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>Writing an LLVM Pass</title> 7 <link rel="stylesheet" href="llvm.css" type="text/css"> 8 </head> 9 <body> 10 11 <h1> 12 Writing an LLVM Pass 13 </h1> 14 15 <ol> 16 <li><a href="#introduction">Introduction - What is a pass?</a></li> 17 <li><a href="#quickstart">Quick Start - Writing hello world</a> 18 <ul> 19 <li><a href="#makefile">Setting up the build environment</a></li> 20 <li><a href="#basiccode">Basic code required</a></li> 21 <li><a href="#running">Running a pass with <tt>opt</tt></a></li> 22 </ul></li> 23 <li><a href="#passtype">Pass classes and requirements</a> 24 <ul> 25 <li><a href="#ImmutablePass">The <tt>ImmutablePass</tt> class</a></li> 26 <li><a href="#ModulePass">The <tt>ModulePass</tt> class</a> 27 <ul> 28 <li><a href="#runOnModule">The <tt>runOnModule</tt> method</a></li> 29 </ul></li> 30 <li><a href="#CallGraphSCCPass">The <tt>CallGraphSCCPass</tt> class</a> 31 <ul> 32 <li><a href="#doInitialization_scc">The <tt>doInitialization(CallGraph 33 &)</tt> method</a></li> 34 <li><a href="#runOnSCC">The <tt>runOnSCC</tt> method</a></li> 35 <li><a href="#doFinalization_scc">The <tt>doFinalization(CallGraph 36 &)</tt> method</a></li> 37 </ul></li> 38 <li><a href="#FunctionPass">The <tt>FunctionPass</tt> class</a> 39 <ul> 40 <li><a href="#doInitialization_mod">The <tt>doInitialization(Module 41 &)</tt> method</a></li> 42 <li><a href="#runOnFunction">The <tt>runOnFunction</tt> method</a></li> 43 <li><a href="#doFinalization_mod">The <tt>doFinalization(Module 44 &)</tt> method</a></li> 45 </ul></li> 46 <li><a href="#LoopPass">The <tt>LoopPass</tt> class</a> 47 <ul> 48 <li><a href="#doInitialization_loop">The <tt>doInitialization(Loop *, 49 LPPassManager &)</tt> method</a></li> 50 <li><a href="#runOnLoop">The <tt>runOnLoop</tt> method</a></li> 51 <li><a href="#doFinalization_loop">The <tt>doFinalization() 52 </tt> method</a></li> 53 </ul></li> 54 <li><a href="#RegionPass">The <tt>RegionPass</tt> class</a> 55 <ul> 56 <li><a href="#doInitialization_region">The <tt>doInitialization(Region *, 57 RGPassManager &)</tt> method</a></li> 58 <li><a href="#runOnRegion">The <tt>runOnRegion</tt> method</a></li> 59 <li><a href="#doFinalization_region">The <tt>doFinalization() 60 </tt> method</a></li> 61 </ul></li> 62 <li><a href="#BasicBlockPass">The <tt>BasicBlockPass</tt> class</a> 63 <ul> 64 <li><a href="#doInitialization_fn">The <tt>doInitialization(Function 65 &)</tt> method</a></li> 66 <li><a href="#runOnBasicBlock">The <tt>runOnBasicBlock</tt> 67 method</a></li> 68 <li><a href="#doFinalization_fn">The <tt>doFinalization(Function 69 &)</tt> method</a></li> 70 </ul></li> 71 <li><a href="#MachineFunctionPass">The <tt>MachineFunctionPass</tt> 72 class</a> 73 <ul> 74 <li><a href="#runOnMachineFunction">The 75 <tt>runOnMachineFunction(MachineFunction &)</tt> method</a></li> 76 </ul></li> 77 </ul> 78 <li><a href="#registration">Pass Registration</a> 79 <ul> 80 <li><a href="#print">The <tt>print</tt> method</a></li> 81 </ul></li> 82 <li><a href="#interaction">Specifying interactions between passes</a> 83 <ul> 84 <li><a href="#getAnalysisUsage">The <tt>getAnalysisUsage</tt> 85 method</a></li> 86 <li><a href="#AU::addRequired">The <tt>AnalysisUsage::addRequired<></tt> and <tt>AnalysisUsage::addRequiredTransitive<></tt> methods</a></li> 87 <li><a href="#AU::addPreserved">The <tt>AnalysisUsage::addPreserved<></tt> method</a></li> 88 <li><a href="#AU::examples">Example implementations of <tt>getAnalysisUsage</tt></a></li> 89 <li><a href="#getAnalysis">The <tt>getAnalysis<></tt> and 90 <tt>getAnalysisIfAvailable<></tt> methods</a></li> 91 </ul></li> 92 <li><a href="#analysisgroup">Implementing Analysis Groups</a> 93 <ul> 94 <li><a href="#agconcepts">Analysis Group Concepts</a></li> 95 <li><a href="#registerag">Using <tt>RegisterAnalysisGroup</tt></a></li> 96 </ul></li> 97 <li><a href="#passStatistics">Pass Statistics</a> 98 <li><a href="#passmanager">What PassManager does</a> 99 <ul> 100 <li><a href="#releaseMemory">The <tt>releaseMemory</tt> method</a></li> 101 </ul></li> 102 <li><a href="#registering">Registering dynamically loaded passes</a> 103 <ul> 104 <li><a href="#registering_existing">Using existing registries</a></li> 105 <li><a href="#registering_new">Creating new registries</a></li> 106 </ul></li> 107 <li><a href="#debughints">Using GDB with dynamically loaded passes</a> 108 <ul> 109 <li><a href="#breakpoint">Setting a breakpoint in your pass</a></li> 110 <li><a href="#debugmisc">Miscellaneous Problems</a></li> 111 </ul></li> 112 <li><a href="#future">Future extensions planned</a> 113 <ul> 114 <li><a href="#SMP">Multithreaded LLVM</a></li> 115 </ul></li> 116 </ol> 117 118 <div class="doc_author"> 119 <p>Written by <a href="mailto:sabre (a] nondot.org">Chris Lattner</a> and 120 <a href="mailto:jlaskey (a] mac.com">Jim Laskey</a></p> 121 </div> 122 123 <!-- *********************************************************************** --> 124 <h2> 125 <a name="introduction">Introduction - What is a pass?</a> 126 </h2> 127 <!-- *********************************************************************** --> 128 129 <div> 130 131 <p>The LLVM Pass Framework is an important part of the LLVM system, because LLVM 132 passes are where most of the interesting parts of the compiler exist. Passes 133 perform the transformations and optimizations that make up the compiler, they 134 build the analysis results that are used by these transformations, and they are, 135 above all, a structuring technique for compiler code.</p> 136 137 <p>All LLVM passes are subclasses of the <tt><a 138 href="http://llvm.org/doxygen/classllvm_1_1Pass.html">Pass</a></tt> 139 class, which implement functionality by overriding virtual methods inherited 140 from <tt>Pass</tt>. Depending on how your pass works, you should inherit from 141 the <tt><a href="#ModulePass">ModulePass</a></tt>, <tt><a 142 href="#CallGraphSCCPass">CallGraphSCCPass</a></tt>, <tt><a 143 href="#FunctionPass">FunctionPass</a></tt>, or <tt><a 144 href="#LoopPass">LoopPass</a></tt>, or <tt><a 145 href="#RegionPass">RegionPass</a></tt>, or <tt><a 146 href="#BasicBlockPass">BasicBlockPass</a></tt> classes, which gives the system 147 more information about what your pass does, and how it can be combined with 148 other passes. One of the main features of the LLVM Pass Framework is that it 149 schedules passes to run in an efficient way based on the constraints that your 150 pass meets (which are indicated by which class they derive from).</p> 151 152 <p>We start by showing you how to construct a pass, everything from setting up 153 the code, to compiling, loading, and executing it. After the basics are down, 154 more advanced features are discussed.</p> 155 156 </div> 157 158 <!-- *********************************************************************** --> 159 <h2> 160 <a name="quickstart">Quick Start - Writing hello world</a> 161 </h2> 162 <!-- *********************************************************************** --> 163 164 <div> 165 166 <p>Here we describe how to write the "hello world" of passes. The "Hello" pass 167 is designed to simply print out the name of non-external functions that exist in 168 the program being compiled. It does not modify the program at all, it just 169 inspects it. The source code and files for this pass are available in the LLVM 170 source tree in the <tt>lib/Transforms/Hello</tt> directory.</p> 171 172 <!-- ======================================================================= --> 173 <h3> 174 <a name="makefile">Setting up the build environment</a> 175 </h3> 176 177 <div> 178 179 <p>First, configure and build LLVM. This needs to be done directly inside the 180 LLVM source tree rather than in a separate objects directory. 