li> 863 864<li>a <a href="#ds_string">string</a> container is a specialized sequential 865 container or reference structure that is used for character or byte 866 arrays.</li> 867 868<li>a <a href="#ds_bit">bit</a> container provides an efficient way to store and 869 perform set operations on sets of numeric id's, while automatically 870 eliminating duplicates. Bit containers require a maximum of 1 bit for each 871 identifier you want to store. 872</li> 873</ul> 874 875<p> 876Once the proper category of container is determined, you can fine tune the 877memory use, constant factors, and cache behaviors of access by intelligently 878picking a member of the category. Note that constant factors and cache behavior 879can be a big deal. If you have a vector that usually only contains a few 880elements (but could contain many), for example, it's much better to use 881<a href="#dss_smallvector">SmallVector</a> than <a href="#dss_vector">vector</a> 882. Doing so avoids (relatively) expensive malloc/free calls, which dwarf the 883cost of adding the elements to the container. </p> 884 885<!-- ======================================================================= --> 886<h3> 887 <a name="ds_sequential">Sequential Containers (std::vector, std::list, etc)</a> 888</h3> 889 890<div> 891There are a variety of sequential containers available for you, based on your 892needs. Pick the first in this section that will do what you want. 893 894<!-- _______________________________________________________________________ --> 895<h4> 896 <a name="dss_arrayref">llvm/ADT/ArrayRef.h</a> 897</h4> 898 899<div> 900<p>The llvm::ArrayRef class is the preferred class to use in an interface that 901 accepts a sequential list of elements in memory and just reads from them. By 902 taking an ArrayRef, the API can be passed a fixed size array, an std::vector, 903 an llvm::SmallVector and anything else that is contiguous in memory. 904</p> 905</div> 906 907 908 909<!-- _______________________________________________________________________ --> 910<h4> 911 <a name="dss_fixedarrays">Fixed Size Arrays</a> 912</h4> 913 914<div> 915<p>Fixed size arrays are very simple and very fast. They are good if you know 916exactly how many elements you have, or you have a (low) upper bound on how many 917you have.</p> 918</div> 919 920<!-- _______________________________________________________________________ --> 921<h4> 922 <a name="dss_heaparrays">Heap Allocated Arrays</a> 923</h4> 924 925<div> 926<p>Heap allocated arrays (new[] + delete[]) are also simple. They are good if 927the number of elements is variable, if you know how many elements you will need 928before the array is allocated, and if the array is usually large (if not, 929consider a <a href="#dss_smallvector">SmallVector</a>). The cost of a heap 930allocated array is the cost of the new/delete (aka malloc/free). Also note that 931if you are allocating an array of a type with a constructor, the constructor and 932destructors will be run for every element in the array (re-sizable vectors only 933construct those elements actually used).</p> 934</div> 935 936<!-- _______________________________________________________________________ --> 937<h4> 938 <a name="dss_tinyptrvector">"llvm/ADT/TinyPtrVector.h"</a> 939</h4> 940 941 942<div> 943<p><tt>TinyPtrVector<Type></tt> is a highly specialized collection class 944that is optimized to avoid allocation in the case when a vector has zero or one 945elements. It has two major restrictions: 1) it can only hold values of pointer 946type, and 2) it cannot hold a null pointer.</p> 947 948<p>Since this container is highly specialized, it is rarely used.</p> 949 950</div> 951 952<!-- _______________________________________________________________________ --> 953<h4> 954 <a name="dss_smallvector">"llvm/ADT/SmallVector.h"</a> 955</h4> 956 957<div> 958<p><tt>SmallVector<Type, N></tt> is a simple class that looks and smells 959just like <tt>vector<Type></tt>: 960it supports efficient iteration, lays out elements in memory order (so you can 961do pointer arithmetic between elements), supports efficient push_back/pop_back 962operations, supports efficient random access to its elements, etc.</p> 963 964<p>The advantage of SmallVector is that it allocates space for 965some number of elements (N) <b>in the object itself</b>. Because of this, if 966the SmallVector is dynamically smaller than N, no malloc is performed. This can 967be a big win in cases where the malloc/free call is far more expensive than the 968code that fiddles around with the elements.</p> 969 970<p>This is good for vectors that are "usually small" (e.g. the number of 971predecessors/successors of a block is usually less than 8). On the other hand, 972this makes the size of the SmallVector itself large, so you don't want to 973allocate lots of them (doing so will waste a lot of space). As such, 974SmallVectors are most useful when on the stack.</p> 975 976<p>SmallVector also provides a nice portable and efficient replacement for 977<tt>alloca</tt>.</p> 978 979</div> 980 981<!-- _______________________________________________________________________ --> 982<h4> 983 <a name="dss_vector"><vector></a> 984</h4> 985 986<div> 987<p> 988std::vector is well loved and respected. It is useful when SmallVector isn't: 989when the size of the vector is often large (thus the small optimization will 990rarely be a benefit) or if you will be allocating many instances of the vector 991itself (which would waste space for elements that aren't in the container). 992vector is also useful when interfacing with code that expects vectors :). 993</p> 994 995<p>One worthwhile note about std::vector: avoid code like this:</p> 996 997<div class="doc_code"> 998<pre> 999for ( ... ) { 1000 std::vector<foo> V; 1001 // make use of V. 1002} 1003</pre> 1004</div> 1005 1006<p>Instead, write this as:</p> 1007 1008<div class="doc_code"> 1009<pre> 1010std::vector<foo> V; 1011for ( ... ) { 1012 // make use of V. 1013 V.clear(); 1014} 1015</pre> 1016</div> 1017 1018<p>Doing so will save (at least) one heap allocation and free per iteration of 1019the loop.</p> 1020 1021</div> 1022 1023<!-- _______________________________________________________________________ --> 1024<h4> 1025 <a name="dss_deque"><deque></a> 1026</h4> 1027 1028<div> 1029<p>std::deque is, in some senses, a generalized version of std::vector. Like 1030std::vector, it provides constant time random access and other similar 1031properties, but it also provides efficient access to the front of the list. It 1032does not guarantee continuity of elements within memory.</p> 1033 1034<p>In exchange for this extra flexibility, std::deque has significantly higher 1035constant factor costs than std::vector. If possible, use std::vector or 1036something cheaper.</p> 1037</div> 1038 1039<!-- _______________________________________________________________________ --> 1040<h4> 1041 <a name="dss_list"><list></a> 1042</h4> 1043 1044<div> 1045<p>std::list is an extremely inefficient class that is rarely useful. 1046It performs a heap allocation for every element inserted into it, thus having an 1047extremely high constant factor, particularly for small data types. std::list 1048also only supports bidirectional iteration, not random access iteration.</p> 1049 1050<p>In exchange for this high cost, std::list supports efficient access to both 1051ends of the list (like std::deque, but unlike std::vector or SmallVector). In 1052addition, the iterator invalidation characteristics of std::list are stronger 1053than that of a vector class: inserting or removing an element into the list does 1054not invalidate iterator or pointers to other elements in the list.</p> 1055</div> 1056 1057<!-- _______________________________________________________________________ --> 1058<h4> 1059 <a name="dss_ilist">llvm/ADT/ilist.h</a> 1060</h4> 1061 1062<div> 1063<p><tt>ilist<T></tt> implements an 'intrusive' doubly-linked list. It is 1064intrusive, because it requires the element to store and provide access to the 1065prev/next pointers for the list.</p> 1066 1067<p><tt>ilist</tt> has the same drawbacks as <tt>std::list</tt>, and additionally 1068requires an <tt>ilist_traits</tt> implementation for the element type, but it 1069provides some novel characteristics. In particular, it can efficiently store 1070polymorphic objects, the traits class is informed when an element is inserted or 1071removed from the list, and <tt>ilist</tt>s are guaranteed to support a 1072constant-time splice operation.</p> 1073 1074<p>These properties are exactly what we want for things like 1075<tt>Instruction</tt>s and basic blocks, which is why these are implemented with 1076<tt>ilist</tt>s.</p> 1077 1078Related classes of interest are explained in the following subsections: 1079 <ul> 1080 <li><a href="#dss_ilist_traits">ilist_traits</a></li> 1081 <li><a href="#dss_iplist">iplist</a></li> 1082 <li><a href="#dss_ilist_node">llvm/ADT/ilist_node.h</a></li> 1083 <li><a href="#dss_ilist_sentinel">Sentinels</a></li> 1084 </ul> 1085</div> 1086 1087<!-- _______________________________________________________________________ --> 1088<h4> 1089 <a name="dss_packedvector">llvm/ADT/PackedVector.h</a> 1090</h4> 1091 1092<div> 1093<p> 1094Useful for storing a vector of values using only a few number of bits for each 1095value. Apart from the standard operations of a vector-like container, it can 1096also perform an 'or' set operation. 1097</p> 1098 1099<p>For example:</p> 1100 1101<div class="doc_code"> 1102<pre> 1103enum State { 1104 None = 0x0, 1105 FirstCondition = 0x1, 1106 SecondCondition = 0x2, 1107 Both = 0x3 1108}; 1109 1110State get() { 1111 PackedVector<State, 2> Vec1; 1112 Vec1.push_back(FirstCondition); 1113 1114 PackedVector<State, 2> Vec2; 1115 Vec2.push_back(SecondCondition); 1116 1117 Vec1 |= Vec2; 1118 return Vec1[0]; // returns 'Both'. 1119} 1120</pre> 1121</div> 1122 1123</div> 1124 1125<!-- _______________________________________________________________________ --> 1126<h4> 1127 <a name="dss_ilist_traits">ilist_traits</a> 1128</h4> 1129 1130<div> 1131<p><tt>ilist_traits<T></tt> is <tt>ilist<T></tt>'s customization 1132mechanism. <tt>iplist<T></tt> (and consequently <tt>ilist<T></tt>) 1133publicly derive from this traits class.</p> 1134</div> 1135 1136<!-- _______________________________________________________________________ --> 1137<h4> 1138 <a name="dss_iplist">iplist</a> 1139</h4> 1140 1141<div> 1142<p><tt>iplist<T></tt> is <tt>ilist<T></tt>'s base and as such 1143supports a slightly narrower interface. Notably, inserters from 1144<tt>T&</tt> are absent.</p> 1145 1146<p><tt>ilist_traits<T></tt> is a public base of this class and can be 1147used for a wide variety of customizations.</p> 1148</div> 1149 1150<!-- _______________________________________________________________________ --> 1151<h4> 1152 <a name="dss_ilist_node">llvm/ADT/ilist_node.h</a> 1153</h4> 1154 1155<div> 1156<p><tt>ilist_node<T></tt> implements a the forward and backward links 1157that are expected by the <tt>ilist<T></tt> (and analogous containers) 1158in the default manner.</p> 1159 1160<p><tt>ilist_node<T></tt>s are meant to be embedded in the node type 1161<tt>T</tt>, usually <tt>T</tt> publicly derives from 1162<tt>ilist_node<T></tt>.</p> 1163</div> 1164 1165<!-- _______________________________________________________________________ --> 1166<h4> 1167 <a name="dss_ilist_sentinel">Sentinels</a> 1168</h4> 1169 1170<div> 1171<p><tt>ilist</tt>s have another specialty that must be considered. To be a good 1172citizen in the C++ ecosystem, it needs to support the standard container 1173operations, such as <tt>begin</tt> and <tt>end</tt> iterators, etc. Also, the 1174<tt>operator--</tt> must work correctly on the <tt>end</tt> iterator in the 1175case of non-empty <tt>ilist</tt>s.</p> 1176 1177<p>The only sensible solution to this problem is to allocate a so-called 1178<i>sentinel</i> along with the intrusive list, which serves as the <tt>end</tt> 1179iterator, providing the back-link to the last element. However conforming to the 1180C++ convention it is illegal to <tt>operator++</tt> beyond the sentinel and it 1181also must not be dereferenced.</p> 1182 1183<p>These constraints allow for some implementation freedom to the <tt>ilist</tt> 1184how to allocate and store the sentinel. The corresponding policy is dictated 1185by <tt>ilist_traits<T></tt>. By default a <tt>T</tt> gets heap-allocated 1186whenever the need for a sentinel arises.</p> 1187 1188<p>While the default policy is sufficient in most cases, it may break down when 1189<tt>T</tt> does not provide a default constructor. Also, in the case of many 1190instances of <tt>ilist</tt>s, the memory overhead of the associated sentinels 1191is wasted. To alleviate the situation with numerous and voluminous 1192<tt>T</tt>-sentinels, sometimes a trick is employed, leading to <i>ghostly 1193sentinels</i>.</p> 1194 1195<p>Ghostly sentinels are obtained by specially-crafted <tt>ilist_traits<T></tt> 1196which superpose the sentinel with the <tt>ilist</tt> instance in memory. Pointer 1197arithmetic is used to obtain the sentinel, which is relative to the 1198<tt>ilist</tt>'s <tt>this</tt> pointer. The <tt>ilist</tt> is augmented by an 1199extra pointer, which serves as the back-link of the sentinel. This is the only 1200field in the ghostly sentinel which can be legally accessed.</p> 1201</div> 1202 1203<!-- _______________________________________________________________________ --> 1204<h4> 1205 <a name="dss_other">Other Sequential Container options</a> 1206</h4> 1207 1208<div> 1209<p>Other STL containers are available, such as std::string.</p> 1210 1211<p>There are also various STL adapter classes such as std::queue, 1212std::priority_queue, std::stack, etc. These provide simplified access to an 1213underlying container but don't affect the cost of the container itself.</p> 1214 1215</div> 1216</div> 1217 1218<!-- ======================================================================= --> 1219<h3> 1220 <a name="ds_string">String-like containers</a> 1221</h3> 1222 1223<div> 1224 1225<p> 1226There are a variety of ways to pass around and use strings in C and C++, and 1227LLVM adds a few new options to choose from. Pick the first option on this list 1228that will do what you need, they are ordered according to their relative cost. 1229</p> 1230<p> 1231Note that is is generally preferred to <em>not</em> pass strings around as 1232"<tt>const char*</tt>"'s. These have a number of problems, including the fact 1233that they cannot represent embedded nul ("\0") characters, and do not have a 1234length available efficiently. The general replacement for '<tt>const 1235char*</tt>' is StringRef. 1236</p> 1237 1238<p>For more information on choosing string containers for APIs, please see 1239<a href="#string_apis">Passing strings</a>.</p> 1240 1241 1242<!-- _______________________________________________________________________ --> 1243<h4> 1244 <a name="dss_stringref">llvm/ADT/StringRef.h</a> 1245</h4> 1246 1247<div> 1248<p> 1249The StringRef class is a simple value class that contains a pointer to a 1250character and a length, and is quite related to the <a 1251href="#dss_arrayref">ArrayRef</a> class (but specialized for arrays of 1252characters). Because StringRef carries a length with it, it safely handles 1253strings with embedded nul characters in it, getting the length does not require 1254a strlen call, and it even has very convenient APIs for slicing and dicing the 1255character range that it represents. 1256</p> 1257 1258<p> 1259StringRef is ideal for passing simple strings around that are known to be live, 1260either because they are C string literals, std::string, a C array, or a 1261SmallVector. Each of these cases has an efficient implicit conversion to 1262StringRef, which doesn't result in a dynamic strlen being executed. 1263</p> 1264 1265<p>StringRef has a few major limitations which make more powerful string 1266containers useful:</p> 1267 1268<ol> 1269<li>You cannot directly convert a StringRef to a 'const char*' because there is 1270no way to add a trailing nul (unlike the .c_str() method on various stronger 1271classes).</li> 1272 1273 1274<li>StringRef doesn't own or keep alive the underlying string bytes. 1275As such it can easily lead to dangling pointers, and is not suitable for 1276embedding in datastructures in most cases (instead, use an std::string or 1277something like that).</li> 1278 1279<li>For the same reason, StringRef cannot be used as the return value of a 1280method if the method "computes" the result string. Instead, use 1281std::string.</li> 1282 1283<li>StringRef's do not allow you to mutate the pointed-to string bytes and it 1284doesn't allow you to insert or remove bytes from the range. For editing 1285operations like this, it interoperates with the <a 1286href="#dss_twine">Twine</a> class.