181 Next, you need to create a new directory somewhere in the LLVM source 182 base. For this example, we'll assume that you made 183 <tt>lib/Transforms/Hello</tt>. Finally, you must set up a build script 184 (Makefile) that will compile the source code for the new pass. To do this, 185 copy the following into <tt>Makefile</tt>:</p> 186 <hr> 187 188 <div class="doc_code"><pre> 189 # Makefile for hello pass 190 191 # Path to top level of LLVM hierarchy 192 LEVEL = ../../.. 193 194 # Name of the library to build 195 LIBRARYNAME = Hello 196 197 # Make the shared library become a loadable module so the tools can 198 # dlopen/dlsym on the resulting library. 199 LOADABLE_MODULE = 1 200 201 # Include the makefile implementation stuff 202 include $(LEVEL)/Makefile.common 203 </pre></div> 204 205 <p>This makefile specifies that all of the <tt>.cpp</tt> files in the current 206 directory are to be compiled and linked together into a shared object 207 <tt>$(LEVEL)/Debug+Asserts/lib/Hello.so</tt> that can be dynamically loaded by 208 the <tt>opt</tt> or <tt>bugpoint</tt> tools via their <tt>-load</tt> options. 209 If your operating system uses a suffix other than .so (such as windows or 210 Mac OS/X), the appropriate extension will be used.</p> 211 212 <p>If you are used CMake to build LLVM, see 213 <a href="CMake.html#passdev">Developing an LLVM pass with CMake</a>.</p> 214 215 <p>Now that we have the build scripts set up, we just need to write the code for 216 the pass itself.</p> 217 218 </div> 219 220 <!-- ======================================================================= --> 221 <h3> 222 <a name="basiccode">Basic code required</a> 223 </h3> 224 225 <div> 226 227 <p>Now that we have a way to compile our new pass, we just have to write it. 228 Start out with:</p> 229 230 <div class="doc_code"><pre> 231 <b>#include</b> "<a href="http://llvm.org/doxygen/Pass_8h-source.html">llvm/Pass.h</a>" 232 <b>#include</b> "<a href="http://llvm.org/doxygen/Function_8h-source.html">llvm/Function.h</a>" 233 <b>#include</b> "<a href="http://llvm.org/doxygen/raw__ostream_8h.html">llvm/Support/raw_ostream.h</a>" 234 </pre></div> 235 236 <p>Which are needed because we are writing a <tt><a 237 href="http://llvm.org/doxygen/classllvm_1_1Pass.html">Pass</a></tt>, 238 we are operating on <tt><a 239 href="http://llvm.org/doxygen/classllvm_1_1Function.html">Function</a></tt>'s, 240 and we will be doing some printing.</p> 241 242 <p>Next we have:</p> 243 <div class="doc_code"><pre> 244 <b>using namespace llvm;</b> 245 </pre></div> 246 <p>... which is required because the functions from the include files 247 live in the llvm namespace. 248 </p> 249 250 <p>Next we have:</p> 251 252 <div class="doc_code"><pre> 253 <b>namespace</b> { 254 </pre></div> 255 256 <p>... which starts out an anonymous namespace. Anonymous namespaces are to C++ 257 what the "<tt>static</tt>" keyword is to C (at global scope). It makes the 258 things declared inside of the anonymous namespace only visible to the current 259 file. If you're not familiar with them, consult a decent C++ book for more 260 information.</p> 261 262 <p>Next, we declare our pass itself:</p> 263 264 <div class="doc_code"><pre> 265 <b>struct</b> Hello : <b>public</b> <a href="#FunctionPass">FunctionPass</a> { 266 </pre></div><p> 267 268 <p>This declares a "<tt>Hello</tt>" class that is a subclass of <tt><a 269 href="http://llvm.org/doxygen/classllvm_1_1FunctionPass.html">FunctionPass</a></tt>. 270 The different builtin pass subclasses are described in detail <a 271 href="#passtype">later</a>, but for now, know that <a 272 href="#FunctionPass"><tt>FunctionPass</tt></a>'s operate a function at a 273 time.</p> 274 275 <div class="doc_code"><pre> 276 static char ID; 277 Hello() : FunctionPass(ID) {} 278 </pre></div><p> 279 280 <p> This declares pass identifier used by LLVM to identify pass. This allows LLVM to 281 avoid using expensive C++ runtime information.</p> 282 283 <div class="doc_code"><pre> 284 <b>virtual bool</b> <a href="#runOnFunction">runOnFunction</a>(Function &F) { 285 errs() << "<i>Hello: </i>" << F.getName() << "\n"; 286 <b>return false</b>; 287 } 288 }; <i>// end of struct Hello</i> 289 </pre></div> 290 291 <p>We declare a "<a href="#runOnFunction"><tt>runOnFunction</tt></a>" method, 292 which overloads an abstract virtual method inherited from <a 293 href="#FunctionPass"><tt>FunctionPass</tt></a>. This is where we are supposed 294 to do our thing, so we just print out our message with the name of each 295 function.</p> 296 297 <div class="doc_code"><pre> 298 char Hello::ID = 0; 299 </pre></div> 300 301 <p> We initialize pass ID here. LLVM uses ID's address to identify pass so 302 initialization value is not important.</p> 303 304 <div class="doc_code"><pre> 305 static RegisterPass<Hello> X("<i>hello</i>", "<i>Hello World Pass</i>", 306 false /* Only looks at CFG */, 307 false /* Analysis Pass */); 308 } <i>// end of anonymous namespace</i> 309 </pre></div> 310 311 <p>Lastly, we <a href="#registration">register our class</a> <tt>Hello</tt>, 312 giving it a command line 313 argument "<tt>hello</tt>", and a name "<tt>Hello World Pass</tt>". 314 Last two arguments describe its behavior. 315 If a pass walks CFG without modifying it then third argument is set to true. 316 If a pass is an analysis pass, for example dominator tree pass, then true 317 is supplied as fourth argument. </p> 318 319 <p>As a whole, the <tt>.cpp</tt> file looks like:</p> 320 321 <div class="doc_code"><pre> 322 <b>#include</b> "<a href="http://llvm.org/doxygen/Pass_8h-source.html">llvm/Pass.h</a>" 323 <b>#include</b> "<a href="http://llvm.org/doxygen/Function_8h-source.html">llvm/Function.h</a>" 324 <b>#include</b> "<a href="http://llvm.org/doxygen/raw__ostream_8h.html">llvm/Support/raw_ostream.h</a>" 325 326 <b>using namespace llvm;</b> 327 328 <b>namespace</b> { 329 <b>struct Hello</b> : <b>public</b> <a href="#FunctionPass">FunctionPass</a> { 330 331 static char ID; 332 Hello() : FunctionPass(ID) {} 333 334 <b>virtual bool</b> <a href="#runOnFunction">runOnFunction</a>(Function &F) { 335 errs() << "<i>Hello: </i>" << F.getName() << "\n"; 336 <b>return false</b>; 337 } 338 }; 339 340 char Hello::ID = 0; 341 static RegisterPass<Hello> X("hello", "Hello World Pass", false, false); 342 } 343 344 </pre></div> 345 346 <p>Now that it's all together, compile the file with a simple "<tt>gmake</tt>" 347 command in the local directory and you should get a new file 348 "<tt>Debug+Asserts/lib/Hello.so</tt>" under the top level directory of the LLVM 349 source tree (not in the local directory). Note that everything in this file is 350 contained in an anonymous namespace: this reflects the fact that passes are self 351 contained units that do not need external interfaces (although they can have 352 them) to be useful.</p> 353 354 </div> 355 356 <!-- ======================================================================= --> 357 <h3> 358 <a name="running">Running a pass with <tt>opt</tt></a> 359 </h3> 360 361 <div> 362 363 <p>Now that you have a brand new shiny shared object file, we can use the 364 <tt>opt</tt> command to run an LLVM program through your pass. Because you 365 registered your pass with <tt>RegisterPass</tt>, you will be able to 366 use the <tt>opt</tt> tool to access it, once loaded.</p> 367 368 <p>To test it, follow the example at the end of the <a 369 href="GettingStarted.html">Getting Started Guide</a> to compile "Hello World" to 370 LLVM. We can now run the bitcode file (<tt>hello.bc</tt>) for the program 371 through our transformation like this (or course, any bitcode file will 372 work):</p> 373 374 <div class="doc_code"><pre> 375 $ opt -load ../../../Debug+Asserts/lib/Hello.so -hello < hello.bc > /dev/null 376 Hello: __main 377 Hello: puts 378 Hello: main 379 </pre></div> 380 381 <p>The '<tt>-load</tt>' option specifies that '<tt>opt</tt>' should load your 382 pass as a shared object, which makes '<tt>-hello</tt>' a valid command line 383 argument (which is one reason you need to <a href="#registration">register your 384 pass</a>). Because the hello pass does not modify the program in any 385 interesting way, we just throw away the result of <tt>opt</tt> (sending it to 386 <tt>/dev/null</tt>).