</li> 1287</ol> 1288 1289<p>Because of its strengths and limitations, it is very common for a function to 1290take a StringRef and for a method on an object to return a StringRef that 1291points into some string that it owns.</p> 1292 1293</div> 1294 1295<!-- _______________________________________________________________________ --> 1296<h4> 1297 <a name="dss_twine">llvm/ADT/Twine.h</a> 1298</h4> 1299 1300<div> 1301 <p> 1302 The Twine class is used as an intermediary datatype for APIs that want to take 1303 a string that can be constructed inline with a series of concatenations. 1304 Twine works by forming recursive instances of the Twine datatype (a simple 1305 value object) on the stack as temporary objects, linking them together into a 1306 tree which is then linearized when the Twine is consumed. Twine is only safe 1307 to use as the argument to a function, and should always be a const reference, 1308 e.g.: 1309 </p> 1310 1311 <pre> 1312 void foo(const Twine &T); 1313 ... 1314 StringRef X = ... 1315 unsigned i = ... 1316 foo(X + "." + Twine(i)); 1317 </pre> 1318 1319 <p>This example forms a string like "blarg.42" by concatenating the values 1320 together, and does not form intermediate strings containing "blarg" or 1321 "blarg.". 1322 </p> 1323 1324 <p>Because Twine is constructed with temporary objects on the stack, and 1325 because these instances are destroyed at the end of the current statement, 1326 it is an inherently dangerous API. For example, this simple variant contains 1327 undefined behavior and will probably crash:</p> 1328 1329 <pre> 1330 void foo(const Twine &T); 1331 ... 1332 StringRef X = ... 1333 unsigned i = ... 1334 const Twine &Tmp = X + "." + Twine(i); 1335 foo(Tmp); 1336 </pre> 1337 1338 <p>... because the temporaries are destroyed before the call. That said, 1339 Twine's are much more efficient than intermediate std::string temporaries, and 1340 they work really well with StringRef. Just be aware of their limitations.</p> 1341 1342</div> 1343 1344 1345<!-- _______________________________________________________________________ --> 1346<h4> 1347 <a name="dss_smallstring">llvm/ADT/SmallString.h</a> 1348</h4> 1349 1350<div> 1351 1352<p>SmallString is a subclass of <a href="#dss_smallvector">SmallVector</a> that 1353adds some convenience APIs like += that takes StringRef's. SmallString avoids 1354allocating memory in the case when the preallocated space is enough to hold its 1355data, and it calls back to general heap allocation when required. Since it owns 1356its data, it is very safe to use and supports full mutation of the string.</p> 1357 1358<p>Like SmallVector's, the big downside to SmallString is their sizeof. While 1359they are optimized for small strings, they themselves are not particularly 1360small. This means that they work great for temporary scratch buffers on the 1361stack, but should not generally be put into the heap: it is very rare to 1362see a SmallString as the member of a frequently-allocated heap data structure 1363or returned by-value. 1364</p> 1365 1366</div> 1367 1368<!-- _______________________________________________________________________ --> 1369<h4> 1370 <a name="dss_stdstring">std::string</a> 1371</h4> 1372 1373<div> 1374 1375 <p>The standard C++ std::string class is a very general class that (like 1376 SmallString) owns its underlying data. sizeof(std::string) is very reasonable 1377 so it can be embedded into heap data structures and returned by-value. 1378 On the other hand, std::string is highly inefficient for inline editing (e.g. 1379 concatenating a bunch of stuff together) and because it is provided by the 1380 standard library, its performance characteristics depend a lot of the host 1381 standard library (e.g. libc++ and MSVC provide a highly optimized string 1382 class, GCC contains a really slow implementation). 1383 </p> 1384 1385 <p>The major disadvantage of std::string is that almost every operation that 1386 makes them larger can allocate memory, which is slow. As such, it is better 1387 to use SmallVector or Twine as a scratch buffer, but then use std::string to 1388 persist the result.</p> 1389 1390 1391</div> 1392 1393<!-- end of strings --> 1394</div> 1395 1396 1397<!-- ======================================================================= --> 1398<h3> 1399 <a name="ds_set">Set-Like Containers (std::set, SmallSet, SetVector, etc)</a> 1400</h3> 1401 1402<div> 1403 1404<p>Set-like containers are useful when you need to canonicalize multiple values 1405into a single representation. There are several different choices for how to do 1406this, providing various trade-offs.</p> 1407 1408<!-- _______________________________________________________________________ --> 1409<h4> 1410 <a name="dss_sortedvectorset">A sorted 'vector'</a> 1411</h4> 1412 1413<div> 1414 1415<p>If you intend to insert a lot of elements, then do a lot of queries, a 1416great approach is to use a vector (or other sequential container) with 1417std::sort+std::unique to remove duplicates. This approach works really well if 1418your usage pattern has these two distinct phases (insert then query), and can be 1419coupled with a good choice of <a href="#ds_sequential">sequential container</a>. 1420</p> 1421 1422<p> 1423This combination provides the several nice properties: the result data is 1424contiguous in memory (good for cache locality), has few allocations, is easy to 1425address (iterators in the final vector are just indices or pointers), and can be 1426efficiently queried with a standard binary or radix search.</p> 1427 1428</div> 1429 1430<!-- _______________________________________________________________________ --> 1431<h4> 1432 <a name="dss_smallset">"llvm/ADT/SmallSet.h"</a> 1433</h4> 1434 1435<div> 1436 1437<p>If you have a set-like data structure that is usually small and whose elements 1438are reasonably small, a <tt>SmallSet<Type, N></tt> is a good choice. This set 1439has space for N elements in place (thus, if the set is dynamically smaller than 1440N, no malloc traffic is required) and accesses them with a simple linear search. 1441When the set grows beyond 'N' elements, it allocates a more expensive representation that 1442guarantees efficient access (for most types, it falls back to std::set, but for 1443pointers it uses something far better, <a 1444href="#dss_smallptrset">SmallPtrSet</a>).</p> 1445 1446<p>The magic of this class is that it handles small sets extremely efficiently, 1447but gracefully handles extremely large sets without loss of efficiency. The 1448drawback is that the interface is quite small: it supports insertion, queries 1449and erasing, but does not support iteration.</p> 1450 1451</div> 1452 1453<!-- _______________________________________________________________________ --> 1454<h4> 1455 <a name="dss_smallptrset">"llvm/ADT/SmallPtrSet.h"</a> 1456</h4> 1457 1458<div> 1459 1460<p>SmallPtrSet has all the advantages of <tt>SmallSet</tt> (and a <tt>SmallSet</tt> of pointers is 1461transparently implemented with a <tt>SmallPtrSet</tt>), but also supports iterators. If 1462more than 'N' insertions are performed, a single quadratically 1463probed hash table is allocated and grows as needed, providing extremely 1464efficient access (constant time insertion/deleting/queries with low constant 1465factors) and is very stingy with malloc traffic.</p> 1466 1467<p>Note that, unlike <tt>std::set</tt>, the iterators of <tt>SmallPtrSet</tt> are invalidated 1468whenever an insertion occurs. Also, the values visited by the iterators are not 1469visited in sorted order.</p> 1470 1471</div> 1472 1473<!-- _______________________________________________________________________ --> 1474<h4> 1475 <a name="dss_denseset">"llvm/ADT/DenseSet.h"</a> 1476</h4> 1477 1478<div> 1479 1480<p> 1481DenseSet is a simple quadratically probed hash table. It excels at supporting 1482small values: it uses a single allocation to hold all of the pairs that 1483are currently inserted in the set. DenseSet is a great way to unique small 1484values that are not simple pointers (use <a 1485href="#dss_smallptrset">SmallPtrSet</a> for pointers). Note that DenseSet has 1486the same requirements for the value type that <a 1487href="#dss_densemap">DenseMap</a> has. 1488</p> 1489 1490</div> 1491 1492<!-- _______________________________________________________________________ --> 1493<h4> 1494 <a name="dss_sparseset">"llvm/ADT/SparseSet.h"</a> 1495</h4> 1496 1497<div> 1498 1499<p>SparseSet holds a small number of objects identified by unsigned keys of 1500moderate size. It uses a lot of memory, but provides operations that are 1501almost as fast as a vector. Typical keys are physical registers, virtual 1502registers, or numbered basic blocks.</p> 1503 1504<p>SparseSet is useful for algorithms that need very fast clear/find/insert/erase 1505and fast iteration over small sets. It is not intended for building composite 1506data structures.</p> 1507 1508</div> 1509 1510<!-- _______________________________________________________________________ --> 1511<h4> 1512 <a name="dss_FoldingSet">"llvm/ADT/FoldingSet.h"</a> 1513</h4> 1514 1515<div> 1516 1517<p> 1518FoldingSet is an aggregate class that is really good at uniquing 1519expensive-to-create or polymorphic objects. It is a combination of a chained 1520hash table with intrusive links (uniqued objects are required to inherit from 1521FoldingSetNode) that uses <a href="#dss_smallvector">SmallVector</a> as part of 1522its ID process.</p> 1523 1524<p>Consider a case where you want to implement a "getOrCreateFoo" method for 1525a complex object (for example, a node in the code generator). The client has a 1526description of *what* it wants to generate (it knows the opcode and all the 1527operands), but we don't want to 'new' a node, then try inserting it into a set 1528only to find out it already exists, at which point we would have to delete it 1529and return the node that already exists. 1530</p> 1531 1532<p>To support this style of client, FoldingSet perform a query with a 1533FoldingSetNodeID (which wraps SmallVector) that can be used to describe the 1534element that we want to query for. The query either returns the element 1535matching the ID or it returns an opaque ID that indicates where insertion should 1536take place. Construction of the ID usually does not require heap traffic.</p> 1537 1538<p>Because FoldingSet uses intrusive links, it can support polymorphic objects 1539in the set (for example, you can have SDNode instances mixed with LoadSDNodes). 1540Because the elements are individually allocated, pointers to the elements are 1541stable: inserting or removing elements does not invalidate any pointers to other 1542elements. 1543</p> 1544 1545</div> 1546 1547<!-- _______________________________________________________________________ --> 1548<h4> 1549 <a name="dss_set"><set></a> 1550</h4> 1551 1552<div> 1553 1554<p><tt>std::set</tt> is a reasonable all-around set class, which is decent at 1555many things but great at nothing. std::set allocates memory for each element 1556inserted (thus it is very malloc intensive) and typically stores three pointers 1557per element in the set (thus adding a large amount of per-element space 1558overhead). It offers guaranteed log(n) performance, which is not particularly 1559fast from a complexity standpoint (particularly if the elements of the set are 1560expensive to compare, like strings), and has extremely high constant factors for 1561lookup, insertion and removal.</p> 1562 1563<p>The advantages of std::set are that its iterators are stable (deleting or 1564inserting an element from the set does not affect iterators or pointers to other 1565elements) and that iteration over the set is guaranteed to be in sorted order. 1566If the elements in the set are large, then the relative overhead of the pointers 1567and malloc traffic is not a big deal, but if the elements of the set are small, 1568std::set is almost never a good choice.</p> 1569 1570</div> 1571 1572<!-- _______________________________________________________________________ --> 1573<h4> 1574 <a name="dss_setvector">"llvm/ADT/SetVector.h"</a> 1575</h4> 1576 1577<div> 1578<p>LLVM's SetVector<Type> is an adapter class that combines your choice of 1579a set-like container along with a <a href="#ds_sequential">Sequential 1580Container</a>. The important property 1581that this provides is efficient insertion with uniquing (duplicate elements are 1582ignored) with iteration support. It implements this by inserting elements into 1583both a set-like container and the sequential container, using the set-like 1584container for uniquing and the sequential container for iteration. 1585</p> 1586 1587<p>The difference between SetVector and other sets is that the order of 1588iteration is guaranteed to match the order of insertion into the SetVector. 1589This property is really important for things like sets of pointers. Because 1590pointer values are non-deterministic (e.g. vary across runs of the program on 1591different machines), iterating over the pointers in the set will 1592not be in a well-defined order.</p> 1593 1594<p> 1595The drawback of SetVector is that it requires twice as much space as a normal 1596set and has the sum of constant factors from the set-like container and the 1597sequential container that it uses. Use it *only* if you need to iterate over 1598the elements in a deterministic order. SetVector is also expensive to delete 1599elements out of (linear time), unless you use it's "pop_back" method, which is 1600faster. 1601</p> 1602 1603<p><tt>SetVector</tt> is an adapter class that defaults to 1604 using <tt>std::vector</tt> and a size 16 <tt>SmallSet</tt> for the underlying 1605 containers, so it is quite expensive. However, 1606 <tt>"llvm/ADT/SetVector.h"</tt> also provides a <tt>SmallSetVector</tt> 1607 class, which defaults to using a <tt>SmallVector</tt> and <tt>SmallSet</tt> 1608 of a specified size. If you use this, and if your sets are dynamically 1609 smaller than <tt>N</tt>, you will save a lot of heap traffic.</p> 1610 1611</div> 1612 1613<!-- _______________________________________________________________________ --> 1614<h4> 1615 <a name="dss_uniquevector">"llvm/ADT/UniqueVector.h"</a> 1616</h4> 1617 1618<div> 1619 1620<p> 1621UniqueVector is similar to <a href="#dss_setvector">SetVector</a>, but it 1622retains a unique ID for each element inserted into the set. It internally 1623contains a map and a vector, and it assigns a unique ID for each value inserted 1624into the set.</p> 1625 1626<p>UniqueVector is very expensive: its cost is the sum of the cost of 1627maintaining both the map and vector, it has high complexity, high constant 1628factors, and produces a lot of malloc traffic. It should be avoided.</p> 1629 1630</div> 1631 1632<!-- _______________________________________________________________________ --> 1633<h4> 1634 <a name="dss_immutableset">"llvm/ADT/ImmutableSet.h"</a> 1635</h4> 1636 1637<div> 1638 1639<p> 1640ImmutableSet is an immutable (functional) set implementation based on an AVL 1641tree. 1642Adding or removing elements is done through a Factory object and results in the 1643creation of a new ImmutableSet object. 1644If an ImmutableSet already exists with the given contents, then the existing one 1645is returned; equality is compared with a FoldingSetNodeID. 1646The time and space complexity of add or remove operations is logarithmic in the 1647size of the original set. 1648 1649<p> 1650There is no method for returning an element of the set, you can only check for 1651membership. 1652 1653</div> 1654 1655 1656<!-- _______________________________________________________________________ --> 1657<h4> 1658 <a name="dss_otherset">Other Set-Like Container Options</a> 1659</h4> 1660 1661<div> 1662 1663<p> 1664The STL provides several other options, such as std::multiset and the various 1665"hash_set" like containers (whether from C++ TR1 or from the SGI library). We 1666never use hash_set and unordered_set because they are generally very expensive 1667(each insertion requires a malloc) and very non-portable. 1668</p> 1669 1670<p>std::multiset is useful if you're not interested in elimination of 1671duplicates, but has all the drawbacks of std::set. A sorted vector (where you 1672don't delete duplicate entries) or some other approach is almost always 1673better.</p> 1674 1675</div> 1676 1677</div> 1678 1679<!-- ======================================================================= --> 1680<h3> 1681 <a name="ds_map">Map-Like Containers (std::map, DenseMap, etc)</a> 1682</h3> 1683 1684<div> 1685Map-like containers are useful when you want to associate data to a key. As 1686usual, there are a lot of different ways to do this. :) 1687 1688<!-- _______________________________________________________________________ --> 1689<h4> 1690 <a name="dss_sortedvectormap">A sorted 'vector'</a> 1691</h4> 1692 1693<div> 1694 1695<p> 1696If your usage pattern follows a strict insert-then-query approach, you can 1697trivially use the same approach as <a href="#dss_sortedvectorset">sorted vectors 1698for set-like containers</a>. The only difference is that your query function 1699(which uses std::lower_bound to get efficient log(n) lookup) should only compare 1700the key, not both the key and value. This yields the same advantages as sorted 1701vectors for sets. 1702</p> 1703</div> 1704 1705<!-- _______________________________________________________________________ --> 1706<h4> 1707 <a name="dss_stringmap">"llvm/ADT/StringMap.h"</a> 1708</h4> 1709 1710<div> 1711 1712<p> 1713Strings are commonly used as keys in maps, and they are difficult to support 1714efficiently: they are variable length, inefficient to hash and compare when 1715long, expensive to copy, etc. StringMap is a specialized container designed to 1716cope with these issues. It supports mapping an arbitrary range of bytes to an 1717arbitrary other object.</p> 1718 1719<p>The StringMap implementation uses a quadratically-probed hash table, where 1720the buckets store a pointer to the heap allocated entries (and some other 1721stuff). The entries in the map must be heap allocated because the strings are 1722variable length. The string data (key) and the element object (value) are 1723stored in the same allocation with the string data immediately after the element 1724object. This container guarantees the "<tt>(char*)(&Value+1)</tt>" points 1725to the key string for a value.</p> 1726 1727<p>The StringMap is very fast for several reasons: quadratic probing is very 1728cache efficient for lookups, the hash value of strings in buckets is not 1729recomputed when looking up an element, StringMap rarely has to touch the 1730memory for unrelated objects when looking up a value (even when hash collisions 1731happen), hash table growth does not recompute the hash values for strings 1732already in the table, and each pair in the map is store in a single allocation 1733(the string data is stored in the same allocation as the Value of a pair).</p> 1734 1735<p>StringMap also provides query methods that take byte ranges, so it only ever 1736copies a string if a value is inserted into the table.</p> 1737 1738<p>StringMap iteratation order, however, is not guaranteed to be deterministic, 1739so any uses which require that should instead use a std::map.</p> 1740</div> 1741 1742<!-- _______________________________________________________________________ --> 1743<h4> 1744 <a name="dss_indexedmap">"llvm/ADT/IndexedMap.h"</a> 1745</h4> 1746 1747<div> 1748<p> 1749IndexedMap is a specialized container for mapping small dense integers (or 1750values that can be mapped to small dense integers) to some other type. It is 1751internally implemented as a vector with a mapping function that maps the keys to 1752the dense integer range. 1753</p> 1754 1755<p> 1756This is useful for cases like virtual registers in the LLVM code generator: they 1757have a dense mapping that is offset by a compile-time constant (the first 1758virtual register ID).</p> 1759 1760</div> 1761 1762<!-- _______________________________________________________________________ --> 1763<h4> 1764 <a name="dss_densemap">"llvm/ADT/DenseMap.h"</a> 1765</h4> 1766 1767<div> 1768 1769<p> 1770DenseMap is a simple quadratically probed hash table. It excels at supporting 1771small keys and values: it uses a single allocation to hold all of the pairs that 1772are currently inserted in the map. DenseMap is a great way to map pointers to 1773pointers, or map other small types to each other. 1774</p> 1775 1776<p> 1777There are several aspects of DenseMap that you should be aware of, however. The 1778iterators in a DenseMap are invalidated whenever an insertion occurs, unlike 1779map. Also, because DenseMap allocates space for a large number of key/value 1780pairs (it starts with 64 by default), it will waste a lot of space if your keys 1781or values are large. Finally, you must implement a partial specialization of 1782DenseMapInfo for the key that you want, if it isn't already supported. This 1783is required to tell DenseMap about two special marker values (which can never be 1784inserted into the map) that it needs internally.</p> 1785 1786<p> 1787DenseMap's find_as() method supports lookup operations using an alternate key 1788type. This is useful in cases where the normal key type is expensive to 1789construct, but cheap to compare against. The DenseMapInfo is responsible for 1790defining the appropriate comparison and hashing methods for each alternate 1791key type used. 1792</p> 1793 1794</div> 1795 1796<!-- _______________________________________________________________________ --> 1797<h4> 1798 <a name="dss_valuemap">"llvm/ADT/ValueMap.h"</a> 1799</h4> 1800 1801<div> 1802 1803<p> 1804ValueMap is a wrapper around a <a href="#dss_densemap">DenseMap</a> mapping 1805Value*s (or subclasses) to another type. When a Value is deleted or RAUW'ed, 1806ValueMap will update itself so the new version of the key is mapped to the same 1807value, just as if the key were a WeakVH. You can configure exactly how this 1808happens, and what else happens on these two events, by passing 1809a <code>Config</code> parameter to the ValueMap template.</p> 1810 1811</div> 1812 1813<!-- _______________________________________________________________________ --> 1814<h4> 1815 <a name="dss_intervalmap">"llvm/ADT/IntervalMap.h"</a> 1816</h4> 1817 1818<div> 1819 1820<p> IntervalMap is a compact map for small keys and values. It maps key 1821intervals instead of single keys, and it will automatically coalesce adjacent 1822intervals. When then map only contains a few intervals, they are stored in the 1823map object itself to avoid allocations.</p> 1824 1825<p> The IntervalMap iterators are quite big, so they should not be passed around 1826as STL iterators. The heavyweight iterators allow a smaller data structure.</p> 1827 1828</div> 1829 1830<!-- _______________________________________________________________________ --> 1831<h4> 1832 <a name="dss_map"><map></a> 1833</h4> 1834 1835<div> 1836 1837<p> 1838std::map has similar characteristics to <a href="#dss_set">std::set</a>: it uses 1839a single allocation per pair inserted into the map, it offers log(n) lookup with 1840an extremely large constant factor, imposes a space penalty of 3 pointers per 1841pair in the map, etc.</p> 1842 1843<p>std::map is most useful when your keys or values are very large, if you need 1844to iterate over the collection in sorted order, or if you need stable iterators 1845into the map (i.e. they don't get invalidated if an insertion or deletion of 1846another element takes place).</p> 1847 1848</div> 1849 1850<!-- _______________________________________________________________________ --> 1851<h4> 1852 <a name="dss_inteqclasses">"llvm/ADT/IntEqClasses.h"</a> 1853</h4> 1854 1855<div> 1856 1857<p>IntEqClasses provides a compact representation of equivalence classes of 1858small integers. Initially, each integer in the range 0..n-1 has its own 1859equivalence class. Classes can be joined by passing two class representatives to 1860the join(a, b) method. Two integers are in the same class when findLeader() 1861returns the same representative.</p> 1862 1863<p>Once all equivalence classes are formed, the map can be compressed so each 1864integer 0..n-1 maps to an equivalence class number in the range 0..m-1, where m 1865is the total number of equivalence classes. The map must be uncompressed before 1866it can be edited again.</p> 1867 1868</div> 1869 1870<!-- _______________________________________________________________________ --> 1871<h4> 1872 <a name="dss_immutablemap">"llvm/ADT/ImmutableMap.h"</a> 1873</h4> 1874 1875<div> 1876 1877<p> 1878ImmutableMap is an immutable (functional) map implementation based on an AVL 1879tree. 1880Adding or removing elements is done through a Factory object and results in the 1881creation of a new ImmutableMap object. 1882If an ImmutableMap already exists with the given key set, then the existing one 1883is returned; equality is compared with a FoldingSetNodeID. 1884The time and space complexity of add or remove operations is logarithmic in the 1885size of the original map. 1886 1887</div> 1888 1889<!-- _______________________________________________________________________ --> 1890<h4> 1891 <a name="dss_othermap">Other Map-Like Container Options</a> 1892</h4> 1893 1894<div> 1895 1896<p> 1897The STL provides several other options, such as std::multimap and the various 1898"hash_map" like containers (whether from C++ TR1 or from the SGI library). We 1899never use hash_set and unordered_set because they are generally very expensive 1900(each insertion requires a malloc) and very non-portable.</p> 1901 1902<p>std::multimap is useful if you want to map a key to multiple values, but has 1903all the drawbacks of std::map. A sorted vector or some other approach is almost 1904always better.</p> 1905 1906</div> 1907 1908</div> 1909 1910<!-- ======================================================================= --> 1911<h3> 1912 <a name="ds_bit">Bit storage containers (BitVector, SparseBitVector)</a> 1913</h3> 1914 1915<div> 1916<p>Unlike the other containers, there are only two bit storage containers, and 1917choosing when to use each is relatively straightforward.</p> 1918 1919<p>One additional option is 1920<tt>std::vector<bool></tt>: we discourage its use for two reasons 1) the 1921implementation in many common compilers (e.g. commonly available versions of 1922GCC) is extremely inefficient and 2) the C++ standards committee is likely to 1923deprecate this container and/or change it significantly somehow. In any case, 1924please don't use it.</p> 1925 1926<!-- _______________________________________________________________________ --> 1927<h4> 1928 <a name="dss_bitvector">BitVector</a> 1929</h4> 1930 1931<div> 1932<p> The BitVector container provides a dynamic size set of bits for manipulation. 1933It supports individual bit setting/testing, as well as set operations. The set 1934operations take time O(size of bitvector), but operations are performed one word 1935at a time, instead of one bit at a time. This makes the BitVector very fast for 1936set operations compared to other containers. Use the BitVector when you expect 1937the number of set bits to be high (IE a dense set). 1938</p> 1939</div> 1940 1941<!-- _______________________________________________________________________ --> 1942<h4> 1943 <a name="dss_smallbitvector">SmallBitVector</a> 1944</h4> 1945 1946<div> 1947<p> The SmallBitVector container provides the same interface as BitVector, but 1948it is optimized for the case where only a small number of bits, less than 194925 or so, are needed. It also transparently supports larger bit counts, but 1950slightly less efficiently than a plain BitVector, so SmallBitVector should 1951only be used when larger counts are rare. 1952</p> 1953 1954<p> 1955At this time, SmallBitVector does not support set operations (and, or, xor), 1956and its operator[] does not provide an assignable lvalue. 1957</p> 1958</div> 1959 1960<!-- _______________________________________________________________________ --> 1961<h4> 1962 <a name="dss_sparsebitvector">SparseBitVector</a> 1963</h4> 1964 1965<div> 1966<p> The SparseBitVector container is much like BitVector, with one major 1967difference: Only the bits that are set, are stored. This makes the 1968SparseBitVector much more space efficient than BitVector when the set is sparse, 1969as well as making set operations O(number of set bits) instead of O(size of 1970universe). 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 1971(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). 1972</p> 1973</div> 1974 1975</div> 1976 1977</div> 1978 1979<!-- *********************************************************************** --> 1980<h2> 1981 <a name="common">Helpful Hints for Common Operations</a> 1982</h2> 1983<!-- *********************************************************************** --> 1984 1985<div> 1986 1987<p>This section describes how to perform some very simple transformations of 1988LLVM code. This is meant to give examples of common idioms used, showing the 1989practical side of LLVM transformations. <p> Because this is a "how-to" section, 1990you should also read about the main classes that you will be working with. The 1991<a href="#coreclasses">Core LLVM Class Hierarchy Reference</a> contains details 1992and descriptions of the main classes that you should know about.</p> 1993 1994<!-- NOTE: this section should be heavy on example code --> 1995<!-- ======================================================================= --> 1996<h3> 1997 <a name="inspection">Basic Inspection and Traversal Routines</a> 1998</h3> 1999 2000<div> 2001 2002<p>The LLVM compiler infrastructure have many different data structures that may 2003be traversed. Following the example of the C++ standard template library, the 2004techniques used to traverse these various data structures are all basically the 2005same. For a enumerable sequence of values, the <tt>XXXbegin()</tt> function (or 2006method) returns an iterator to the start of the sequence, the <tt>XXXend()</tt> 2007function returns an iterator pointing to one past the last valid element of the 2008sequence, and there is some <tt>XXXiterator</tt> data type that is common 2009between the two operations.</p> 2010 2011<p>Because the pattern for iteration is common across many different aspects of 2012the program representation, the standard template library algorithms may be used 2013on them, and it is easier to remember how to iterate. First we show a few common 2014examples of the data structures that need to be traversed. Other data 2015structures are traversed in very similar ways.</p> 2016 2017<!-- _______________________________________________________________________ --> 2018<h4> 2019 <a name="iterate_function">Iterating over the </a><a 2020 href="#BasicBlock"><tt>BasicBlock</tt></a>s in a <a 2021 href="#Function"><tt>Function</tt></a> 2022</h4> 2023 2024<div> 2025 2026<p>It's quite common to have a <tt>Function</tt> instance that you'd like to 2027transform in some way; in particular, you'd like to manipulate its 2028<tt>BasicBlock</tt>s. To facilitate this, you'll need to iterate over all of 2029the <tt>BasicBlock</tt>s that constitute the <tt>Function</tt>. The following is 2030an example that prints the name of a <tt>BasicBlock</tt> and the number of 2031<tt>Instruction</tt>s it contains:</p> 2032 2033<div class="doc_code"> 2034<pre> 2035// <i>func is a pointer to a Function instance</i> 2036for (Function::iterator i = func->begin(), e = func->end(); i != e; ++i) 2037 // <i>Print out the name of the basic block if it has one, and then the</i> 2038 // <i>number of instructions that it contains</i> 2039 errs() << "Basic block (name=" << i->getName() << ") has " 2040 << i->size() << " instructions.\n"; 2041</pre> 2042</div> 2043 2044<p>Note that i can be used as if it were a pointer for the purposes of 2045invoking member functions of the <tt>Instruction</tt> class. This is 2046because the indirection operator is overloaded for the iterator 2047classes. In the above code, the expression <tt>i->size()</tt> is 2048exactly equivalent to <tt>(*i).size()</tt> just like you'd expect.</p> 2049 2050</div> 2051 2052<!-- _______________________________________________________________________ --> 2053<h4> 2054 <a name="iterate_basicblock">Iterating over the </a><a 2055 href="#Instruction"><tt>Instruction</tt></a>s in a <a 2056 href="#BasicBlock"><tt>BasicBlock</tt></a> 2057</h4> 2058 2059<div> 2060 2061<p>Just like when dealing with <tt>BasicBlock</tt>s in <tt>Function</tt>s, it's 2062easy to iterate over the individual instructions that make up 2063<tt>BasicBlock</tt>s. Here's a code snippet that prints out each instruction in 2064a <tt>BasicBlock</tt>:</p> 2065 2066<div class="doc_code"> 2067<pre> 2068// <i>blk is a pointer to a BasicBlock instance</i> 2069for (BasicBlock::iterator i = blk->begin(), e = blk->end(); i != e; ++i) 2070 // <i>The next statement works since operator<<(ostream&,...)</i> 2071 // <i>is overloaded for Instruction&</i> 2072 errs() << *i << "\n"; 2073</pre> 2074</div> 2075 2076<p>However, this isn't really the best way to print out the contents of a 2077<tt>BasicBlock</tt>! Since the ostream operators are overloaded for virtually 2078anything you'll care about, you could have just invoked the print routine on the 2079basic block itself: <tt>errs() << *blk << "\n";</tt>.</p> 2080 2081</div> 2082 2083<!