</p> 387 388 <p>To see what happened to the other string you registered, try running 389 <tt>opt</tt> with the <tt>-help</tt> option:</p> 390 391 <div class="doc_code"><pre> 392 $ opt -load ../../../Debug+Asserts/lib/Hello.so -help 393 OVERVIEW: llvm .bc -> .bc modular optimizer 394 395 USAGE: opt [options] <input bitcode> 396 397 OPTIONS: 398 Optimizations available: 399 ... 400 -funcresolve - Resolve Functions 401 -gcse - Global Common Subexpression Elimination 402 -globaldce - Dead Global Elimination 403 <b>-hello - Hello World Pass</b> 404 -indvars - Canonicalize Induction Variables 405 -inline - Function Integration/Inlining 406 -instcombine - Combine redundant instructions 407 ... 408 </pre></div> 409 410 <p>The pass name get added as the information string for your pass, giving some 411 documentation to users of <tt>opt</tt>. Now that you have a working pass, you 412 would go ahead and make it do the cool transformations you want. Once you get 413 it all working and tested, it may become useful to find out how fast your pass 414 is. The <a href="#passManager"><tt>PassManager</tt></a> provides a nice command 415 line option (<tt>--time-passes</tt>) that allows you to get information about 416 the execution time of your pass along with the other passes you queue up. For 417 example:</p> 418 419 <div class="doc_code"><pre> 420 $ opt -load ../../../Debug+Asserts/lib/Hello.so -hello -time-passes < hello.bc > /dev/null 421 Hello: __main 422 Hello: puts 423 Hello: main 424 =============================================================================== 425 ... Pass execution timing report ... 426 =============================================================================== 427 Total Execution Time: 0.02 seconds (0.0479059 wall clock) 428 429 ---User Time--- --System Time-- --User+System-- ---Wall Time--- --- Pass Name --- 430 0.0100 (100.0%) 0.0000 ( 0.0%) 0.0100 ( 50.0%) 0.0402 ( 84.0%) Bitcode Writer 431 0.0000 ( 0.0%) 0.0100 (100.0%) 0.0100 ( 50.0%) 0.0031 ( 6.4%) Dominator Set Construction 432 0.0000 ( 0.0%) 0.0000 ( 0.0%) 0.0000 ( 0.0%) 0.0013 ( 2.7%) Module Verifier 433 <b> 0.0000 ( 0.0%) 0.0000 ( 0.0%) 0.0000 ( 0.0%) 0.0033 ( 6.9%) Hello World Pass</b> 434 0.0100 (100.0%) 0.0100 (100.0%) 0.0200 (100.0%) 0.0479 (100.0%) TOTAL 435 </pre></div> 436 437 <p>As you can see, our implementation above is pretty fast :). The additional 438 passes listed are automatically inserted by the '<tt>opt</tt>' tool to verify 439 that the LLVM emitted by your pass is still valid and well formed LLVM, which 440 hasn't been broken somehow.</p> 441 442 <p>Now that you have seen the basics of the mechanics behind passes, we can talk 443 about some more details of how they work and how to use them.</p> 444 445 </div> 446 447 </div> 448 449 <!-- *********************************************************************** --> 450 <h2> 451 <a name="passtype">Pass classes and requirements</a> 452 </h2> 453 <!-- *********************************************************************** --> 454 455 <div> 456 457 <p>One of the first things that you should do when designing a new pass is to 458 decide what class you should subclass for your pass. The <a 459 href="#basiccode">Hello World</a> example uses the <tt><a 460 href="#FunctionPass">FunctionPass</a></tt> class for its implementation, but we 461 did not discuss why or when this should occur. Here we talk about the classes 462 available, from the most general to the most specific.</p> 463 464 <p>When choosing a superclass for your Pass, you should choose the <b>most 465 specific</b> class possible, while still being able to meet the requirements 466 listed. This gives the LLVM Pass Infrastructure information necessary to 467 optimize how passes are run, so that the resultant compiler isn't unnecessarily 468 slow.</p> 469 470 <!-- ======================================================================= --> 471 <h3> 472 <a name="ImmutablePass">The <tt>ImmutablePass</tt> class</a> 473 </h3> 474 475 <div> 476 477 <p>The most plain and boring type of pass is the "<tt><a 478 href="http://llvm.org/doxygen/classllvm_1_1ImmutablePass.html">ImmutablePass</a></tt>" 479 class. This pass type is used for passes that do not have to be run, do not 480 change state, and never need to be updated. This is not a normal type of 481 transformation or analysis, but can provide information about the current 482 compiler configuration.</p> 483 484 <p>Although this pass class is very infrequently used, it is important for 485 providing information about the current target machine being compiled for, and 486 other static information that can affect the various transformations.</p> 487 488 <p><tt>ImmutablePass</tt>es never invalidate other transformations, are never 489 invalidated, and are never "run".</p> 490 491 </div> 492 493 <!-- ======================================================================= --> 494 <h3> 495 <a name="ModulePass">The <tt>ModulePass</tt> class</a> 496 </h3> 497 498 <div> 499 500 <p>The "<tt><a 501 href="http://llvm.org/doxygen/classllvm_1_1ModulePass.html">ModulePass</a></tt>" 502 class is the most general of all superclasses that you can use. Deriving from 503 <tt>ModulePass</tt> indicates that your pass uses the entire program as a unit, 504 referring to function bodies in no predictable order, or adding and removing 505 functions. Because nothing is known about the behavior of <tt>ModulePass</tt> 506 subclasses, no optimization can be done for their execution.</p> 507 508 <p>A module pass can use function level passes (e.g. dominators) using 509 the getAnalysis interface 510 <tt>getAnalysis<DominatorTree>(llvm::Function *)</tt> to provide the 511 function to retrieve analysis result for, if the function pass does not require 512 any module or immutable passes. Note that this can only be done for functions for which the 513 analysis ran, e.g. in the case of dominators you should only ask for the 514 DominatorTree for function definitions, not declarations.</p> 515 516 <p>To write a correct <tt>ModulePass</tt> subclass, derive from 517 <tt>ModulePass</tt> and overload the <tt>runOnModule</tt> method with the 518 following signature:</p> 519 520 <!-- _______________________________________________________________________ --> 521 <h4> 522 <a name="runOnModule">The <tt>runOnModule</tt> method</a> 523 </h4> 524 525 <div> 526 527 <div class="doc_code"><pre> 528 <b>virtual bool</b> runOnModule(Module &M) = 0; 529 </pre></div> 530 531 <p>The <tt>runOnModule</tt> method performs the interesting work of the pass. 532 It should return true if the module was modified by the transformation and 533 false otherwise.</p> 534 535 </div> 536 537 </div> 538 539 <!-- ======================================================================= --> 540 <h3> 541 <a name="CallGraphSCCPass">The <tt>CallGraphSCCPass</tt> class</a> 542 </h3> 543 544 <div> 545 546 <p>The "<tt><a 547 href="http://llvm.org/doxygen/classllvm_1_1CallGraphSCCPass.html">CallGraphSCCPass</a></tt>" 548 is used by passes that need to traverse the program bottom-up on the call graph 549 (callees before callers). Deriving from CallGraphSCCPass provides some 550 mechanics for building and traversing the CallGraph, but also allows the system 551 to optimize execution of CallGraphSCCPass's. If your pass meets the 552 requirements outlined below, and doesn't meet the requirements of a <tt><a 553 href="#FunctionPass">FunctionPass</a></tt> or <tt><a 554 href="#BasicBlockPass">BasicBlockPass</a></tt>, you should derive from 555 <tt>CallGraphSCCPass</tt>.</p> 556 557 <p><b>TODO</b>: explain briefly what SCC, Tarjan's algo, and B-U mean.</p> 558 559 <p>To be explicit, <tt>CallGraphSCCPass</tt> subclasses are:</p> 560 561 <ol> 562 563 <li>... <em>not allowed</em> to inspect or modify any <tt>Function</tt>s other 564 than those in the current SCC and the direct callers and direct callees of the 565 SCC.</li> 566 567 <li>... <em>required</em> to preserve the current CallGraph object, updating it 568 to reflect any changes made to the program.</li> 569 570 <li>... <em>not allowed</em> to add or remove SCC's from the current Module, 571 though they may change the contents of an SCC.</li> 572 573 <li>... <em>allowed</em> to add or remove global variables from the current 574 Module.</li> 575 576 <li>... <em>allowed</em> to maintain state across invocations of 577 <a href="#runOnSCC"><tt>runOnSCC</tt></a> (including global data).</li> 578 </ol> 579 580 <p>Implementing a <tt>CallGraphSCCPass</tt> is slightly tricky in some cases 581 because it has to handle SCCs with more than one node in it. All of the virtual 582 methods described below should return true if they modified the program, or 583 false if they didn't.</p> 584 585 <!-- _______________________________________________________________________ --> 586 <h4> 587 <a name="doInitialization_scc"> 588 The <tt>doInitialization(CallGraph &)</tt> method 589 </a> 590 </h4> 591 592 <div> 593 594 <div class="doc_code"><pre> 595 <b>virtual bool</b> doInitialization(CallGraph &CG); 596 </pre></div> 597 598 <p>The <tt>doIninitialize</tt> method is allowed to do most of the things that 599 <tt>CallGraphSCCPass</tt>'s are not allowed to do. They can add and remove 600 functions, get pointers to functions, etc. The <tt>doInitialization</tt> method 601 is designed to do simple initialization type of stuff that does not depend on 602 the SCCs being processed. The <tt>doInitialization</tt> method call is not 603 scheduled to overlap with any other pass executions (thus it should be very 604 fast).</p> 605 606 </div> 607 608 <!-- _______________________________________________________________________ --> 609 <h4> 610 <a name="runOnSCC">The <tt>runOnSCC</tt> method</a> 611 </h4> 612 613 <div> 614 615 <div class="doc_code"><pre> 616 <b>virtual bool</b> runOnSCC(CallGraphSCC &SCC) = 0; 617 </pre></div> 618 619 <p>The <tt>runOnSCC</tt> method performs the interesting work of the pass, and 620 should return true if the module was modified by the transformation, false 621 otherwise.</p> 622 623 </div> 624 625 <!-- _______________________________________________________________________ --> 626 <h4> 627 <a name="doFinalization_scc"> 628 The <tt>doFinalization(CallGraph &)</tt> method 629 </a> 630 </h4> 631 632 <div> 633 634 <div class="doc_code"><pre> 635 <b>virtual bool</b> doFinalization(CallGraph &CG); 636 </pre></div> 637 638 <p>The <tt>doFinalization</tt> method is an infrequently used method that is 639 called when the pass framework has finished calling <a 640 href="#runOnFunction"><tt>runOnFunction</tt></a> for every function in the 641 program being compiled.</p> 642 643 </div> 644 645 </div> 646 647 <!-- ======================================================================= --> 648 <h3> 649 <a name="FunctionPass">The <tt>FunctionPass</tt> class</a> 650 </h3> 651 652 <div> 653 654 <p>In contrast to <tt>ModulePass</tt> subclasses, <tt><a 655 href="http://llvm.org/doxygen/classllvm_1_1Pass.html">FunctionPass</a></tt> 656 subclasses do have a predictable, local behavior that can be expected by the 657 system. All <tt>FunctionPass</tt> execute on each function in the program 658 independent of all of the other functions in the program. 659 <tt>FunctionPass</tt>'s do not require that they are executed in a particular 660 order, and <tt>FunctionPass</tt>'s do not modify external functions.</p> 661 662 <p>To be explicit, <tt>FunctionPass</tt> subclasses are not allowed to:</p> 663 664 <ol> 665 <li>Modify a Function other than the one currently being processed.</li> 666 <li>Add or remove Function's from the current Module.</li> 667 <li>Add or remove global variables from the current Module.</li> 668 <li>Maintain state across invocations of 669 <a href="#runOnFunction"><tt>runOnFunction</tt></a> (including global data)</li> 670 </ol> 671 672 <p>Implementing a <tt>FunctionPass</tt> is usually straightforward (See the <a 673 href="#basiccode">Hello World</a> pass for example). <tt>FunctionPass</tt>'s 674 may overload three virtual methods to do their work. All of these methods 675 should return true if they modified the program, or false if they didn't.</p> 676 677 <!-- _______________________________________________________________________ --> 678 <h4> 679 <a name="doInitialization_mod"> 680 The <tt>doInitialization(Module &)</tt> method 681 </a> 682 </h4> 683 684 <div> 685 686 <div class="doc_code"><pre> 687 <b>virtual bool</b> doInitialization(Module &M); 688 </pre></div> 689 690 <p>The <tt>doIninitialize</tt> method is allowed to do most of the things that 691 <tt>FunctionPass</tt>'s are not allowed to do. They can add and remove 692 functions, get pointers to functions, etc. The <tt>doInitialization</tt> method 693 is designed to do simple initialization type of stuff that does not depend on 694 the functions being processed. The <tt>doInitialization</tt> method call is not 695 scheduled to overlap with any other pass executions (thus it should be very 696 fast).</p> 697 698 <p>A good example of how this method should be used is the <a 699 href="http://llvm.org/doxygen/LowerAllocations_8cpp-source.html">LowerAllocations</a> 700 pass. This pass converts <tt>malloc</tt> and <tt>free</tt> instructions into 701 platform dependent <tt>malloc()</tt> and <tt>free()</tt> function calls. It 702 uses the <tt>doInitialization</tt> method to get a reference to the malloc and 703 free functions that it needs, adding prototypes to the module if necessary.</p> 704 705 </div> 706 707 <!-- _______________________________________________________________________ --> 708 <h4> 709 <a name="runOnFunction">The <tt>runOnFunction</tt> method</a> 710 </h4> 711 712 <div> 713 714 <div class="doc_code"><pre> 715 <b>virtual bool</b> runOnFunction(Function &F) = 0; 716 </pre></div><p> 717 718 <p>The <tt>runOnFunction</tt> method must be implemented by your subclass to do 719 the transformation or analysis work of your pass. As usual, a true value should 720 be returned if the function is modified.</p> 721 722 </div> 723 724 <!-- _______________________________________________________________________ --> 725 <h4> 726 <a name="doFinalization_mod"> 727 The <tt>doFinalization(Module &)</tt> method 728 </a> 729 </h4> 730 731 <div> 732 733 <div class="doc_code"><pre> 734 <b>virtual bool</b> doFinalization(Module &M); 735 </pre></div> 736 737 <p>The <tt>doFinalization</tt> method is an infrequently used method that is 738 called when the pass framework has finished calling <a 739 href="#runOnFunction"><tt>runOnFunction</tt></a> for every function in the 740 program being compiled.</p> 741 742 </div> 743 744 </div> 745 746 <!-- ======================================================================= --> 747 <h3> 748 <a name="LoopPass">The <tt>LoopPass</tt> class </a> 749 </h3> 750 751 <div> 752 753 <p> All <tt>LoopPass</tt> execute on each loop in the function independent of 754 all of the other loops in the function. <tt>LoopPass</tt> processes loops in 755 loop nest order such that outer most loop is processed last. </p> 756 757 <p> <tt>LoopPass</tt> subclasses are allowed to update loop nest using 758 <tt>LPPassManager</tt> interface. Implementing a loop pass is usually 759 straightforward. <tt>LoopPass</tt>'s may overload three virtual methods to 760 do their work. All these methods should return true if they modified the 761 program, or false if they didn't. </p> 762 763 <!