-- _______________________________________________________________________ --> 2084<h4> 2085 <a name="iterate_institer">Iterating over the </a><a 2086 href="#Instruction"><tt>Instruction</tt></a>s in a <a 2087 href="#Function"><tt>Function</tt></a> 2088</h4> 2089 2090<div> 2091 2092<p>If you're finding that you commonly iterate over a <tt>Function</tt>'s 2093<tt>BasicBlock</tt>s and then that <tt>BasicBlock</tt>'s <tt>Instruction</tt>s, 2094<tt>InstIterator</tt> should be used instead. You'll need to include <a 2095href="/doxygen/InstIterator_8h-source.html"><tt>llvm/Support/InstIterator.h</tt></a>, 2096and then instantiate <tt>InstIterator</tt>s explicitly in your code. Here's a 2097small example that shows how to dump all instructions in a function to the standard error stream:<p> 2098 2099<div class="doc_code"> 2100<pre> 2101#include "<a href="/doxygen/InstIterator_8h-source.html">llvm/Support/InstIterator.h</a>" 2102 2103// <i>F is a pointer to a Function instance</i> 2104for (inst_iterator I = inst_begin(F), E = inst_end(F); I != E; ++I) 2105 errs() << *I << "\n"; 2106</pre> 2107</div> 2108 2109<p>Easy, isn't it? You can also use <tt>InstIterator</tt>s to fill a 2110work list with its initial contents. For example, if you wanted to 2111initialize a work list to contain all instructions in a <tt>Function</tt> 2112F, all you would need to do is something like:</p> 2113 2114<div class="doc_code"> 2115<pre> 2116std::set<Instruction*> worklist; 2117// or better yet, SmallPtrSet<Instruction*, 64> worklist; 2118 2119for (inst_iterator I = inst_begin(F), E = inst_end(F); I != E; ++I) 2120 worklist.insert(&*I); 2121</pre> 2122</div> 2123 2124<p>The STL set <tt>worklist</tt> would now contain all instructions in the 2125<tt>Function</tt> pointed to by F.</p> 2126 2127</div> 2128 2129<!-- _______________________________________________________________________ --> 2130<h4> 2131 <a name="iterate_convert">Turning an iterator into a class pointer (and 2132 vice-versa)</a> 2133</h4> 2134 2135<div> 2136 2137<p>Sometimes, it'll be useful to grab a reference (or pointer) to a class 2138instance when all you've got at hand is an iterator. Well, extracting 2139a reference or a pointer from an iterator is very straight-forward. 2140Assuming that <tt>i</tt> is a <tt>BasicBlock::iterator</tt> and <tt>j</tt> 2141is a <tt>BasicBlock::const_iterator</tt>:</p> 2142 2143<div class="doc_code"> 2144<pre> 2145Instruction& inst = *i; // <i>Grab reference to instruction reference</i> 2146Instruction* pinst = &*i; // <i>Grab pointer to instruction reference</i> 2147const Instruction& inst = *j; 2148</pre> 2149</div> 2150 2151<p>However, the iterators you'll be working with in the LLVM framework are 2152special: they will automatically convert to a ptr-to-instance type whenever they 2153need to. Instead of dereferencing the iterator and then taking the address of 2154the result, you can simply assign the iterator to the proper pointer type and 2155you get the dereference and address-of operation as a result of the assignment 2156(behind the scenes, this is a result of overloading casting mechanisms). Thus 2157the last line of the last example,</p> 2158 2159<div class="doc_code"> 2160<pre> 2161Instruction *pinst = &*i; 2162</pre> 2163</div> 2164 2165<p>is semantically equivalent to</p> 2166 2167<div class="doc_code"> 2168<pre> 2169Instruction *pinst = i; 2170</pre> 2171</div> 2172 2173<p>It's also possible to turn a class pointer into the corresponding iterator, 2174and this is a constant time operation (very efficient). The following code 2175snippet illustrates use of the conversion constructors provided by LLVM 2176iterators. By using these, you can explicitly grab the iterator of something 2177without actually obtaining it via iteration over some structure:</p> 2178 2179<div class="doc_code"> 2180<pre> 2181void printNextInstruction(Instruction* inst) { 2182 BasicBlock::iterator it(inst); 2183 ++it; // <i>After this line, it refers to the instruction after *inst</i> 2184 if (it != inst->getParent()->end()) errs() << *it << "\n"; 2185} 2186</pre> 2187</div> 2188 2189<p>Unfortunately, these implicit conversions come at a cost; they prevent 2190these iterators from conforming to standard iterator conventions, and thus 2191from being usable with standard algorithms and containers. For example, they 2192prevent the following code, where <tt>B</tt> is a <tt>BasicBlock</tt>, 2193from compiling:</p> 2194 2195<div class="doc_code"> 2196<pre> 2197 llvm::SmallVector<llvm::Instruction *, 16>(B->begin(), B->end()); 2198</pre> 2199</div> 2200 2201<p>Because of this, these implicit conversions may be removed some day, 2202and <tt>operator*</tt> changed to return a pointer instead of a reference.</p> 2203 2204</div> 2205 2206<!--_______________________________________________________________________--> 2207<h4> 2208 <a name="iterate_complex">Finding call sites: a slightly more complex 2209 example</a> 2210</h4> 2211 2212<div> 2213 2214<p>Say that you're writing a FunctionPass and would like to count all the 2215locations in the entire module (that is, across every <tt>Function</tt>) where a 2216certain function (i.e., some <tt>Function</tt>*) is already in scope. As you'll 2217learn later, you may want to use an <tt>InstVisitor</tt> to accomplish this in a 2218much more straight-forward manner, but this example will allow us to explore how 2219you'd do it if you didn't have <tt>InstVisitor</tt> around. In pseudo-code, this 2220is what we want to do:</p> 2221 2222<div class="doc_code"> 2223<pre> 2224initialize callCounter to zero 2225for each Function f in the Module 2226 for each BasicBlock b in f 2227 for each Instruction i in b 2228 if (i is a CallInst and calls the given function) 2229 increment callCounter 2230</pre> 2231</div> 2232 2233<p>And the actual code is (remember, because we're writing a 2234<tt>FunctionPass</tt>, our <tt>FunctionPass</tt>-derived class simply has to 2235override the <tt>runOnFunction</tt> method):</p> 2236 2237<div class="doc_code"> 2238<pre> 2239Function* targetFunc = ...; 2240 2241class OurFunctionPass : public FunctionPass { 2242 public: 2243 OurFunctionPass(): callCounter(0) { } 2244 2245 virtual runOnFunction(Function& F) { 2246 for (Function::iterator b = F.begin(), be = F.end(); b != be; ++b) { 2247 for (BasicBlock::iterator i = b->begin(), ie = b->end(); i != ie; ++i) { 2248 if (<a href="#CallInst">CallInst</a>* callInst = <a href="#isa">dyn_cast</a><<a 2249 href="#CallInst">CallInst</a>>(&*i)) { 2250 // <i>We know we've encountered a call instruction, so we</i> 2251 // <i>need to determine if it's a call to the</i> 2252 // <i>function pointed to by m_func or not.</i> 2253 if (callInst->getCalledFunction() == targetFunc) 2254 ++callCounter; 2255 } 2256 } 2257 } 2258 } 2259 2260 private: 2261 unsigned callCounter; 2262}; 2263</pre> 2264</div> 2265 2266</div> 2267 2268<!--_______________________________________________________________________--> 2269<h4> 2270 <a name="calls_and_invokes">Treating calls and invokes the same way</a> 2271</h4> 2272 2273<div> 2274 2275<p>You may have noticed that the previous example was a bit oversimplified in 2276that it did not deal with call sites generated by 'invoke' instructions. In 2277this, and in other situations, you may find that you want to treat 2278<tt>CallInst</tt>s and <tt>InvokeInst</tt>s the same way, even though their 2279most-specific common base class is <tt>Instruction</tt>, which includes lots of 2280less closely-related things. For these cases, LLVM provides a handy wrapper 2281class called <a 2282href="http://llvm.org/doxygen/classllvm_1_1CallSite.html"><tt>CallSite</tt></a>. 2283It is essentially a wrapper around an <tt>Instruction</tt> pointer, with some 2284methods that provide functionality common to <tt>CallInst</tt>s and 2285<tt>InvokeInst</tt>s.</p> 2286 2287<p>This class has "value semantics": it should be passed by value, not by 2288reference and it should not be dynamically allocated or deallocated using 2289<tt>operator new</tt> or <tt>operator delete</tt>. It is efficiently copyable, 2290assignable and constructable, with costs equivalents to that of a bare pointer. 2291If you look at its definition, it has only a single pointer member.</p> 2292 2293</div> 2294 2295<!--_______________________________________________________________________--> 2296<h4> 2297 <a name="iterate_chains">Iterating over def-use & use-def chains</a> 2298</h4> 2299 2300<div> 2301 2302<p>Frequently, we might have an instance of the <a 2303href="/doxygen/classllvm_1_1Value.html">Value Class</a> and we want to 2304determine which <tt>User</tt>s use the <tt>Value</tt>. The list of all 2305<tt>User</tt>s of a particular <tt>Value</tt> is called a <i>def-use</i> chain. 2306For example, let's say we have a <tt>Function*</tt> named <tt>F</tt> to a 2307particular function <tt>foo</tt>. Finding all of the instructions that 2308<i>use</i> <tt>foo</tt> is as simple as iterating over the <i>def-use</i> chain 2309of <tt>F</tt>:</p> 2310 2311<div class="doc_code"> 2312<pre> 2313Function *F = ...; 2314 2315for (Value::use_iterator i = F->use_begin(), e = F->use_end(); i != e; ++i) 2316 if (Instruction *Inst = dyn_cast<Instruction>(*i)) { 2317 errs() << "F is used in instruction:\n"; 2318 errs() << *Inst << "\n"; 2319 } 2320</pre> 2321</div> 2322 2323<p>Note that dereferencing a <tt>Value::use_iterator</tt> is not a very cheap 2324operation. Instead of performing <tt>*i</tt> above several times, consider 2325doing it only once in the loop body and reusing its result.</p> 2326 2327<p>Alternatively, it's common to have an instance of the <a 2328href="/doxygen/classllvm_1_1User.html">User Class</a> and need to know what 2329<tt>Value</tt>s are used by it. The list of all <tt>Value</tt>s used by a 2330<tt>User</tt> is known as a <i>use-def</i> chain. Instances of class 2331<tt>Instruction</tt> are common <tt>User</tt>s, so we might want to iterate over 2332all of the values that a particular instruction uses (that is, the operands of 2333the particular <tt>Instruction</tt>):</p> 2334 2335<div class="doc_code"> 2336<pre> 2337Instruction *pi = ...; 2338 2339for (User::op_iterator i = pi->op_begin(), e = pi->op_end(); i != e; ++i) { 2340 Value *v = *i; 2341 // <i>...</i> 2342} 2343</pre> 2344</div> 2345 2346<p>Declaring objects as <tt>const</tt> is an important tool of enforcing 2347mutation free algorithms (such as analyses, etc.). For this purpose above 2348iterators come in constant flavors as <tt>Value::const_use_iterator</tt> 2349and <tt>Value::const_op_iterator</tt>. They automatically arise when 2350calling <tt>use/op_begin()</tt> on <tt>const Value*</tt>s or 2351<tt>const User*</tt>s respectively. Upon dereferencing, they return 2352<tt>const Use*</tt>s. Otherwise the above patterns remain unchanged.</p> 2353 2354</div> 2355 2356<!--_______________________________________________________________________--> 2357<h4> 2358 <a name="iterate_preds">Iterating over predecessors & 2359successors of blocks</a> 2360</h4> 2361 2362<div> 2363 2364<p>Iterating over the predecessors and successors of a block is quite easy 2365with the routines defined in <tt>"llvm/Support/CFG.h"</tt>. Just use code like 2366this to iterate over all predecessors of BB:</p> 2367 2368<div class="doc_code"> 2369<pre> 2370#include "llvm/Support/CFG.h" 2371BasicBlock *BB = ...; 2372 2373for (pred_iterator PI = pred_begin(BB), E = pred_end(BB); PI != E; ++PI) { 2374 BasicBlock *Pred = *PI; 2375 // <i>...</i> 2376} 2377</pre> 2378</div> 2379 2380<p>Similarly, to iterate over successors use 2381succ_iterator/succ_begin/succ_end.</p> 2382 2383</div> 2384 2385</div> 2386 2387<!-- ======================================================================= --> 2388<h3> 2389 <a name="simplechanges">Making simple changes</a> 2390</h3> 2391 2392<div> 2393 2394<p>There are some primitive transformation operations present in the LLVM 2395infrastructure that are worth knowing about. When performing 2396transformations, it's fairly common to manipulate the contents of basic 2397blocks. This section describes some of the common methods for doing so 2398and gives example code.</p> 2399 2400<!--_______________________________________________________________________--> 2401<h4> 2402 <a name="schanges_creating">Creating and inserting new 2403 <tt>Instruction</tt>s</a> 2404</h4> 2405 2406<div> 2407 2408<p><i>Instantiating Instructions</i></p> 2409 2410<p>Creation of <tt>Instruction</tt>s is straight-forward: simply call the 2411constructor for the kind of instruction to instantiate and provide the necessary 2412parameters. For example, an <tt>AllocaInst</tt> only <i>requires</i> a 2413(const-ptr-to) <tt>Type</tt>. Thus:</p> 2414 2415<div class="doc_code"> 2416<pre> 2417AllocaInst* ai = new AllocaInst(Type::Int32Ty); 2418</pre> 2419</div> 2420 2421<p>will create an <tt>AllocaInst</tt> instance that represents the allocation of 2422one integer in the current stack frame, at run time. Each <tt>Instruction</tt> 2423subclass is likely to have varying default parameters which change the semantics 2424of the instruction, so refer to the <a 2425href="/doxygen/classllvm_1_1Instruction.html">doxygen documentation for the subclass of 2426Instruction</a> that you're interested in instantiating.</p> 2427 2428<p><i>Naming values</i></p> 2429 2430<p>It is very useful to name the values of instructions when you're able to, as 2431this facilitates the debugging of your transformations. If you end up looking 2432at generated LLVM machine code, you definitely want to have logical names 2433associated with the results of instructions! By supplying a value for the 2434<tt>Name</tt> (default) parameter of the <tt>Instruction</tt> constructor, you 2435associate a logical name with the result of the instruction's execution at 2436run time. For example, say that I'm writing a transformation that dynamically 2437allocates space for an integer on the stack, and that integer is going to be 2438used as some kind of index by some other code. To accomplish this, I place an 2439<tt>AllocaInst</tt> at the first point in the first <tt>BasicBlock</tt> of some 2440<tt>Function</tt>, and I'm intending to use it within the same 2441<tt>Function</tt>. I might do:</p> 2442 2443<div class="doc_code"> 2444<pre> 2445AllocaInst* pa = new AllocaInst(Type::Int32Ty, 0, "indexLoc"); 2446</pre> 2447</div> 2448 2449<p>where <tt>indexLoc</tt> is now the logical name of the instruction's 2450execution value, which is a pointer to an integer on the run time stack.</p> 2451 2452<p><i>Inserting instructions</i></p> 2453 2454<p>There are essentially two ways to insert an <tt>Instruction</tt> 2455into an existing sequence of instructions that form a <tt>BasicBlock</tt>:</p> 2456 2457<ul> 2458 <li>Insertion into an explicit instruction list 2459 2460 <p>Given a <tt>BasicBlock* pb</tt>, an <tt>Instruction* pi</tt> within that 2461 <tt>BasicBlock</tt>, and a newly-created instruction we wish to insert 2462 before <tt>*pi</tt>, we do the following: </p> 2463 2464<div class="doc_code"> 2465<pre> 2466BasicBlock *pb = ...; 2467Instruction *pi = ...; 2468Instruction *newInst = new Instruction(...); 2469 2470pb->getInstList().insert(pi, newInst); // <i>Inserts newInst before pi in pb</i> 2471</pre> 2472</div> 2473 2474 <p>Appending to the end of a <tt>BasicBlock</tt> is so common that 2475 the <tt>Instruction</tt> class and <tt>Instruction</tt>-derived 2476 classes provide constructors which take a pointer to a 2477 <tt>BasicBlock</tt> to be appended to. For example code that 2478 looked like: </p> 2479 2480<div class="doc_code"> 2481<pre> 2482BasicBlock *pb = ...; 2483Instruction *newInst = new Instruction(...); 2484 2485pb->getInstList().push_back(newInst); // <i>Appends newInst to pb</i> 2486</pre> 2487</div> 2488 2489 <p>becomes: </p> 2490 2491<div class="doc_code"> 2492<pre> 2493BasicBlock *pb = ...; 2494Instruction *newInst = new Instruction(..., pb); 2495</pre> 2496</div> 2497 2498 <p>which is much cleaner, especially if you are creating 2499 long instruction streams.</p></li> 2500 2501 <li>Insertion into an implicit instruction list 2502 2503 <p><tt>Instruction</tt> instances that are already in <tt>BasicBlock</tt>s 2504 are implicitly associated with an existing instruction list: the instruction 2505 list of the enclosing basic block. Thus, we could have accomplished the same 2506 thing as the above code without being given a <tt>BasicBlock</tt> by doing: 2507 </p> 2508 2509<div class="doc_code"> 2510<pre> 2511Instruction *pi = ...; 2512Instruction *newInst = new Instruction(...); 2513 2514pi->getParent()->getInstList().insert(pi, newInst); 2515</pre> 2516</div> 2517 2518 <p>In fact, this sequence of steps occurs so frequently that the 2519 <tt>Instruction</tt> class and <tt>Instruction</tt>-derived classes provide 2520 constructors which take (as a default parameter) a pointer to an 2521 <tt>Instruction</tt> which the newly-created <tt>Instruction</tt> should 2522 precede. That is, <tt>Instruction</tt> constructors are capable of 2523 inserting the newly-created instance into the <tt>BasicBlock</tt> of a 2524 provided instruction, immediately before that instruction. Using an 2525 <tt>Instruction</tt> constructor with a <tt>insertBefore</tt> (default) 2526 parameter, the above code becomes:</p> 2527 2528<div class="doc_code"> 2529<pre> 2530Instruction* pi = ...; 2531Instruction* newInst = new Instruction(..., pi); 2532</pre> 2533</div> 2534 2535 <p>which is much cleaner, especially if you're creating a lot of 2536 instructions and adding them to <tt>BasicBlock</tt>s.</p></li> 2537</ul> 2538 2539</div> 2540 2541<!--_______________________________________________________________________--> 2542<h4> 2543 <a name="schanges_deleting">Deleting <tt>Instruction</tt>s</a> 2544</h4> 2545 2546<div> 2547 2548<p>Deleting an instruction from an existing sequence of instructions that form a 2549<a href="#BasicBlock"><tt>BasicBlock</tt></a> is very straight-forward: just 2550call the instruction's eraseFromParent() method. For example:</p> 2551 2552<div class="doc_code"> 2553<pre> 2554<a href="#Instruction">Instruction</a> *I = .. ; 2555I->eraseFromParent(); 2556</pre> 2557</div> 2558 2559<p>This unlinks the instruction from its containing basic block and deletes 2560it. If you'd just like to unlink the instruction from its containing basic 2561block but not delete it, you can use the <tt>removeFromParent()</tt> method.</p> 2562 2563</div> 2564 2565<!--_______________________________________________________________________--> 2566<h4> 2567 <a name="schanges_replacing">Replacing an <tt>Instruction</tt> with another 2568 <tt>Value</tt></a> 2569</h4> 2570 2571<div> 2572 2573<h5><i>Replacing individual instructions</i></h5> 2574 2575<p>Including "<a href="/doxygen/BasicBlockUtils_8h-source.html">llvm/Transforms/Utils/BasicBlockUtils.h</a>" 2576permits use of two very useful replace functions: <tt>ReplaceInstWithValue</tt> 2577and <tt>ReplaceInstWithInst</tt>.</p> 2578 2579<h5><a name="schanges_deleting">Deleting <tt>Instruction</tt>s</a></h5> 2580 2581<div> 2582<ul> 2583 <li><tt>ReplaceInstWithValue</tt> 2584 2585 <p>This function replaces all uses of a given instruction with a value, 2586 and then removes the original instruction. The following example 2587 illustrates the replacement of the result of a particular 2588 <tt>AllocaInst</tt> that allocates memory for a single integer with a null 2589 pointer to an integer.</p> 2590 2591<div class="doc_code"> 2592<pre> 2593AllocaInst* instToReplace = ...; 2594BasicBlock::iterator ii(instToReplace); 2595 2596ReplaceInstWithValue(instToReplace->getParent()->getInstList(), ii, 2597 Constant::getNullValue(PointerType::getUnqual(Type::Int32Ty))); 2598</pre></div></li> 2599 2600 <li><tt>ReplaceInstWithInst</tt> 2601 2602 <p>This function replaces a particular instruction with another 2603 instruction, inserting the new instruction into the basic block at the 2604 location where the old instruction was, and replacing any uses of the old 2605 instruction with the new instruction. The following example illustrates 2606 the replacement of one <tt>AllocaInst</tt> with another.</p> 2607 2608<div class="doc_code"> 2609<pre> 2610AllocaInst* instToReplace = ...; 2611BasicBlock::iterator ii(instToReplace); 2612 2613ReplaceInstWithInst(instToReplace->getParent()->getInstList(), ii, 2614 new AllocaInst(Type::Int32Ty, 0, "ptrToReplacedInt")); 2615</pre></div></li> 2616</ul> 2617 2618</div> 2619 2620<h5><i>Replacing multiple uses of <tt>User</tt>s and <tt>Value</tt>s</i></h5> 2621 2622<p>You can use <tt>Value::replaceAllUsesWith</tt> and 2623<tt>User::replaceUsesOfWith</tt> to change more than one use at a time. See the 2624doxygen documentation for the <a href="/doxygen/classllvm_1_1Value.html">Value Class</a> 2625and <a href="/doxygen/classllvm_1_1User.html">User Class</a>, respectively, for more 2626information.</p> 2627 2628<!-- Value::replaceAllUsesWith User::replaceUsesOfWith Point out: 2629include/llvm/Transforms/Utils/ especially BasicBlockUtils.h with: 2630ReplaceInstWithValue, ReplaceInstWithInst --> 2631 2632</div> 2633 2634<!--_______________________________________________________________________--> 2635<h4> 2636 <a name="schanges_deletingGV">Deleting <tt>GlobalVariable</tt>s</a> 2637</h4> 2638 2639<div> 2640 2641<p>Deleting a global variable from a module is just as easy as deleting an 2642Instruction. First, you must have a pointer to the global variable that you wish 2643 to delete. You use this pointer to erase it from its parent, the module. 2644 For example:</p> 2645 2646<div class="doc_code"> 2647<pre> 2648<a href="#GlobalVariable">GlobalVariable</a> *GV = .. ; 2649 2650GV->eraseFromParent(); 2651</pre> 2652</div> 2653 2654</div> 2655 2656</div> 2657 2658<!-- ======================================================================= --> 2659<h3> 2660 <a name="create_types">How to Create Types</a> 2661</h3> 2662 2663<div> 2664 2665<p>In generating IR, you may need some complex types. If you know these types 2666statically, you can use <tt>TypeBuilder<...>::get()</tt>, defined 2667in <tt>llvm/Support/TypeBuilder.h</tt>, to retrieve them. <tt>TypeBuilder</tt> 2668has two forms depending on whether you're building types for cross-compilation 2669or native library use. <tt>TypeBuilder<T, true></tt> requires 2670that <tt>T</tt> be independent of the host environment, meaning that it's built 2671out of types from 2672the <a href="/doxygen/namespacellvm_1_1types.html"><tt>llvm::types</tt></a> 2673namespace and pointers, functions, arrays, etc. built of 2674those. <tt>TypeBuilder<T, false></tt> additionally allows native C types 2675whose size may depend on the host compiler. For example,</p> 2676 2677<div class="doc_code"> 2678<pre> 2679FunctionType *ft = TypeBuilder<types::i<8>(types::i<32>*), true>::get(); 2680</pre> 2681</div> 2682 2683<p>is easier to read and write than the equivalent</p> 2684 2685<div class="doc_code"> 2686<pre> 2687std::vector<const Type*> params; 2688params.push_back(PointerType::getUnqual(Type::Int32Ty)); 2689FunctionType *ft = FunctionType::get(Type::Int8Ty, params, false); 2690</pre> 2691</div> 2692 2693<p>See the <a href="/doxygen/TypeBuilder_8h-source.html#l00001">class 2694comment</a> for more details.</p> 2695 2696</div> 2697 2698</div> 2699 2700<!-- *********************************************************************** --> 2701<h2> 2702 <a name="threading">Threads and LLVM</a> 2703</h2> 2704<!-- *********************************************************************** --> 2705 2706<div> 2707<p> 2708This section describes the interaction of the LLVM APIs with multithreading, 2709both on the part of client applications, and in the JIT, in the hosted 2710application. 2711</p> 2712 2713<p> 2714Note that LLVM's support for multithreading is still relatively young. Up 2715through version 2.5, the execution of threaded hosted applications was 2716supported, but not threaded client access to the APIs. While this use case is 2717now supported, clients <em>must</em> adhere to the guidelines specified below to 2718ensure proper operation in multithreaded mode. 2719</p> 2720 2721<p> 2722Note that, on Unix-like platforms, LLVM requires the presence of GCC's atomic 2723intrinsics in order to support threaded operation. If you need a 2724multhreading-capable LLVM on a platform without a suitably modern system 2725compiler, consider compiling LLVM and LLVM-GCC in single-threaded mode, and 2726using the resultant compiler to build a copy of LLVM with multithreading 2727support. 2728</p> 2729 2730<!-- ======================================================================= --> 2731<h3> 2732 <a name="startmultithreaded">Entering and Exiting Multithreaded Mode</a> 2733</h3> 2734 2735<div> 2736 2737<p> 2738In order to properly protect its internal data structures while avoiding 2739excessive locking overhead in the single-threaded case, the LLVM must intialize 2740certain data structures necessary to provide guards around its internals. To do 2741so, the client program must invoke <tt>llvm_start_multithreaded()</tt> before 2742making any concurrent LLVM API calls. To subsequently tear down these 2743structures, use the <tt>llvm_stop_multithreaded()</tt> call. You can also use 2744the <tt>llvm_is_multithreaded()</tt> call to check the status of multithreaded 2745mode. 2746</p> 2747 2748<p> 2749Note that both of these calls must be made <em>in isolation</em>. That is to 2750say that no other LLVM API calls may be executing at any time during the 2751execution of <tt>llvm_start_multithreaded()</tt> or <tt>llvm_stop_multithreaded 2752</tt>. It's is the client's responsibility to enforce this isolation. 2753</p> 2754 2755<p> 2756The return value of <tt>llvm_start_multithreaded()</tt> indicates the success or 2757failure of the initialization. Failure typically indicates that your copy of 2758LLVM was built without multithreading support, typically because GCC atomic 2759intrinsics were not found in your system compiler. In this case, the LLVM API 2760will not be safe for concurrent calls. However, it <em>will</em> be safe for 2761hosting threaded applications in the JIT, though <a href="#jitthreading">care 2762must be taken</a> to ensure that side exits and the like do not accidentally 2763result in concurrent LLVM API calls. 2764</p> 2765</div> 2766 2767<!-- ======================================================================= --> 2768<h3> 2769 <a name="shutdown">Ending Execution with <tt>llvm_shutdown()</tt></a> 2770</h3> 2771 2772<div> 2773<p> 2774When you are done using the LLVM APIs, you should call <tt>llvm_shutdown()</tt> 2775to deallocate memory used for internal structures. This will also invoke 2776<tt>llvm_stop_multithreaded()</tt> if LLVM is operating in multithreaded mode. 2777As such, <tt>llvm_shutdown()</tt> requires the same isolation guarantees as 2778<tt>llvm_stop_multithreaded()</tt>. 2779</p> 2780 2781<p> 2782Note that, if you use scope-based shutdown, you can use the 2783<tt>llvm_shutdown_obj</tt> class, which calls <tt>llvm_shutdown()</tt> in its 2784destructor. 2785</div> 2786 2787<!-- ======================================================================= --> 2788<h3> 2789 <a name="managedstatic">Lazy Initialization with <tt>ManagedStatic</tt></a> 2790</h3> 2791 2792<div> 2793<p> 2794<tt>ManagedStatic</tt> is a utility class in LLVM used to implement static 2795initialization of static resources, such as the global type tables. Before the 2796invocation of <tt>llvm_shutdown()</tt>, it implements a simple lazy 2797initialization scheme. Once <tt>llvm_start_multithreaded()</tt> returns, 2798however, it uses double-checked locking to implement thread-safe lazy 2799initialization. 2800</p> 2801 2802<p> 2803Note that, because no other threads are allowed to issue LLVM API calls before 2804<tt>llvm_start_multithreaded()</tt> returns, it is possible to have 2805<tt>ManagedStatic</tt>s of <tt>llvm::sys::Mutex</tt>s. 2806</p> 2807 2808<p> 2809The <tt>llvm_acquire_global_lock()</tt> and <tt>llvm_release_global_lock</tt> 2810APIs provide access to the global lock used to implement the double-checked 2811locking for lazy initialization. These should only be used internally to LLVM, 2812and only if you know what you're doing! 2813</p> 2814</div> 2815 2816<!-- ======================================================================= --> 2817<h3> 2818 <a name="llvmcontext">Achieving Isolation with <tt>LLVMContext</tt></a> 2819</h3> 2820 2821<div> 2822<p> 2823<tt>LLVMContext</tt> is an opaque class in the LLVM API which clients can use 2824to operate multiple, isolated instances of LLVM concurrently within the same 2825address space. For instance, in a hypothetical compile-server, the compilation 2826of an individual translation unit is conceptually independent from all the 2827others, and it would be desirable to be able to compile incoming translation 2828units concurrently on independent server threads. Fortunately, 2829<tt>LLVMContext</tt> exists to enable just this kind of scenario! 2830</p> 2831 2832<p> 2833Conceptually, <tt>LLVMContext</tt> provides isolation. Every LLVM entity 2834(<tt>Module</tt>s, <tt>Value</tt>s, <tt>Type</tt>s, <tt>Constant</tt>s, etc.) 2835in LLVM's in-memory IR belongs to an <tt>LLVMContext</tt>. Entities in 2836different contexts <em>cannot</em> interact with each other: <tt>Module</tt>s in 2837different contexts cannot be linked together, <tt>Function</tt>s cannot be added 2838to <tt>Module</tt>s in different contexts, etc. What this means is that is is 2839safe to compile on multiple threads simultaneously, as long as no two threads 2840operate on entities within the same context. 2841</p> 2842 2843<p> 2844In practice, very few places in the API require the explicit specification of a 2845<tt>LLVMContext</tt>, other than the <tt>Type</tt> creation/lookup APIs. 2846Because every <tt>Type</tt> carries a reference to its owning context, most 2847other entities can determine what context they belong to by looking at their 2848own <tt>Type</tt>. If you are adding new entities to LLVM IR, please try to 2849maintain this interface design. 2850</p> 2851 2852<p> 2853For clients that do <em>not</em> require the benefits of isolation, LLVM 2854provides a convenience API <tt>getGlobalContext()</tt>. This returns a global, 2855lazily initialized <tt>LLVMContext</tt> that may be used in situations where 2856isolation is not a concern. 2857</p> 2858</div> 2859 2860<!-- ======================================================================= --> 2861<h3> 2862 <a name="jitthreading">Threads and the JIT</a> 2863</h3> 2864 2865<div> 2866<p> 2867LLVM's "eager" JIT compiler is safe to use in threaded programs. Multiple 2868threads can call <tt>ExecutionEngine::getPointerToFunction()</tt> or 2869<tt>ExecutionEngine::runFunction()</tt> concurrently, and multiple threads can 2870run code output by the JIT concurrently. The user must still ensure that only 2871one thread accesses IR in a given <tt>LLVMContext</tt> while another thread 2872might be modifying it. One way to do that is to always hold the JIT lock while 2873accessing IR outside the JIT (the JIT <em>modifies</em> the IR by adding 2874<tt>CallbackVH</tt>s). Another way is to only 2875call <tt>getPointerToFunction()</tt> from the <tt>LLVMContext</tt>'s thread. 2876</p> 2877 2878<p>When the JIT is configured to compile lazily (using 2879<tt>ExecutionEngine::DisableLazyCompilation(false)</tt>), there is currently a 2880<a href="http://llvm.org/bugs/show_bug.cgi?id=5184">race condition</a> in 2881updating call sites after a function is lazily-jitted. It's still possible to 2882use the lazy JIT in a threaded program if you ensure that only one thread at a 2883time can call any particular lazy stub and that the JIT lock guards any IR 2884access, but we suggest using only the eager JIT in threaded programs. 2885</p> 2886</div> 2887 2888</div> 2889 2890<!-- *********************************************************************** --> 2891<h2> 2892 <a name="advanced">Advanced Topics</a> 2893</h2> 2894<!-- *********************************************************************** --> 2895 2896<div> 2897<p> 2898This section describes some of the advanced or obscure API's that most clients 2899do not need to be aware of. These API's tend manage the inner workings of the 2900LLVM system, and only need to be accessed in unusual circumstances. 2901</p> 2902 2903 2904<!-- ======================================================================= --> 2905<h3> 2906 <a name="SymbolTable">The <tt>ValueSymbolTable</tt> class</a> 2907</h3> 2908 2909<div> 2910<p>The <tt><a href="http://llvm.org/doxygen/classllvm_1_1ValueSymbolTable.html"> 2911ValueSymbolTable</a></tt> class provides a symbol table that the <a 2912href="#Function"><tt>Function</tt></a> and <a href="#Module"> 2913<tt>Module</tt></a> classes use for naming value definitions. The symbol table 2914can provide a name for any <a href="#Value"><tt>Value</tt></a>. 2915</p> 2916 2917<p>Note that the <tt>SymbolTable</tt> class should not be directly accessed 2918by most clients. It should only be used when iteration over the symbol table 2919names themselves are required, which is very special purpose. Note that not 2920all LLVM 2921<tt><a href="#Value">Value</a></tt>s have names, and those without names (i.e. they have 2922an empty name) do not exist in the symbol table. 2923</p> 2924 2925<p>Symbol tables support iteration over the values in the symbol 2926table with <tt>begin/end/iterator</tt> and supports querying to see if a 2927specific name is in the symbol table (with <tt>lookup</tt>). The 2928<tt>ValueSymbolTable</tt> class exposes no public mutator methods, instead, 2929simply call <tt>setName</tt> on a value, which will autoinsert it into the 2930appropriate symbol table.