-- _______________________________________________________________________ --> 764 <h4> 765 <a name="doInitialization_loop"> 766 The <tt>doInitialization(Loop *,LPPassManager &)</tt> method 767 </a> 768 </h4> 769 770 <div> 771 772 <div class="doc_code"><pre> 773 <b>virtual bool</b> doInitialization(Loop *, LPPassManager &LPM); 774 </pre></div> 775 776 <p>The <tt>doInitialization</tt> method is designed to do simple initialization 777 type of stuff that does not depend on the functions being processed. The 778 <tt>doInitialization</tt> method call is not scheduled to overlap with any 779 other pass executions (thus it should be very fast). LPPassManager 780 interface should be used to access Function or Module level analysis 781 information.</p> 782 783 </div> 784 785 786 <!-- _______________________________________________________________________ --> 787 <h4> 788 <a name="runOnLoop">The <tt>runOnLoop</tt> method</a> 789 </h4> 790 791 <div> 792 793 <div class="doc_code"><pre> 794 <b>virtual bool</b> runOnLoop(Loop *, LPPassManager &LPM) = 0; 795 </pre></div><p> 796 797 <p>The <tt>runOnLoop</tt> method must be implemented by your subclass to do 798 the transformation or analysis work of your pass. As usual, a true value should 799 be returned if the function is modified. <tt>LPPassManager</tt> interface 800 should be used to update loop nest.</p> 801 802 </div> 803 804 <!-- _______________________________________________________________________ --> 805 <h4> 806 <a name="doFinalization_loop">The <tt>doFinalization()</tt> method</a> 807 </h4> 808 809 <div> 810 811 <div class="doc_code"><pre> 812 <b>virtual bool</b> doFinalization(); 813 </pre></div> 814 815 <p>The <tt>doFinalization</tt> method is an infrequently used method that is 816 called when the pass framework has finished calling <a 817 href="#runOnLoop"><tt>runOnLoop</tt></a> for every loop in the 818 program being compiled. </p> 819 820 </div> 821 822 </div> 823 824 <!-- ======================================================================= --> 825 <h3> 826 <a name="RegionPass">The <tt>RegionPass</tt> class </a> 827 </h3> 828 829 <div> 830 831 <p> <tt>RegionPass</tt> is similar to <a href="#LoopPass"><tt>LoopPass</tt></a>, 832 but executes on each single entry single exit region in the function. 833 <tt>RegionPass</tt> processes regions in nested order such that the outer most 834 region is processed last. </p> 835 836 <p> <tt>RegionPass</tt> subclasses are allowed to update the region tree by using 837 the <tt>RGPassManager</tt> interface. You may overload three virtual methods of 838 <tt>RegionPass</tt> to implement your own region pass. All these 839 methods should return true if they modified the program, or false if they didn not. 840 </p> 841 842 <!-- _______________________________________________________________________ --> 843 <h4> 844 <a name="doInitialization_region"> 845 The <tt>doInitialization(Region *, RGPassManager &)</tt> method 846 </a> 847 </h4> 848 849 <div> 850 851 <div class="doc_code"><pre> 852 <b>virtual bool</b> doInitialization(Region *, RGPassManager &RGM); 853 </pre></div> 854 855 <p>The <tt>doInitialization</tt> method is designed to do simple initialization 856 type of stuff that does not depend on the functions being processed. The 857 <tt>doInitialization</tt> method call is not scheduled to overlap with any 858 other pass executions (thus it should be very fast). RPPassManager 859 interface should be used to access Function or Module level analysis 860 information.</p> 861 862 </div> 863 864 865 <!-- _______________________________________________________________________ --> 866 <h4> 867 <a name="runOnRegion">The <tt>runOnRegion</tt> method</a> 868 </h4> 869 870 <div> 871 872 <div class="doc_code"><pre> 873 <b>virtual bool</b> runOnRegion(Region *, RGPassManager &RGM) = 0; 874 </pre></div><p> 875 876 <p>The <tt>runOnRegion</tt> method must be implemented by your subclass to do 877 the transformation or analysis work of your pass. As usual, a true value should 878 be returned if the region is modified. <tt>RGPassManager</tt> interface 879 should be used to update region tree.</p> 880 881 </div> 882 883 <!-- _______________________________________________________________________ --> 884 <h4> 885 <a name="doFinalization_region">The <tt>doFinalization()</tt> method</a> 886 </h4> 887 888 <div> 889 890 <div class="doc_code"><pre> 891 <b>virtual bool</b> doFinalization(); 892 </pre></div> 893 894 <p>The <tt>doFinalization</tt> method is an infrequently used method that is 895 called when the pass framework has finished calling <a 896 href="#runOnRegion"><tt>runOnRegion</tt></a> for every region in the 897 program being compiled. </p> 898 899 </div> 900 901 </div> 902 903 <!-- ======================================================================= --> 904 <h3> 905 <a name="BasicBlockPass">The <tt>BasicBlockPass</tt> class</a> 906 </h3> 907 908 <div> 909 910 <p><tt>BasicBlockPass</tt>'s are just like <a 911 href="#FunctionPass"><tt>FunctionPass</tt></a>'s, except that they must limit 912 their scope of inspection and modification to a single basic block at a time. 913 As such, they are <b>not</b> allowed to do any of the following:</p> 914 915 <ol> 916 <li>Modify or inspect any basic blocks outside of the current one</li> 917 <li>Maintain state across invocations of 918 <a href="#runOnBasicBlock"><tt>runOnBasicBlock</tt></a></li> 919 <li>Modify the control flow graph (by altering terminator instructions)</li> 920 <li>Any of the things forbidden for 921 <a href="#FunctionPass"><tt>FunctionPass</tt></a>es.</li> 922 </ol> 923 924 <p><tt>BasicBlockPass</tt>es are useful for traditional local and "peephole" 925 optimizations. They may override the same <a 926 href="#doInitialization_mod"><tt>doInitialization(Module &)</tt></a> and <a 927 href="#doFinalization_mod"><tt>doFinalization(Module &)</tt></a> methods that <a 928 href="#FunctionPass"><tt>FunctionPass</tt></a>'s have, but also have the following virtual methods that may also be implemented:</p> 929 930 <!-- _______________________________________________________________________ --> 931 <h4> 932 <a name="doInitialization_fn"> 933 The <tt>doInitialization(Function &)</tt> method 934 </a> 935 </h4> 936 937 <div> 938 939 <div class="doc_code"><pre> 940 <b>virtual bool</b> doInitialization(Function &F); 941 </pre></div> 942 943 <p>The <tt>doIninitialize</tt> method is allowed to do most of the things that 944 <tt>BasicBlockPass</tt>'s are not allowed to do, but that 945 <tt>FunctionPass</tt>'s can. The <tt>doInitialization</tt> method is designed 946 to do simple initialization that does not depend on the 947 BasicBlocks being processed. The <tt>doInitialization</tt> method call is not 948 scheduled to overlap with any other pass executions (thus it should be very 949 fast).</p> 950 951 </div> 952 953 <!-- _______________________________________________________________________ --> 954 <h4> 955 <a name="runOnBasicBlock">The <tt>runOnBasicBlock</tt> method</a> 956 </h4> 957 958 <div> 959 960 <div class="doc_code"><pre> 961 <b>virtual bool</b> runOnBasicBlock(BasicBlock &BB) = 0; 962 </pre></div> 963 964 <p>Override this function to do the work of the <tt>BasicBlockPass</tt>. This 965 function is not allowed to inspect or modify basic blocks other than the 966 parameter, and are not allowed to modify the CFG. A true value must be returned 967 if the basic block is modified.</p> 968 969 </div> 970 971 <!-- _______________________________________________________________________ --> 972 <h4> 973 <a name="doFinalization_fn"> 974 The <tt>doFinalization(Function &)</tt> method 975 </a> 976 </h4> 977 978 <div> 979 980 <div class="doc_code"><pre> 981 <b>virtual bool</b> doFinalization(Function &F); 982 </pre></div> 983 984 <p>The <tt>doFinalization</tt> method is an infrequently used method that is 985 called when the pass framework has finished calling <a 986 href="#runOnBasicBlock"><tt>runOnBasicBlock</tt></a> for every BasicBlock in the 987 program being compiled. This can be used to perform per-function 988 finalization.</p> 989 990 </div> 991 992 </div> 993 994 <!-- ======================================================================= --> 995 <h3> 996 <a name="MachineFunctionPass">The <tt>MachineFunctionPass</tt> class</a> 997 </h3> 998 999 <div> 1000 1001 <p>A <tt>MachineFunctionPass</tt> is a part of the LLVM code generator that 1002 executes on the machine-dependent representation of each LLVM function in the 1003 program.</p> 1004 1005 <p>Code generator passes are registered and initialized specially by 1006 <tt>TargetMachine::addPassesToEmitFile</tt> and similar routines, so they 1007 cannot generally be run from the <tt>opt</tt> or <tt>bugpoint</tt> 1008 commands.</p> 1009 1010 <p>A <tt>MachineFunctionPass</tt> is also a <tt>FunctionPass</tt>, so all 1011 the restrictions that apply to a <tt>FunctionPass</tt> also apply to it. 1012 <tt>MachineFunctionPass</tt>es also have additional restrictions. In particular, 1013 <tt>MachineFunctionPass</tt>es are not allowed to do any of the following:</p> 1014 1015 <ol> 1016 <li>Modify or create any LLVM IR Instructions, BasicBlocks, Arguments, 1017 Functions, GlobalVariables, GlobalAliases, or Modules.</li> 1018 <li>Modify a MachineFunction other than the one currently being processed.</li> 1019 <li>Maintain state across invocations of <a 1020 href="#runOnMachineFunction"><tt>runOnMachineFunction</tt></a> (including global 1021 data)</li> 1022 </ol> 1023 1024 <!-- _______________________________________________________________________ --> 1025 <h4> 1026 <a name="runOnMachineFunction"> 1027 The <tt>runOnMachineFunction(MachineFunction &MF)</tt> method 1028 </a> 1029 </h4> 1030 1031 <div> 1032 1033 <div class="doc_code"><pre> 1034 <b>virtual bool</b> runOnMachineFunction(MachineFunction &MF) = 0; 1035 </pre></div> 1036 1037 <p><tt>runOnMachineFunction</tt> can be considered the main entry point of a 1038 <tt>MachineFunctionPass</tt>; that is, you should override this method to do the 1039 work of your <tt>MachineFunctionPass</tt>.</p> 1040 1041 <p>The <tt>runOnMachineFunction</tt> method is called on every 1042 <tt>MachineFunction</tt> in a <tt>Module</tt>, so that the 1043 <tt>MachineFunctionPass</tt> may perform optimizations on the machine-dependent 1044 representation of the function. If you want to get at the LLVM <tt>Function</tt> 1045 for the <tt>MachineFunction</tt> you're working on, use 1046 <tt>MachineFunction</tt>'s <tt>getFunction()</tt> accessor method -- but 1047 remember, you may not modify the LLVM <tt>Function</tt> or its contents from a 1048 <tt>MachineFunctionPass</tt>.</p> 1049 1050 </div> 1051 1052 </div> 1053 1054 </div> 1055 1056 <!-- *********************************************************************** --> 1057 <h2> 1058 <a name="registration">Pass registration</a> 1059 </h2> 1060 <!-- *********************************************************************** --> 1061 1062 <div> 1063 1064 <p>In the <a href="#basiccode">Hello World</a> example pass we illustrated how 1065 pass registration works, and discussed some of the reasons that it is used and 1066 what it does. Here we discuss how and why passes are registered.</p> 1067 1068 <p>As we saw above, passes are registered with the <b><tt>RegisterPass</tt></b> 1069 template. The template parameter is the name of the pass that is to be used on 1070 the command line to specify that the pass should be added to a program (for 1071 example, with <tt>opt</tt> or <tt>bugpoint</tt>). The first argument is the 1072 name of the pass, which is to be used for the <tt>-help</tt> output of 1073 programs, as 1074 well as for debug output generated by the <tt>--debug-pass</tt> option.</p> 1075 1076 <p>If you want your pass to be easily dumpable, you should 1077 implement the virtual <tt>print</tt> method:</p> 1078 1079 <!-- _______________________________________________________________________ --> 1080 <h4> 1081 <a name="print">The <tt>print</tt> method</a> 1082 </h4> 1083 1084 <div> 1085 1086 <div class="doc_code"><pre> 1087 <b>virtual void</b> print(std::ostream &O, <b>const</b> Module *M) <b>const</b>; 1088 </pre></div> 1089 1090 <p>The <tt>print</tt> method must be implemented by "analyses" in order to print 1091 a human readable version of the analysis results. This is useful for debugging 1092 an analysis itself, as well as for other people to figure out how an analysis 1093 works. Use the <tt>opt -analyze</tt> argument to invoke this method.</p> 1094 1095 <p>The <tt>llvm::OStream</tt> parameter specifies the stream to write the results on, 1096 and the <tt>Module</tt> parameter gives a pointer to the top level module of the 1097 program that has been analyzed. Note however that this pointer may be null in 1098 certain circumstances (such as calling the <tt>Pass::dump()</tt> from a 1099 debugger), so it should only be used to enhance debug output, it should not be 1100 depended on.</p> 1101 1102 </div> 1103 1104 </div> 1105 1106 <!-- *********************************************************************** --> 1107 <h2> 1108 <a name="interaction">Specifying interactions between passes</a> 1109 </h2> 1110 <!-- *********************************************************************** --> 1111 1112 <div> 1113 1114 <p>One of the main responsibilities of the <tt>PassManager</tt> is to make sure 1115 that passes interact with each other correctly. Because <tt>PassManager</tt> 1116 tries to <a href="#passmanager">optimize the execution of passes</a> it must 1117 know how the passes interact with each other and what dependencies exist between 1118 the various passes. To track this, each pass can declare the set of passes that 1119 are required to be executed before the current pass, and the passes which are 1120 invalidated by the current pass.</p> 1121 1122 <p>Typically this functionality is used to require that analysis results are 1123 computed before your pass is run. Running arbitrary transformation passes can 1124 invalidate the computed analysis results, which is what the invalidation set 1125 specifies. If a pass does not implement the <tt><a 1126 href="#getAnalysisUsage">getAnalysisUsage</a></tt> method, it defaults to not 1127 having any prerequisite passes, and invalidating <b>all</b> other passes.</p> 1128 1129 <!-- _______________________________________________________________________ --> 1130 <h4> 1131 <a name="getAnalysisUsage">The <tt>getAnalysisUsage</tt> method</a> 1132 </h4> 1133 1134 <div> 1135 1136 <div class="doc_code"><pre> 1137 <b>virtual void</b> getAnalysisUsage(AnalysisUsage &Info) <b>const</b>; 1138 </pre></div> 1139 1140 <p>By implementing the <tt>getAnalysisUsage</tt> method, the required and 1141 invalidated sets may be specified for your transformation. The implementation 1142 should fill in the <tt><a 1143 href="http://llvm.org/doxygen/classllvm_1_1AnalysisUsage.html">AnalysisUsage</a></tt> 1144 object with information about which passes are required and not invalidated. To 1145 do this, a pass may call any of the following methods on the AnalysisUsage 1146 object:</p> 1147 </div> 1148 1149 <!-- _______________________________________________________________________ --> 1150 <h4> 1151 <a name="AU::addRequired"> 1152 The <tt>AnalysisUsage::addRequired<></tt> 1153 and <tt>AnalysisUsage::addRequiredTransitive<></tt> methods 1154 </a> 1155 </h4> 1156 1157 <div> 1158 <p> 1159 If your pass requires a previous pass to be executed (an analysis for example), 1160 it can use one of these methods to arrange for it to be run before your pass. 1161 LLVM has many different types of analyses and passes that can be required, 1162 spanning the range from <tt>DominatorSet</tt> to <tt>BreakCriticalEdges</tt>. 