</p> 2931 2932</div> 2933 2934 2935 2936<!-- ======================================================================= --> 2937<h3> 2938 <a name="UserLayout">The <tt>User</tt> and owned <tt>Use</tt> classes' memory layout</a> 2939</h3> 2940 2941<div> 2942<p>The <tt><a href="http://llvm.org/doxygen/classllvm_1_1User.html"> 2943User</a></tt> class provides a basis for expressing the ownership of <tt>User</tt> 2944towards other <tt><a href="http://llvm.org/doxygen/classllvm_1_1Value.html"> 2945Value</a></tt>s. The <tt><a href="http://llvm.org/doxygen/classllvm_1_1Use.html"> 2946Use</a></tt> helper class is employed to do the bookkeeping and to facilitate <i>O(1)</i> 2947addition and removal.</p> 2948 2949<!-- ______________________________________________________________________ --> 2950<h4> 2951 <a name="Use2User"> 2952 Interaction and relationship between <tt>User</tt> and <tt>Use</tt> objects 2953 </a> 2954</h4> 2955 2956<div> 2957<p> 2958A subclass of <tt>User</tt> can choose between incorporating its <tt>Use</tt> objects 2959or refer to them out-of-line by means of a pointer. A mixed variant 2960(some <tt>Use</tt>s inline others hung off) is impractical and breaks the invariant 2961that the <tt>Use</tt> objects belonging to the same <tt>User</tt> form a contiguous array. 2962</p> 2963 2964<p> 2965We have 2 different layouts in the <tt>User</tt> (sub)classes: 2966<ul> 2967<li><p>Layout a) 2968The <tt>Use</tt> object(s) are inside (resp. at fixed offset) of the <tt>User</tt> 2969object and there are a fixed number of them.</p> 2970 2971<li><p>Layout b) 2972The <tt>Use</tt> object(s) are referenced by a pointer to an 2973array from the <tt>User</tt> object and there may be a variable 2974number of them.</p> 2975</ul> 2976<p> 2977As of v2.4 each layout still possesses a direct pointer to the 2978start of the array of <tt>Use</tt>s. Though not mandatory for layout a), 2979we stick to this redundancy for the sake of simplicity. 2980The <tt>User</tt> object also stores the number of <tt>Use</tt> objects it 2981has. (Theoretically this information can also be calculated 2982given the scheme presented below.)</p> 2983<p> 2984Special forms of allocation operators (<tt>operator new</tt>) 2985enforce the following memory layouts:</p> 2986 2987<ul> 2988<li><p>Layout a) is modelled by prepending the <tt>User</tt> object by the <tt>Use[]</tt> array.</p> 2989 2990<pre> 2991...---.---.---.---.-------... 2992 | P | P | P | P | User 2993'''---'---'---'---'-------''' 2994</pre> 2995 2996<li><p>Layout b) is modelled by pointing at the <tt>Use[]</tt> array.</p> 2997<pre> 2998.-------... 2999| User 3000'-------''' 3001 | 3002 v 3003 .---.---.---.---... 3004 | P | P | P | P | 3005 '---'---'---'---''' 3006</pre> 3007</ul> 3008<i>(In the above figures '<tt>P</tt>' stands for the <tt>Use**</tt> that 3009 is stored in each <tt>Use</tt> object in the member <tt>Use::Prev</tt>)</i> 3010 3011</div> 3012 3013<!-- ______________________________________________________________________ --> 3014<h4> 3015 <a name="Waymarking">The waymarking algorithm</a> 3016</h4> 3017 3018<div> 3019<p> 3020Since the <tt>Use</tt> objects are deprived of the direct (back)pointer to 3021their <tt>User</tt> objects, there must be a fast and exact method to 3022recover it. This is accomplished by the following scheme:</p> 3023 3024A bit-encoding in the 2 LSBits (least significant bits) of the <tt>Use::Prev</tt> allows to find the 3025start of the <tt>User</tt> object: 3026<ul> 3027<li><tt>00</tt> —> binary digit 0</li> 3028<li><tt>01</tt> —> binary digit 1</li> 3029<li><tt>10</tt> —> stop and calculate (<tt>s</tt>)</li> 3030<li><tt>11</tt> —> full stop (<tt>S</tt>)</li> 3031</ul> 3032<p> 3033Given a <tt>Use*</tt>, all we have to do is to walk till we get 3034a stop and we either have a <tt>User</tt> immediately behind or 3035we have to walk to the next stop picking up digits 3036and calculating the offset:</p> 3037<pre> 3038.---.---.---.---.---.---.---.---.---.---.---.---.---.---.---.---.---------------- 3039| 1 | s | 1 | 0 | 1 | 0 | s | 1 | 1 | 0 | s | 1 | 1 | s | 1 | S | User (or User*) 3040'---'---'---'---'---'---'---'---'---'---'---'---'---'---'---'---'---------------- 3041 |+15 |+10 |+6 |+3 |+1 3042 | | | | |__> 3043 | | | |__________> 3044 | | |______________________> 3045 | |______________________________________> 3046 |__________________________________________________________> 3047</pre> 3048<p> 3049Only the significant number of bits need to be stored between the 3050stops, so that the <i>worst case is 20 memory accesses</i> when there are 30511000 <tt>Use</tt> objects associated with a <tt>User</tt>.</p> 3052 3053</div> 3054 3055<!-- ______________________________________________________________________ --> 3056<h4> 3057 <a name="ReferenceImpl">Reference implementation</a> 3058</h4> 3059 3060<div> 3061<p> 3062The following literate Haskell fragment demonstrates the concept:</p> 3063 3064<div class="doc_code"> 3065<pre> 3066> import Test.QuickCheck 3067> 3068> digits :: Int -> [Char] -> [Char] 3069> digits 0 acc = '0' : acc 3070> digits 1 acc = '1' : acc 3071> digits n acc = digits (n `div` 2) $ digits (n `mod` 2) acc 3072> 3073> dist :: Int -> [Char] -> [Char] 3074> dist 0 [] = ['S'] 3075> dist 0 acc = acc 3076> dist 1 acc = let r = dist 0 acc in 's' : digits (length r) r 3077> dist n acc = dist (n - 1) $ dist 1 acc 3078> 3079> takeLast n ss = reverse $ take n $ reverse ss 3080> 3081> test = takeLast 40 $ dist 20 [] 3082> 3083</pre> 3084</div> 3085<p> 3086Printing <test> gives: <tt>"1s100000s11010s10100s1111s1010s110s11s1S"</tt></p> 3087<p> 3088The reverse algorithm computes the length of the string just by examining 3089a certain prefix:</p> 3090 3091<div class="doc_code"> 3092<pre> 3093> pref :: [Char] -> Int 3094> pref "S" = 1 3095> pref ('s':'1':rest) = decode 2 1 rest 3096> pref (_:rest) = 1 + pref rest 3097> 3098> decode walk acc ('0':rest) = decode (walk + 1) (acc * 2) rest 3099> decode walk acc ('1':rest) = decode (walk + 1) (acc * 2 + 1) rest 3100> decode walk acc _ = walk + acc 3101> 3102</pre> 3103</div> 3104<p> 3105Now, as expected, printing <pref test> gives <tt>40</tt>.</p> 3106<p> 3107We can <i>quickCheck</i> this with following property:</p> 3108 3109<div class="doc_code"> 3110<pre> 3111> testcase = dist 2000 [] 3112> testcaseLength = length testcase 3113> 3114> identityProp n = n > 0 && n <= testcaseLength ==> length arr == pref arr 3115> where arr = takeLast n testcase 3116> 3117</pre> 3118</div> 3119<p> 3120As expected <quickCheck identityProp> gives:</p> 3121 3122<pre> 3123*Main> quickCheck identityProp 3124OK, passed 100 tests. 3125</pre> 3126<p> 3127Let's be a bit more exhaustive:</p> 3128 3129<div class="doc_code"> 3130<pre> 3131> 3132> deepCheck p = check (defaultConfig { configMaxTest = 500 }) p 3133> 3134</pre> 3135</div> 3136<p> 3137And here is the result of <deepCheck identityProp>:</p> 3138 3139<pre> 3140*Main> deepCheck identityProp 3141OK, passed 500 tests. 3142</pre> 3143 3144</div> 3145 3146<!-- ______________________________________________________________________ --> 3147<h4> 3148 <a name="Tagging">Tagging considerations</a> 3149</h4> 3150 3151<div> 3152 3153<p> 3154To maintain the invariant that the 2 LSBits of each <tt>Use**</tt> in <tt>Use</tt> 3155never change after being set up, setters of <tt>Use::Prev</tt> must re-tag the 3156new <tt>Use**</tt> on every modification. Accordingly getters must strip the 3157tag bits.</p> 3158<p> 3159For layout b) instead of the <tt>User</tt> we find a pointer (<tt>User*</tt> with LSBit set). 3160Following this pointer brings us to the <tt>User</tt>. A portable trick ensures 3161that the first bytes of <tt>User</tt> (if interpreted as a pointer) never has 3162the LSBit set. (Portability is relying on the fact that all known compilers place the 3163<tt>vptr</tt> in the first word of the instances.)</p> 3164 3165</div> 3166 3167</div> 3168 3169</div> 3170 3171<!-- *********************************************************************** --> 3172<h2> 3173 <a name="coreclasses">The Core LLVM Class Hierarchy Reference </a> 3174</h2> 3175<!-- *********************************************************************** --> 3176 3177<div> 3178<p><tt>#include "<a href="/doxygen/Type_8h-source.html">llvm/Type.h</a>"</tt> 3179<br>doxygen info: <a href="/doxygen/classllvm_1_1Type.html">Type Class</a></p> 3180 3181<p>The Core LLVM classes are the primary means of representing the program 3182being inspected or transformed. The core LLVM classes are defined in 3183header files in the <tt>include/llvm/</tt> directory, and implemented in 3184the <tt>lib/VMCore</tt> directory.</p> 3185 3186<!-- ======================================================================= --> 3187<h3> 3188 <a name="Type">The <tt>Type</tt> class and Derived Types</a> 3189</h3> 3190 3191<div> 3192 3193 <p><tt>Type</tt> is a superclass of all type classes. Every <tt>Value</tt> has 3194 a <tt>Type</tt>. <tt>Type</tt> cannot be instantiated directly but only 3195 through its subclasses. Certain primitive types (<tt>VoidType</tt>, 3196 <tt>LabelType</tt>, <tt>FloatType</tt> and <tt>DoubleType</tt>) have hidden 3197 subclasses. They are hidden because they offer no useful functionality beyond 3198 what the <tt>Type</tt> class offers except to distinguish themselves from 3199 other subclasses of <tt>Type</tt>.</p> 3200 <p>All other types are subclasses of <tt>DerivedType</tt>. Types can be 3201 named, but this is not a requirement. There exists exactly 3202 one instance of a given shape at any one time. This allows type equality to 3203 be performed with address equality of the Type Instance. That is, given two 3204 <tt>Type*</tt> values, the types are identical if the pointers are identical. 3205 </p> 3206 3207<!-- _______________________________________________________________________ --> 3208<h4> 3209 <a name="m_Type">Important Public Methods</a> 3210</h4> 3211 3212<div> 3213 3214<ul> 3215 <li><tt>bool isIntegerTy() const</tt>: Returns true for any integer type.</li> 3216 3217 <li><tt>bool isFloatingPointTy()</tt>: Return true if this is one of the five 3218 floating point types.</li> 3219 3220 <li><tt>bool isSized()</tt>: Return true if the type has known size. Things 3221 that don't have a size are abstract types, labels and void.</li> 3222 3223</ul> 3224</div> 3225 3226<!-- _______________________________________________________________________ --> 3227<h4> 3228 <a name="derivedtypes">Important Derived Types</a> 3229</h4> 3230<div> 3231<dl> 3232 <dt><tt>IntegerType</tt></dt> 3233 <dd>Subclass of DerivedType that represents integer types of any bit width. 3234 Any bit width between <tt>IntegerType::MIN_INT_BITS</tt> (1) and 3235 <tt>IntegerType::MAX_INT_BITS</tt> (~8 million) can be represented. 3236 <ul> 3237 <li><tt>static const IntegerType* get(unsigned NumBits)</tt>: get an integer 3238 type of a specific bit width.</li> 3239 <li><tt>unsigned getBitWidth() const</tt>: Get the bit width of an integer 3240 type.</li> 3241 </ul> 3242 </dd> 3243 <dt><tt>SequentialType</tt></dt> 3244 <dd>This is subclassed by ArrayType, PointerType and VectorType. 3245 <ul> 3246 <li><tt>const Type * getElementType() const</tt>: Returns the type of each 3247 of the elements in the sequential type. </li> 3248 </ul> 3249 </dd> 3250 <dt><tt>ArrayType</tt></dt> 3251 <dd>This is a subclass of SequentialType and defines the interface for array 3252 types. 3253 <ul> 3254 <li><tt>unsigned getNumElements() const</tt>: Returns the number of 3255 elements in the array. </li> 3256 </ul> 3257 </dd> 3258 <dt><tt>PointerType</tt></dt> 3259 <dd>Subclass of SequentialType for pointer types.</dd> 3260 <dt><tt>VectorType</tt></dt> 3261 <dd>Subclass of SequentialType for vector types. A 3262 vector type is similar to an ArrayType but is distinguished because it is 3263 a first class type whereas ArrayType is not. Vector types are used for 3264 vector operations and are usually small vectors of of an integer or floating 3265 point type.</dd> 3266 <dt><tt>StructType</tt></dt> 3267 <dd>Subclass of DerivedTypes for struct types.</dd> 3268 <dt><tt><a name="FunctionType">FunctionType</a></tt></dt> 3269 <dd>Subclass of DerivedTypes for function types. 3270 <ul> 3271 <li><tt>bool isVarArg() const</tt>: Returns true if it's a vararg 3272 function</li> 3273 <li><tt> const Type * getReturnType() const</tt>: Returns the 3274 return type of the function.</li> 3275 <li><tt>const Type * getParamType (unsigned i)</tt>: Returns 3276 the type of the ith parameter.</li> 3277 <li><tt> const unsigned getNumParams() const</tt>: Returns the 3278 number of formal parameters.</li> 3279 </ul> 3280 </dd> 3281</dl> 3282</div> 3283 3284</div> 3285 3286<!-- ======================================================================= --> 3287<h3> 3288 <a name="Module">The <tt>Module</tt> class</a> 3289</h3> 3290 3291<div> 3292 3293<p><tt>#include "<a 3294href="/doxygen/Module_8h-source.html">llvm/Module.h</a>"</tt><br> doxygen info: 3295<a href="/doxygen/classllvm_1_1Module.html">Module Class</a></p> 3296 3297<p>The <tt>Module</tt> class represents the top level structure present in LLVM 3298programs. An LLVM module is effectively either a translation unit of the 3299original program or a combination of several translation units merged by the 3300linker. The <tt>Module</tt> class keeps track of a list of <a 3301href="#Function"><tt>Function</tt></a>s, a list of <a 3302href="#GlobalVariable"><tt>GlobalVariable</tt></a>s, and a <a 3303href="#SymbolTable"><tt>SymbolTable</tt></a>. Additionally, it contains a few 3304helpful member functions that try to make common operations easy.</p> 3305 3306<!-- _______________________________________________________________________ --> 3307<h4> 3308 <a name="m_Module">Important Public Members of the <tt>Module</tt> class</a> 3309</h4> 3310 3311<div> 3312 3313<ul> 3314 <li><tt>Module::Module(std::string name = "")</tt> 3315 3316 <p>Constructing a <a href="#Module">Module</a> is easy. You can optionally 3317provide a name for it (probably based on the name of the translation unit).</p> 3318 </li> 3319 3320 <li><tt>Module::iterator</tt> - Typedef for function list iterator<br> 3321 <tt>Module::const_iterator</tt> - Typedef for const_iterator.<br> 3322 3323 <tt>begin()</tt>, <tt>end()</tt> 3324 <tt>size()</tt>, <tt>empty()</tt> 3325 3326 <p>These are forwarding methods that make it easy to access the contents of 3327 a <tt>Module</tt> object's <a href="#Function"><tt>Function</tt></a> 3328 list.</p></li> 3329 3330 <li><tt>Module::FunctionListType &getFunctionList()</tt> 3331 3332 <p> Returns the list of <a href="#Function"><tt>Function</tt></a>s. This is 3333 necessary to use when you need to update the list or perform a complex 3334 action that doesn't have a forwarding method.</p> 3335 3336 <p><!-- Global Variable --></p></li> 3337</ul> 3338 3339<hr> 3340 3341<ul> 3342 <li><tt>Module::global_iterator</tt> - Typedef for global variable list iterator<br> 3343 3344 <tt>Module::const_global_iterator</tt> - Typedef for const_iterator.<br> 3345 3346 <tt>global_begin()</tt>, <tt>global_end()</tt> 3347 <tt>global_size()</tt>, <tt>global_empty()</tt> 3348 3349 <p> These are forwarding methods that make it easy to access the contents of 3350 a <tt>Module</tt> object's <a 3351 href="#GlobalVariable"><tt>GlobalVariable</tt></a> list.</p></li> 3352 3353 <li><tt>Module::GlobalListType &getGlobalList()</tt> 3354 3355 <p>Returns the list of <a 3356 href="#GlobalVariable"><tt>GlobalVariable</tt></a>s. This is necessary to 3357 use when you need to update the list or perform a complex action that 3358 doesn't have a forwarding method.</p> 3359 3360 <p><!-- Symbol table stuff --> </p></li> 3361</ul> 3362 3363<hr> 3364 3365<ul> 3366 <li><tt><a href="#SymbolTable">SymbolTable</a> *getSymbolTable()</tt> 3367 3368 <p>Return a reference to the <a href="#SymbolTable"><tt>SymbolTable</tt></a> 3369 for this <tt>Module</tt>.</p> 3370 3371 <p><!-- Convenience methods --></p></li> 3372</ul> 3373 3374<hr> 3375 3376<ul> 3377 <li><tt><a href="#Function">Function</a> *getFunction(const std::string 3378 &Name, const <a href="#FunctionType">FunctionType</a> *Ty)</tt> 3379 3380 <p>Look up the specified function in the <tt>Module</tt> <a 3381 href="#SymbolTable"><tt>SymbolTable</tt></a>. If it does not exist, return 3382 <tt>null</tt>.</p></li> 3383 3384 <li><tt><a href="#Function">Function</a> *getOrInsertFunction(const 3385 std::string &Name, const <a href="#FunctionType">FunctionType</a> *T)</tt> 3386 3387 <p>Look up the specified function in the <tt>Module</tt> <a 3388 href="#SymbolTable"><tt>SymbolTable</tt></a>. If it does not exist, add an 3389 external declaration for the function and return it.</p></li> 3390 3391 <li><tt>std::string getTypeName(const <a href="#Type">Type</a> *Ty)</tt> 3392 3393 <p>If there is at least one entry in the <a 3394 href="#SymbolTable"><tt>SymbolTable</tt></a> for the specified <a 3395 href="#Type"><tt>Type</tt></a>, return it. Otherwise return the empty 3396 string.</p></li> 3397 3398 <li><tt>bool addTypeName(const std::string &Name, const <a 3399 href="#Type">Type</a> *Ty)</tt> 3400 3401 <p>Insert an entry in the <a href="#SymbolTable"><tt>SymbolTable</tt></a> 3402 mapping <tt>Name</tt> to <tt>Ty</tt>. If there is already an entry for this 3403 name, true is returned and the <a 3404 href="#SymbolTable"><tt>SymbolTable</tt></a> is not modified.</p></li> 3405</ul> 3406 3407</div> 3408 3409</div> 3410 3411<!-- ======================================================================= --> 3412<h3> 3413 <a name="Value">The <tt>Value</tt> class</a> 3414</h3> 3415 3416<div> 3417 3418<p><tt>#include "<a href="/doxygen/Value_8h-source.html">llvm/Value.h</a>"</tt> 3419<br> 3420doxygen info: <a href="/doxygen/classllvm_1_1Value.html">Value Class</a></p> 3421 3422<p>The <tt>Value</tt> class is the most important class in the LLVM Source 3423base. It represents a typed value that may be used (among other things) as an 3424operand to an instruction. There are many different types of <tt>Value</tt>s, 3425such as <a href="#Constant"><tt>Constant</tt></a>s,<a 3426href="#Argument"><tt>Argument</tt></a>s. Even <a 3427href="#Instruction"><tt>Instruction</tt></a>s and <a 3428href="#Function"><tt>Function</tt></a>s are <tt>Value</tt>s.</p> 3429 3430<p>A particular <tt>Value</tt> may be used many times in the LLVM representation 3431for a program. For example, an incoming argument to a function (represented 3432with an instance of the <a href="#Argument">Argument</a> class) is "used" by 3433every instruction in the function that references the argument. To keep track 3434of this relationship, the <tt>Value</tt> class keeps a list of all of the <a 3435href="#User"><tt>User</tt></a>s that is using it (the <a 3436href="#User"><tt>User</tt></a> class is a base class for all nodes in the LLVM 3437graph that can refer to <tt>Value</tt>s). This use list is how LLVM represents 3438def-use information in the program, and is accessible through the <tt>use_</tt>* 3439methods, shown below.</p> 3440 3441<p>Because LLVM is a typed representation, every LLVM <tt>Value</tt> is typed, 3442and this <a href="#Type">Type</a> is available through the <tt>getType()</tt> 3443method. In addition, all LLVM values can be named. The "name" of the 3444<tt>Value</tt> is a symbolic string printed in the LLVM code:</p> 3445 3446<div class="doc_code"> 3447<pre> 3448%<b>foo</b> = add i32 1, 2 3449</pre> 3450</div> 3451 3452<p><a name="nameWarning">The name of this instruction is "foo".</a> <b>NOTE</b> 3453that the name of any value may be missing (an empty string), so names should 3454<b>ONLY</b> be used for debugging (making the source code easier to read, 3455debugging printouts), they should not be used to keep track of values or map 3456between them. For this purpose, use a <tt>std::map</tt> of pointers to the 3457<tt>Value</tt> itself instead.</p> 3458 3459<p>One important aspect of LLVM is that there is no distinction between an SSA 3460variable and the operation that produces it. Because of this, any reference to 3461the value produced by an instruction (or the value available as an incoming 3462argument, for example) is represented as a direct pointer to the instance of 3463the class that 3464represents this value. Although this may take some getting used to, it 3465simplifies the representation and makes it easier to manipulate.</p> 3466 3467<!-- _______________________________________________________________________ --> 3468<h4> 3469 <a name="m_Value">Important Public Members of the <tt>Value</tt> class</a> 3470</h4> 3471 3472<div> 3473 3474<ul> 3475 <li><tt>Value::use_iterator</tt> - Typedef for iterator over the 3476use-list<br> 3477 <tt>Value::const_use_iterator</tt> - Typedef for const_iterator over 3478the use-list<br> 3479 <tt>unsigned use_size()</tt> - Returns the number of users of the 3480value.<br> 3481 <tt>bool use_empty()</tt> - Returns true if there are no users.<br> 3482 <tt>use_iterator use_begin()</tt> - Get an iterator to the start of 3483the use-list.<br> 3484 <tt>use_iterator use_end()</tt> - Get an iterator to the end of the 3485use-list.<br> 3486 <tt><a href="#User">User</a> *use_back()</tt> - Returns the last 3487element in the list. 3488 <p> These methods are the interface to access the def-use 3489information in LLVM. As with all other iterators in LLVM, the naming 3490conventions follow the conventions defined by the <a href="#stl">STL</a>.</p> 3491 </li> 3492 <li><tt><a href="#Type">Type</a> *getType() const</tt> 3493 <p>This method returns the Type of the Value.</p> 3494 </li> 3495 <li><tt>bool hasName() const</tt><br> 3496 <tt>std::string getName() const</tt><br> 3497 <tt>void setName(const std::string &Name)</tt> 3498 <p> This family of methods is used to access and assign a name to a <tt>Value</tt>, 3499be aware of the <a href="#nameWarning">precaution above</a>.</p> 3500 </li> 3501 <li><tt>void replaceAllUsesWith(Value *V)</tt> 3502 3503 <p>This method traverses the use list of a <tt>Value</tt> changing all <a 3504 href="#User"><tt>User</tt>s</a> of the current value to refer to 3505 "<tt>V</tt>" instead. For example, if you detect that an instruction always 3506 produces a constant value (for example through constant folding), you can 3507 replace all uses of the instruction with the constant like this:</p> 3508 3509<div class="doc_code"> 3510<pre> 3511Inst->replaceAllUsesWith(ConstVal); 3512</pre> 3513</div> 3514 3515</ul> 3516 3517</div> 3518 3519</div> 3520 3521<!-- ======================================================================= --> 3522<h3> 3523 <a name="User">The <tt>User</tt> class</a> 3524</h3> 3525 3526<div> 3527 3528<p> 3529<tt>#include "<a href="/doxygen/User_8h-source.html">llvm/User.h</a>"</tt><br> 3530doxygen info: <a href="/doxygen/classllvm_1_1User.html">User Class</a><br> 3531Superclass: <a href="#Value"><tt>Value</tt></a></p> 3532 3533<p>The <tt>User</tt> class is the common base class of all LLVM nodes that may 3534refer to <a href="#Value"><tt>Value</tt></a>s. It exposes a list of "Operands" 3535that are all of the <a href="#Value"><tt>Value</tt></a>s that the User is 3536referring to. The <tt>User</tt> class itself is a subclass of 3537<tt>Value</tt>.</p> 3538 3539<p>The operands of a <tt>User</tt> point directly to the LLVM <a 3540href="#Value"><tt>Value</tt></a> that it refers to. Because LLVM uses Static 3541Single Assignment (SSA) form, there can only be one definition referred to, 3542allowing this direct connection. This connection provides the use-def 3543information in LLVM.</p> 3544 3545<!-- _______________________________________________________________________ --> 3546<h4> 3547 <a name="m_User">Important Public Members of the <tt>User</tt> class</a> 3548</h4> 3549 3550<div> 3551 3552<p>The <tt>User</tt> class exposes the operand list in two ways: through 3553an index access interface and through an iterator based interface.</p> 3554 3555<ul> 3556 <li><tt>Value *getOperand(unsigned i)</tt><br> 3557 <tt>unsigned getNumOperands()</tt> 3558 <p> These two methods expose the operands of the <tt>User</tt> in a 3559convenient form for direct access.</p></li> 3560 3561 <li><tt>User::op_iterator</tt> - Typedef for iterator over the operand 3562list<br> 3563 <tt>op_iterator op_begin()</tt> - Get an iterator to the start of 3564the operand list.<br> 3565 <tt>op_iterator op_end()</tt> - Get an iterator to the end of the 3566operand list. 3567 <p> Together, these methods make up the iterator based interface to 3568the operands of a <tt>User</tt>.</p></li> 3569</ul> 3570 3571</div> 3572 3573</div> 3574 3575<!-- ======================================================================= --> 3576<h3> 3577 <a name="Instruction">The <tt>Instruction</tt> class</a> 3578</h3> 3579 3580<div> 3581 3582<p><tt>#include "</tt><tt><a 3583href="/doxygen/Instruction_8h-source.html">llvm/Instruction.h</a>"</tt><br> 3584doxygen info: <a href="/doxygen/classllvm_1_1Instruction.html">Instruction Class</a><br> 3585Superclasses: <a href="#User"><tt>User</tt></a>, <a 3586href="#Value"><tt>Value</tt></a></p> 3587 3588<p>The <tt>Instruction</tt> class is the common base class for all LLVM 3589instructions. It provides only a few methods, but is a very commonly used 3590class. The primary data tracked by the <tt>Instruction</tt> class itself is the 3591opcode (instruction type) and the parent <a 3592href="#BasicBlock"><tt>BasicBlock</tt></a> the <tt>Instruction</tt> is embedded 3593into. To represent a specific type of instruction, one of many subclasses of 3594<tt>Instruction</tt> are used.</p> 3595 3596<p> Because the <tt>Instruction</tt> class subclasses the <a 3597href="#User"><tt>User</tt></a> class, its operands can be accessed in the same 3598way as for other <a href="#User"><tt>User</tt></a>s (with the 3599<tt>getOperand()</tt>/<tt>getNumOperands()</tt> and 3600<tt>op_begin()</tt>/<tt>op_end()</tt> methods).</p> <p> An important file for 3601the <tt>Instruction</tt> class is the <tt>llvm/Instruction.def</tt> file. This 3602file contains some meta-data about the various different types of instructions 3603in LLVM. It describes the enum values that are used as opcodes (for example 3604<tt>Instruction::Add</tt> and <tt>Instruction::ICmp</tt>), as well as the 3605concrete sub-classes of <tt>Instruction</tt> that implement the instruction (for 3606example <tt><a href="#BinaryOperator">BinaryOperator</a></tt> and <tt><a 3607href="#CmpInst">CmpInst</a></tt>). Unfortunately, the use of macros in 3608this file confuses doxygen, so these enum values don't show up correctly in the 3609<a href="/doxygen/classllvm_1_1Instruction.html">doxygen output</a>.</p> 3610 3611<!-- _______________________________________________________________________ --> 3612<h4> 3613 <a name="s_Instruction"> 3614 Important Subclasses of the <tt>Instruction</tt> class 3615 </a> 3616</h4> 3617<div> 3618 <ul> 3619 <li><tt><a name="BinaryOperator">BinaryOperator</a></tt> 3620 <p>This subclasses represents all two operand instructions whose operands 3621 must be the same type, except for the comparison instructions.</p></li> 3622 <li><tt><a name="CastInst">CastInst</a></tt> 3623 <p>This subclass is the parent of the 12 casting instructions. It provides 3624 common operations on cast instructions.</p> 3625 <li><tt><a name="CmpInst">CmpInst</a></tt> 3626 <p>This subclass respresents the two comparison instructions, 3627 <a href="LangRef.html#i_icmp">ICmpInst</a> (integer opreands), and 3628 <a href="LangRef.html#i_fcmp">FCmpInst</a> (floating point operands).</p> 3629 <li><tt><a name="TerminatorInst">TerminatorInst</a></tt> 3630 <p>This subclass is the parent of all terminator instructions (those which 3631 can terminate a block).</p> 3632 </ul> 3633 </div> 3634 3635<!-- _______________________________________________________________________ --> 3636<h4> 3637 <a name="m_Instruction"> 3638 Important Public Members of the <tt>Instruction</tt> class 3639 </a> 3640</h4> 3641 3642<div> 3643 3644<ul> 3645 <li><tt><a href="#BasicBlock">BasicBlock</a> *getParent()</tt> 3646 <p>Returns the <a href="#BasicBlock"><tt>BasicBlock</tt></a> that 3647this <tt>Instruction</tt> is embedded into.</p></li> 3648 <li><tt>bool mayWriteToMemory()</tt> 3649 <p>Returns true if the instruction writes to memory, i.e. it is a 3650 <tt>call</tt>,<tt>free</tt>,<tt>invoke</tt>, or <tt>store</tt>.</p></li> 3651 <li><tt>unsigned getOpcode()</tt> 3652 <p>Returns the opcode for the <tt>Instruction</tt>.</p></li> 3653 <li><tt><a href="#Instruction">Instruction</a> *clone() const</tt> 3654 <p>Returns another instance of the specified instruction, identical 3655in all ways to the original except that the instruction has no parent 3656(ie it's not embedded into a <a href="#BasicBlock"><tt>BasicBlock</tt></a>), 3657and it has no name</p></li> 3658</ul> 3659 3660</div> 3661 3662</div> 3663 3664<!-- ======================================================================= --> 3665<h3> 3666 <a name="Constant">The <tt>Constant</tt> class and subclasses</a> 3667</h3> 3668 3669<div> 3670 3671<p>Constant represents a base class for different types of constants. It 3672is subclassed by ConstantInt, ConstantArray, etc. for representing 3673the various types of Constants. <a href="#GlobalValue">GlobalValue</a> is also 3674a subclass, which represents the address of a global variable or function. 3675</p> 3676 3677<!-- _______________________________________________________________________ --> 3678<h4>Important Subclasses of Constant</h4> 3679<div> 3680<ul> 3681 <li>ConstantInt : This subclass of Constant represents an integer constant of 3682 any width. 3683 <ul> 3684 <li><tt>const APInt& getValue() const</tt>: Returns the underlying 3685 value of this constant, an APInt value.</li> 3686 <li><tt>int64_t getSExtValue() const</tt>: Converts the underlying APInt 3687 value to an int64_t via sign extension. If the value (not the bit width) 3688 of the APInt is too large to fit in an int64_t, an assertion will result. 3689 For this reason, use of this method is discouraged.</li> 3690 <li><tt>uint64_t getZExtValue() const</tt>: Converts the underlying APInt 3691 value to a uint64_t via zero extension. IF the value (not the bit width) 3692 of the APInt is too large to fit in a uint64_t, an assertion will result. 3693 For this reason, use of this method is discouraged.</li> 3694 <li><tt>static ConstantInt* get(const APInt& Val)</tt>: Returns the 3695 ConstantInt object that represents the value provided by <tt>Val</tt>. 3696 The type is implied as the IntegerType that corresponds to the bit width 3697 of <tt>Val</tt>.</li> 3698 <li><tt>static ConstantInt* get(const Type *Ty, uint64_t Val)</tt>: 3699 Returns the ConstantInt object that represents the value provided by 3700 <tt>Val</tt> for integer type <tt>Ty</tt>.</li> 3701 </ul> 3702 </li> 3703 <li>ConstantFP : This class represents a floating point constant. 3704 <ul> 3705 <li><tt>double getValue() const</tt>: Returns the underlying value of 3706 this constant. </li> 3707 </ul> 3708 </li> 3709 <li>ConstantArray : This represents a constant array. 3710 <ul> 3711 <li><tt>const std::vector<Use> &getValues() const</tt>: Returns 3712 a vector of component constants that makeup this array. </li> 3713 </ul> 3714 </li> 3715 <li>ConstantStruct : This represents a constant struct. 3716 <ul> 3717 <li><tt>const std::vector<Use> &getValues() const</tt>: Returns 3718 a vector of component constants that makeup this array. </li> 3719 </ul> 3720 </li> 3721 <li>GlobalValue : This represents either a global variable or a function. In 3722 either case, the value is a constant fixed address (after linking). 3723 </li> 3724</ul> 3725</div> 3726 3727</div> 3728 3729<!-- ======================================================================= --> 3730<h3> 3731 <a name="GlobalValue">The <tt>GlobalValue</tt> class</a> 3732</h3> 3733 3734<div> 3735 3736<p><tt>#include "<a 3737href="/doxygen/GlobalValue_8h-source.html">llvm/GlobalValue.h</a>"</tt><br> 3738doxygen info: <a href="/doxygen/classllvm_1_1GlobalValue.html">GlobalValue 3739Class</a><br> 3740Superclasses: <a href="#Constant"><tt>Constant</tt></a>, 3741<a href="#User"><tt>User</tt></a>, <a href="#Value"><tt>Value</tt></a></p> 3742 3743<p>Global values (<a href="#GlobalVariable"><tt>GlobalVariable</tt></a>s or <a 3744href="#Function"><tt>Function</tt></a>s) are the only LLVM values that are 3745visible in the bodies of all <a href="#Function"><tt>Function</tt></a>s. 3746Because they are visible at global scope, they are also subject to linking with 3747other globals defined in different translation units. To control the linking 3748process, <tt>GlobalValue</tt>s know their linkage rules. Specifically, 3749<tt>GlobalValue</tt>s know whether they have internal or external linkage, as 3750defined by the <tt>LinkageTypes</tt> enumeration.</p> 3751 3752<p>If a <tt>GlobalValue</tt> has internal linkage (equivalent to being 3753<tt>static</tt> in C), it is not visible to code outside the current translation 3754unit, and does not participate in linking. If it has external linkage, it is 3755visible to external code, and does participate in linking. In addition to 3756linkage information, <tt>GlobalValue</tt>s keep track of which <a 3757href="#Module"><tt>Module</tt></a> they are currently part of.</p> 3758 3759<p>Because <tt>GlobalValue</tt>s are memory objects, they are always referred to 3760by their <b>address</b>. As such, the <a href="#Type"><tt>Type</tt></a> of a 3761global is always a pointer to its contents. It is important to remember this 3762when using the <tt>GetElementPtrInst</tt> instruction because this pointer must 3763be dereferenced first. For example, if you have a <tt>GlobalVariable</tt> (a 3764subclass of <tt>GlobalValue)</tt> that is an array of 24 ints, type <tt>[24 x 3765i32]</tt>, then the <tt>GlobalVariable</tt> is a pointer to that array. Although 3766the address of the first element of this array and the value of the 3767<tt>GlobalVariable</tt> are the same, they have different types. The 3768<tt>GlobalVariable</tt>'s type is <tt>[24 x i32]</tt>. The first element's type 3769is <tt>i32.