1163 Requiring <tt>BreakCriticalEdges</tt>, for example, guarantees that there will 1164 be no critical edges in the CFG when your pass has been run. 1165 </p> 1166 1167 <p> 1168 Some analyses chain to other analyses to do their job. For example, an <a 1169 href="AliasAnalysis.html">AliasAnalysis</a> implementation is required to <a 1170 href="AliasAnalysis.html#chaining">chain</a> to other alias analysis passes. In 1171 cases where analyses chain, the <tt>addRequiredTransitive</tt> method should be 1172 used instead of the <tt>addRequired</tt> method. This informs the PassManager 1173 that the transitively required pass should be alive as long as the requiring 1174 pass is. 1175 </p> 1176 </div> 1177 1178 <!-- _______________________________________________________________________ --> 1179 <h4> 1180 <a name="AU::addPreserved"> 1181 The <tt>AnalysisUsage::addPreserved<></tt> method 1182 </a> 1183 </h4> 1184 1185 <div> 1186 <p> 1187 One of the jobs of the PassManager is to optimize how and when analyses are run. 1188 In particular, it attempts to avoid recomputing data unless it needs to. For 1189 this reason, passes are allowed to declare that they preserve (i.e., they don't 1190 invalidate) an existing analysis if it's available. For example, a simple 1191 constant folding pass would not modify the CFG, so it can't possibly affect the 1192 results of dominator analysis. By default, all passes are assumed to invalidate 1193 all others. 1194 </p> 1195 1196 <p> 1197 The <tt>AnalysisUsage</tt> class provides several methods which are useful in 1198 certain circumstances that are related to <tt>addPreserved</tt>. In particular, 1199 the <tt>setPreservesAll</tt> method can be called to indicate that the pass does 1200 not modify the LLVM program at all (which is true for analyses), and the 1201 <tt>setPreservesCFG</tt> method can be used by transformations that change 1202 instructions in the program but do not modify the CFG or terminator instructions 1203 (note that this property is implicitly set for <a 1204 href="#BasicBlockPass">BasicBlockPass</a>'s). 1205 </p> 1206 1207 <p> 1208 <tt>addPreserved</tt> is particularly useful for transformations like 1209 <tt>BreakCriticalEdges</tt>. This pass knows how to update a small set of loop 1210 and dominator related analyses if they exist, so it can preserve them, despite 1211 the fact that it hacks on the CFG. 1212 </p> 1213 </div> 1214 1215 <!-- _______________________________________________________________________ --> 1216 <h4> 1217 <a name="AU::examples"> 1218 Example implementations of <tt>getAnalysisUsage</tt> 1219 </a> 1220 </h4> 1221 1222 <div> 1223 1224 <div class="doc_code"><pre> 1225 <i>// This example modifies the program, but does not modify the CFG</i> 1226 <b>void</b> <a href="http://llvm.org/doxygen/structLICM.html">LICM</a>::getAnalysisUsage(AnalysisUsage &AU) <b>const</b> { 1227 AU.setPreservesCFG(); 1228 AU.addRequired<<a href="http://llvm.org/doxygen/classllvm_1_1LoopInfo.html">LoopInfo</a>>(); 1229 } 1230 </pre></div> 1231 1232 </div> 1233 1234 <!-- _______________________________________________________________________ --> 1235 <h4> 1236 <a name="getAnalysis"> 1237 The <tt>getAnalysis<></tt> and 1238 <tt>getAnalysisIfAvailable<></tt> methods 1239 </a> 1240 </h4> 1241 1242 <div> 1243 1244 <p>The <tt>Pass::getAnalysis<></tt> method is automatically inherited by 1245 your class, providing you with access to the passes that you declared that you 1246 required with the <a href="#getAnalysisUsage"><tt>getAnalysisUsage</tt></a> 1247 method. It takes a single template argument that specifies which pass class you 1248 want, and returns a reference to that pass. For example:</p> 1249 1250 <div class="doc_code"><pre> 1251 bool LICM::runOnFunction(Function &F) { 1252 LoopInfo &LI = getAnalysis<LoopInfo>(); 1253 ... 1254 } 1255 </pre></div> 1256 1257 <p>This method call returns a reference to the pass desired. You may get a 1258 runtime assertion failure if you attempt to get an analysis that you did not 1259 declare as required in your <a 1260 href="#getAnalysisUsage"><tt>getAnalysisUsage</tt></a> implementation. This 1261 method can be called by your <tt>run*</tt> method implementation, or by any 1262 other local method invoked by your <tt>run*</tt> method. 1263 1264 A module level pass can use function level analysis info using this interface. 1265 For example:</p> 1266 1267 <div class="doc_code"><pre> 1268 bool ModuleLevelPass::runOnModule(Module &M) { 1269 ... 1270 DominatorTree &DT = getAnalysis<DominatorTree>(Func); 1271 ... 1272 } 1273 </pre></div> 1274 1275 <p>In above example, runOnFunction for DominatorTree is called by pass manager 1276 before returning a reference to the desired pass.</p> 1277 1278 <p> 1279 If your pass is capable of updating analyses if they exist (e.g., 1280 <tt>BreakCriticalEdges</tt>, as described above), you can use the 1281 <tt>getAnalysisIfAvailable</tt> method, which returns a pointer to the analysis 1282 if it is active. For example:</p> 1283 1284 <div class="doc_code"><pre> 1285 ... 1286 if (DominatorSet *DS = getAnalysisIfAvailable<DominatorSet>()) { 1287 <i>// A DominatorSet is active. This code will update it.</i> 1288 } 1289 ... 1290 </pre></div> 1291 1292 </div> 1293 1294 </div> 1295 1296 <!-- *********************************************************************** --> 1297 <h2> 1298 <a name="analysisgroup">Implementing Analysis Groups</a> 1299 </h2> 1300 <!-- *********************************************************************** --> 1301 1302 <div> 1303 1304 <p>Now that we understand the basics of how passes are defined, how they are 1305 used, and how they are required from other passes, it's time to get a little bit 1306 fancier. All of the pass relationships that we have seen so far are very 1307 simple: one pass depends on one other specific pass to be run before it can run. 1308 For many applications, this is great, for others, more flexibility is 1309 required.</p> 1310 1311 <p>In particular, some analyses are defined such that there is a single simple 1312 interface to the analysis results, but multiple ways of calculating them. 1313 Consider alias analysis for example. The most trivial alias analysis returns 1314 "may alias" for any alias query. The most sophisticated analysis a 1315 flow-sensitive, context-sensitive interprocedural analysis that can take a 1316 significant amount of time to execute (and obviously, there is a lot of room 1317 between these two extremes for other implementations). To cleanly support 1318 situations like this, the LLVM Pass Infrastructure supports the notion of 1319 Analysis Groups.</p> 1320 1321 <!-- _______________________________________________________________________ --> 1322 <h4> 1323 <a name="agconcepts">Analysis Group Concepts</a> 1324 </h4> 1325 1326 <div> 1327 1328 <p>An Analysis Group is a single simple interface that may be implemented by 1329 multiple different passes. Analysis Groups can be given human readable names 1330 just like passes, but unlike passes, they need not derive from the <tt>Pass</tt> 1331 class. An analysis group may have one or more implementations, one of which is 1332 the "default" implementation.</p> 1333 1334 <p>Analysis groups are used by client passes just like other passes are: the 1335 <tt>AnalysisUsage::addRequired()</tt> and <tt>Pass::getAnalysis()</tt> methods. 1336 In order to resolve this requirement, the <a href="#passmanager">PassManager</a> 1337 scans the available passes to see if any implementations of the analysis group 1338 are available. If none is available, the default implementation is created for 1339 the pass to use. All standard rules for <A href="#interaction">interaction 1340 between passes</a> still apply.