</tt> Because of this, accessing a global value requires you to 3770dereference the pointer with <tt>GetElementPtrInst</tt> first, then its elements 3771can be accessed. This is explained in the <a href="LangRef.html#globalvars">LLVM 3772Language Reference Manual</a>.</p> 3773 3774<!-- _______________________________________________________________________ --> 3775<h4> 3776 <a name="m_GlobalValue"> 3777 Important Public Members of the <tt>GlobalValue</tt> class 3778 </a> 3779</h4> 3780 3781<div> 3782 3783<ul> 3784 <li><tt>bool hasInternalLinkage() const</tt><br> 3785 <tt>bool hasExternalLinkage() const</tt><br> 3786 <tt>void setInternalLinkage(bool HasInternalLinkage)</tt> 3787 <p> These methods manipulate the linkage characteristics of the <tt>GlobalValue</tt>.</p> 3788 <p> </p> 3789 </li> 3790 <li><tt><a href="#Module">Module</a> *getParent()</tt> 3791 <p> This returns the <a href="#Module"><tt>Module</tt></a> that the 3792GlobalValue is currently embedded into.</p></li> 3793</ul> 3794 3795</div> 3796 3797</div> 3798 3799<!-- ======================================================================= --> 3800<h3> 3801 <a name="Function">The <tt>Function</tt> class</a> 3802</h3> 3803 3804<div> 3805 3806<p><tt>#include "<a 3807href="/doxygen/Function_8h-source.html">llvm/Function.h</a>"</tt><br> doxygen 3808info: <a href="/doxygen/classllvm_1_1Function.html">Function Class</a><br> 3809Superclasses: <a href="#GlobalValue"><tt>GlobalValue</tt></a>, 3810<a href="#Constant"><tt>Constant</tt></a>, 3811<a href="#User"><tt>User</tt></a>, 3812<a href="#Value"><tt>Value</tt></a></p> 3813 3814<p>The <tt>Function</tt> class represents a single procedure in LLVM. It is 3815actually one of the more complex classes in the LLVM hierarchy because it must 3816keep track of a large amount of data. The <tt>Function</tt> class keeps track 3817of a list of <a href="#BasicBlock"><tt>BasicBlock</tt></a>s, a list of formal 3818<a href="#Argument"><tt>Argument</tt></a>s, and a 3819<a href="#SymbolTable"><tt>SymbolTable</tt></a>.</p> 3820 3821<p>The list of <a href="#BasicBlock"><tt>BasicBlock</tt></a>s is the most 3822commonly used part of <tt>Function</tt> objects. The list imposes an implicit 3823ordering of the blocks in the function, which indicate how the code will be 3824laid out by the backend. Additionally, the first <a 3825href="#BasicBlock"><tt>BasicBlock</tt></a> is the implicit entry node for the 3826<tt>Function</tt>. It is not legal in LLVM to explicitly branch to this initial 3827block. There are no implicit exit nodes, and in fact there may be multiple exit 3828nodes from a single <tt>Function</tt>. If the <a 3829href="#BasicBlock"><tt>BasicBlock</tt></a> list is empty, this indicates that 3830the <tt>Function</tt> is actually a function declaration: the actual body of the 3831function hasn't been linked in yet.</p> 3832 3833<p>In addition to a list of <a href="#BasicBlock"><tt>BasicBlock</tt></a>s, the 3834<tt>Function</tt> class also keeps track of the list of formal <a 3835href="#Argument"><tt>Argument</tt></a>s that the function receives. This 3836container manages the lifetime of the <a href="#Argument"><tt>Argument</tt></a> 3837nodes, just like the <a href="#BasicBlock"><tt>BasicBlock</tt></a> list does for 3838the <a href="#BasicBlock"><tt>BasicBlock</tt></a>s.</p> 3839 3840<p>The <a href="#SymbolTable"><tt>SymbolTable</tt></a> is a very rarely used 3841LLVM feature that is only used when you have to look up a value by name. Aside 3842from that, the <a href="#SymbolTable"><tt>SymbolTable</tt></a> is used 3843internally to make sure that there are not conflicts between the names of <a 3844href="#Instruction"><tt>Instruction</tt></a>s, <a 3845href="#BasicBlock"><tt>BasicBlock</tt></a>s, or <a 3846href="#Argument"><tt>Argument</tt></a>s in the function body.</p> 3847 3848<p>Note that <tt>Function</tt> is a <a href="#GlobalValue">GlobalValue</a> 3849and therefore also a <a href="#Constant">Constant</a>. The value of the function 3850is its address (after linking) which is guaranteed to be constant.</p> 3851 3852<!-- _______________________________________________________________________ --> 3853<h4> 3854 <a name="m_Function"> 3855 Important Public Members of the <tt>Function</tt> class 3856 </a> 3857</h4> 3858 3859<div> 3860 3861<ul> 3862 <li><tt>Function(const </tt><tt><a href="#FunctionType">FunctionType</a> 3863 *Ty, LinkageTypes Linkage, const std::string &N = "", Module* Parent = 0)</tt> 3864 3865 <p>Constructor used when you need to create new <tt>Function</tt>s to add 3866 the the program. The constructor must specify the type of the function to 3867 create and what type of linkage the function should have. The <a 3868 href="#FunctionType"><tt>FunctionType</tt></a> argument 3869 specifies the formal arguments and return value for the function. The same 3870 <a href="#FunctionType"><tt>FunctionType</tt></a> value can be used to 3871 create multiple functions. The <tt>Parent</tt> argument specifies the Module 3872 in which the function is defined. If this argument is provided, the function 3873 will automatically be inserted into that module's list of 3874 functions.</p></li> 3875 3876 <li><tt>bool isDeclaration()</tt> 3877 3878 <p>Return whether or not the <tt>Function</tt> has a body defined. If the 3879 function is "external", it does not have a body, and thus must be resolved 3880 by linking with a function defined in a different translation unit.</p></li> 3881 3882 <li><tt>Function::iterator</tt> - Typedef for basic block list iterator<br> 3883 <tt>Function::const_iterator</tt> - Typedef for const_iterator.<br> 3884 3885 <tt>begin()</tt>, <tt>end()</tt> 3886 <tt>size()</tt>, <tt>empty()</tt> 3887 3888 <p>These are forwarding methods that make it easy to access the contents of 3889 a <tt>Function</tt> object's <a href="#BasicBlock"><tt>BasicBlock</tt></a> 3890 list.</p></li> 3891 3892 <li><tt>Function::BasicBlockListType &getBasicBlockList()</tt> 3893 3894 <p>Returns the list of <a href="#BasicBlock"><tt>BasicBlock</tt></a>s. This 3895 is necessary to use when you need to update the list or perform a complex 3896 action that doesn't have a forwarding method.</p></li> 3897 3898 <li><tt>Function::arg_iterator</tt> - Typedef for the argument list 3899iterator<br> 3900 <tt>Function::const_arg_iterator</tt> - Typedef for const_iterator.<br> 3901 3902 <tt>arg_begin()</tt>, <tt>arg_end()</tt> 3903 <tt>arg_size()</tt>, <tt>arg_empty()</tt> 3904 3905 <p>These are forwarding methods that make it easy to access the contents of 3906 a <tt>Function</tt> object's <a href="#Argument"><tt>Argument</tt></a> 3907 list.</p></li> 3908 3909 <li><tt>Function::ArgumentListType &getArgumentList()</tt> 3910 3911 <p>Returns the list of <a href="#Argument"><tt>Argument</tt></a>s. This is 3912 necessary to use when you need to update the list or perform a complex 3913 action that doesn't have a forwarding method.</p></li> 3914 3915 <li><tt><a href="#BasicBlock">BasicBlock</a> &getEntryBlock()</tt> 3916 3917 <p>Returns the entry <a href="#BasicBlock"><tt>BasicBlock</tt></a> for the 3918 function. Because the entry block for the function is always the first 3919 block, this returns the first block of the <tt>Function</tt>.</p></li> 3920 3921 <li><tt><a href="#Type">Type</a> *getReturnType()</tt><br> 3922 <tt><a href="#FunctionType">FunctionType</a> *getFunctionType()</tt> 3923 3924 <p>This traverses the <a href="#Type"><tt>Type</tt></a> of the 3925 <tt>Function</tt> and returns the return type of the function, or the <a 3926 href="#FunctionType"><tt>FunctionType</tt></a> of the actual 3927 function.</p></li> 3928 3929 <li><tt><a href="#SymbolTable">SymbolTable</a> *getSymbolTable()</tt> 3930 3931 <p> Return a pointer to the <a href="#SymbolTable"><tt>SymbolTable</tt></a> 3932 for this <tt>Function</tt>.</p></li> 3933</ul> 3934 3935</div> 3936 3937</div> 3938 3939<!-- ======================================================================= --> 3940<h3> 3941 <a name="GlobalVariable">The <tt>GlobalVariable</tt> class</a> 3942</h3> 3943 3944<div> 3945 3946<p><tt>#include "<a 3947href="/doxygen/GlobalVariable_8h-source.html">llvm/GlobalVariable.h</a>"</tt> 3948<br> 3949doxygen info: <a href="/doxygen/classllvm_1_1GlobalVariable.html">GlobalVariable 3950 Class</a><br> 3951Superclasses: <a href="#GlobalValue"><tt>GlobalValue</tt></a>, 3952<a href="#Constant"><tt>Constant</tt></a>, 3953<a href="#User"><tt>User</tt></a>, 3954<a href="#Value"><tt>Value</tt></a></p> 3955 3956<p>Global variables are represented with the (surprise surprise) 3957<tt>GlobalVariable</tt> class. Like functions, <tt>GlobalVariable</tt>s are also 3958subclasses of <a href="#GlobalValue"><tt>GlobalValue</tt></a>, and as such are 3959always referenced by their address (global values must live in memory, so their 3960"name" refers to their constant address). See 3961<a href="#GlobalValue"><tt>GlobalValue</tt></a> for more on this. Global 3962variables may have an initial value (which must be a 3963<a href="#Constant"><tt>Constant</tt></a>), and if they have an initializer, 3964they may be marked as "constant" themselves (indicating that their contents 3965never change at runtime).</p> 3966 3967<!-- _______________________________________________________________________ --> 3968<h4> 3969 <a name="m_GlobalVariable"> 3970 Important Public Members of the <tt>GlobalVariable</tt> class 3971 </a> 3972</h4> 3973 3974<div> 3975 3976<ul> 3977 <li><tt>GlobalVariable(const </tt><tt><a href="#Type">Type</a> *Ty, bool 3978 isConstant, LinkageTypes& Linkage, <a href="#Constant">Constant</a> 3979 *Initializer = 0, const std::string &Name = "", Module* Parent = 0)</tt> 3980 3981 <p>Create a new global variable of the specified type. If 3982 <tt>isConstant</tt> is true then the global variable will be marked as 3983 unchanging for the program. The Linkage parameter specifies the type of 3984 linkage (internal, external, weak, linkonce, appending) for the variable. 3985 If the linkage is InternalLinkage, WeakAnyLinkage, WeakODRLinkage, 3986 LinkOnceAnyLinkage or LinkOnceODRLinkage, then the resultant 3987 global variable will have internal linkage. AppendingLinkage concatenates 3988 together all instances (in different translation units) of the variable 3989 into a single variable but is only applicable to arrays. See 3990 the <a href="LangRef.html#modulestructure">LLVM Language Reference</a> for 3991 further details on linkage types. Optionally an initializer, a name, and the 3992 module to put the variable into may be specified for the global variable as 3993 well.</p></li> 3994 3995 <li><tt>bool isConstant() const</tt> 3996 3997 <p>Returns true if this is a global variable that is known not to 3998 be modified at runtime.</p></li> 3999 4000 <li><tt>bool hasInitializer()</tt> 4001 4002 <p>Returns true if this <tt>GlobalVariable</tt> has an intializer.</p></li> 4003 4004 <li><tt><a href="#Constant">Constant</a> *getInitializer()</tt> 4005 4006 <p>Returns the initial value for a <tt>GlobalVariable</tt>. It is not legal 4007 to call this method if there is no initializer.</p></li> 4008</ul> 4009 4010</div> 4011 4012</div> 4013 4014<!-- ======================================================================= --> 4015<h3> 4016 <a name="BasicBlock">The <tt>BasicBlock</tt> class</a> 4017</h3> 4018 4019<div> 4020 4021<p><tt>#include "<a 4022href="/doxygen/BasicBlock_8h-source.html">llvm/BasicBlock.h</a>"</tt><br> 4023doxygen info: <a href="/doxygen/classllvm_1_1BasicBlock.html">BasicBlock 4024Class</a><br> 4025Superclass: <a href="#Value"><tt>Value</tt></a></p> 4026 4027<p>This class represents a single entry single exit section of the code, 4028commonly known as a basic block by the compiler community. The 4029<tt>BasicBlock</tt> class maintains a list of <a 4030href="#Instruction"><tt>Instruction</tt></a>s, which form the body of the block. 4031Matching the language definition, the last element of this list of instructions 4032is always a terminator instruction (a subclass of the <a 4033href="#TerminatorInst"><tt>TerminatorInst</tt></a> class).</p> 4034 4035<p>In addition to tracking the list of instructions that make up the block, the 4036<tt>BasicBlock</tt> class also keeps track of the <a 4037href="#Function"><tt>Function</tt></a> that it is embedded into.</p> 4038 4039<p>Note that <tt>BasicBlock</tt>s themselves are <a 4040href="#Value"><tt>Value</tt></a>s, because they are referenced by instructions 4041like branches and can go in the switch tables. <tt>BasicBlock</tt>s have type 4042<tt>label</tt>.</p> 4043 4044<!-- _______________________________________________________________________ --> 4045<h4> 4046 <a name="m_BasicBlock"> 4047 Important Public Members of the <tt>BasicBlock</tt> class 4048 </a> 4049</h4> 4050 4051<div> 4052<ul> 4053 4054<li><tt>BasicBlock(const std::string &Name = "", </tt><tt><a 4055 href="#Function">Function</a> *Parent = 0)</tt> 4056 4057<p>The <tt>BasicBlock</tt> constructor is used to create new basic blocks for 4058insertion into a function. The constructor optionally takes a name for the new 4059block, and a <a href="#Function"><tt>Function</tt></a> to insert it into. If 4060the <tt>Parent</tt> parameter is specified, the new <tt>BasicBlock</tt> is 4061automatically inserted at the end of the specified <a 4062href="#Function"><tt>Function</tt></a>, if not specified, the BasicBlock must be 4063manually inserted into the <a href="#Function"><tt>Function</tt></a>.</p></li> 4064 4065<li><tt>BasicBlock::iterator</tt> - Typedef for instruction list iterator<br> 4066<tt>BasicBlock::const_iterator</tt> - Typedef for const_iterator.<br> 4067<tt>begin()</tt>, <tt>end()</tt>, <tt>front()</tt>, <tt>back()</tt>, 4068<tt>size()</tt>, <tt>empty()</tt> 4069STL-style functions for accessing the instruction list. 4070 4071<p>These methods and typedefs are forwarding functions that have the same 4072semantics as the standard library methods of the same names. These methods 4073expose the underlying instruction list of a basic block in a way that is easy to 4074manipulate. To get the full complement of container operations (including 4075operations to update the list), you must use the <tt>getInstList()</tt> 4076method.</p></li> 4077 4078<li><tt>BasicBlock::InstListType &getInstList()</tt> 4079 4080<p>This method is used to get access to the underlying container that actually 4081holds the Instructions. This method must be used when there isn't a forwarding 4082function in the <tt>BasicBlock</tt> class for the operation that you would like 4083to perform. Because there are no forwarding functions for "updating" 4084operations, you need to use this if you want to update the contents of a 4085<tt>BasicBlock</tt>.</p></li> 4086 4087<li><tt><a href="#Function">Function</a> *getParent()</tt> 4088 4089<p> Returns a pointer to <a href="#Function"><tt>Function</tt></a> the block is 4090embedded into, or a null pointer if it is homeless.</p></li> 4091 4092<li><tt><a href="#TerminatorInst">TerminatorInst</a> *getTerminator()</tt> 4093 4094<p> Returns a pointer to the terminator instruction that appears at the end of 4095the <tt>BasicBlock</tt>. If there is no terminator instruction, or if the last 4096instruction in the block is not a terminator, then a null pointer is 4097returned.</p></li> 4098 4099</ul> 4100 4101</div> 4102 4103</div> 4104 4105<!-- ======================================================================= --> 4106<h3> 4107 <a name="Argument">The <tt>Argument</tt> class</a> 4108</h3> 4109 4110<div> 4111 4112<p>This subclass of Value defines the interface for incoming formal 4113arguments to a function. A Function maintains a list of its formal 4114arguments. An argument has a pointer to the parent Function.</p> 4115 4116</div> 4117 4118</div> 4119 4120<!-- *********************************************************************** --> 4121<hr> 4122<address> 4123 <a href="http://jigsaw.w3.org/css-validator/check/referer"><img 4124 src="http://jigsaw.w3.org/css-validator/images/vcss-blue" alt="Valid CSS"></a> 4125 <a href="http://validator.w3.org/check/referer"><img 4126 src="http://www.w3.org/Icons/valid-html401" alt="Valid HTML 4.01 Strict"></a> 4127 4128 <a href="mailto:dhurjati@cs.uiuc.edu">Dinakar Dhurjati</a> and 4129 <a href="mailto:sabre@nondot.org">Chris Lattner</a><br> 4130 <a href="http://llvm.org/">The LLVM Compiler Infrastructure</a><br> 4131 Last modified: $Date$ 4132</address> 4133 4134</body> 4135</html> 4136