</p> 1341 1342 <p>Although <a href="#registration">Pass Registration</a> is optional for normal 1343 passes, all analysis group implementations must be registered, and must use the 1344 <A href="#registerag"><tt>INITIALIZE_AG_PASS</tt></a> template to join the 1345 implementation pool. Also, a default implementation of the interface 1346 <b>must</b> be registered with <A 1347 href="#registerag"><tt>RegisterAnalysisGroup</tt></a>.</p> 1348 1349 <p>As a concrete example of an Analysis Group in action, consider the <a 1350 href="http://llvm.org/doxygen/classllvm_1_1AliasAnalysis.html">AliasAnalysis</a> 1351 analysis group. The default implementation of the alias analysis interface (the 1352 <tt><a 1353 href="http://llvm.org/doxygen/structBasicAliasAnalysis.html">basicaa</a></tt> 1354 pass) just does a few simple checks that don't require significant analysis to 1355 compute (such as: two different globals can never alias each other, etc). 1356 Passes that use the <tt><a 1357 href="http://llvm.org/doxygen/classllvm_1_1AliasAnalysis.html">AliasAnalysis</a></tt> 1358 interface (for example the <tt><a 1359 href="http://llvm.org/doxygen/structGCSE.html">gcse</a></tt> pass), do 1360 not care which implementation of alias analysis is actually provided, they just 1361 use the designated interface.</p> 1362 1363 <p>From the user's perspective, commands work just like normal. Issuing the 1364 command '<tt>opt -gcse ...</tt>' will cause the <tt>basicaa</tt> class to be 1365 instantiated and added to the pass sequence. Issuing the command '<tt>opt 1366 -somefancyaa -gcse ...</tt>' will cause the <tt>gcse</tt> pass to use the 1367 <tt>somefancyaa</tt> alias analysis (which doesn't actually exist, it's just a 1368 hypothetical example) instead.</p> 1369 1370 </div> 1371 1372 <!-- _______________________________________________________________________ --> 1373 <h4> 1374 <a name="registerag">Using <tt>RegisterAnalysisGroup</tt></a> 1375 </h4> 1376 1377 <div> 1378 1379 <p>The <tt>RegisterAnalysisGroup</tt> template is used to register the analysis 1380 group itself, while the <tt>INITIALIZE_AG_PASS</tt> is used to add pass 1381 implementations to the analysis group. First, 1382 an analysis group should be registered, with a human readable name 1383 provided for it. 1384 Unlike registration of passes, there is no command line argument to be specified 1385 for the Analysis Group Interface itself, because it is "abstract":</p> 1386 1387 <div class="doc_code"><pre> 1388 <b>static</b> RegisterAnalysisGroup<<a href="http://llvm.org/doxygen/classllvm_1_1AliasAnalysis.html">AliasAnalysis</a>> A("<i>Alias Analysis</i>"); 1389 </pre></div> 1390 1391 <p>Once the analysis is registered, passes can declare that they are valid 1392 implementations of the interface by using the following code:</p> 1393 1394 <div class="doc_code"><pre> 1395 <b>namespace</b> { 1396 //<i> Declare that we implement the AliasAnalysis interface</i> 1397 INITIALIZE_AG_PASS(FancyAA, <a href="http://llvm.org/doxygen/classllvm_1_1AliasAnalysis.html">AliasAnalysis</a>, "<i>somefancyaa</i>", 1398 "<i>A more complex alias analysis implementation</i>", 1399 false, // <i>Is CFG Only?</i> 1400 true, // <i>Is Analysis?</i> 1401 false, // <i>Is default Analysis Group implementation?</i> 1402 ); 1403 } 1404 </pre></div> 1405 1406 <p>This just shows a class <tt>FancyAA</tt> that 1407 uses the <tt>INITIALIZE_AG_PASS</tt> macro both to register and 1408 to "join" the <tt><a href="http://llvm.org/doxygen/classllvm_1_1AliasAnalysis.html">AliasAnalysis</a></tt> 1409 analysis group. Every implementation of an analysis group should join using 1410 this macro.</p> 1411 1412 <div class="doc_code"><pre> 1413 <b>namespace</b> { 1414 //<i> Declare that we implement the AliasAnalysis interface</i> 1415 INITIALIZE_AG_PASS(BasicAA, <a href="http://llvm.org/doxygen/classllvm_1_1AliasAnalysis.html">AliasAnalysis</a>, "<i>basicaa</i>", 1416 "<i>Basic Alias Analysis (default AA impl)</i>", 1417 false, // <i>Is CFG Only?</i> 1418 true, // <i>Is Analysis?</i> 1419 true, // <i>Is default Analysis Group implementation?</i> 1420 ); 1421 } 1422 </pre></div> 1423 1424 <p>Here we show how the default implementation is specified (using the final 1425 argument to the <tt>INITIALIZE_AG_PASS</tt> template). There must be exactly 1426 one default implementation available at all times for an Analysis Group to be 1427 used. Only default implementation can derive from <tt>ImmutablePass</tt>. 1428 Here we declare that the 1429 <tt><a href="http://llvm.org/doxygen/structBasicAliasAnalysis.html">BasicAliasAnalysis</a></tt> 1430 pass is the default implementation for the interface.</p> 1431 1432 </div> 1433 1434 </div> 1435 1436 <!-- *********************************************************************** --> 1437 <h2> 1438 <a name="passStatistics">Pass Statistics</a> 1439 </h2> 1440 <!-- *********************************************************************** --> 1441 1442 <div> 1443 <p>The <a 1444 href="http://llvm.org/doxygen/Statistic_8h-source.html"><tt>Statistic</tt></a> 1445 class is designed to be an easy way to expose various success 1446 metrics from passes. These statistics are printed at the end of a 1447 run, when the -stats command line option is enabled on the command 1448 line. See the <a href="http://llvm.org/docs/ProgrammersManual.html#Statistic">Statistics section</a> in the Programmer's Manual for details. 1449 1450 </div> 1451 1452 1453 <!-- *********************************************************************** --> 1454 <h2> 1455 <a name="passmanager">What PassManager does</a> 1456 </h2> 1457 <!-- *********************************************************************** --> 1458 1459 <div> 1460 1461 <p>The <a 1462 href="http://llvm.org/doxygen/PassManager_8h-source.html"><tt>PassManager</tt></a> 1463 <a 1464 href="http://llvm.org/doxygen/classllvm_1_1PassManager.html">class</a> 1465 takes a list of passes, ensures their <a href="#interaction">prerequisites</a> 1466 are set up correctly, and then schedules passes to run efficiently. All of the 1467 LLVM tools that run passes use the <tt>PassManager</tt> for execution of these 1468 passes.</p> 1469 1470 <p>The <tt>PassManager</tt> does two main things to try to reduce the execution 1471 time of a series of passes:</p> 1472 1473 <ol> 1474 <li><b>Share analysis results</b> - The PassManager attempts to avoid 1475 recomputing analysis results as much as possible. This means keeping track of 1476 which analyses are available already, which analyses get invalidated, and which 1477 analyses are needed to be run for a pass. An important part of work is that the 1478 <tt>PassManager</tt> tracks the exact lifetime of all analysis results, allowing 1479 it to <a href="#releaseMemory">free memory</a> allocated to holding analysis 1480 results as soon as they are no longer needed.</li> 1481 1482 <li><b>Pipeline the execution of passes on the program</b> - The 1483 <tt>PassManager</tt> attempts to get better cache and memory usage behavior out 1484 of a series of passes by pipelining the passes together. This means that, given 1485 a series of consecutive <a href="#FunctionPass"><tt>FunctionPass</tt></a>'s, it 1486 will execute all of the <a href="#FunctionPass"><tt>FunctionPass</tt></a>'s on 1487 the first function, then all of the <a 1488 href="#FunctionPass"><tt>FunctionPass</tt></a>es on the second function, 1489 etc... until the entire program has been run through the passes. 1490 1491 <p>This improves the cache behavior of the compiler, because it is only touching 1492 the LLVM program representation for a single function at a time, instead of 1493 traversing the entire program. It reduces the memory consumption of compiler, 1494 because, for example, only one <a 1495 href="http://llvm.org/doxygen/classllvm_1_1DominatorSet.html"><tt>DominatorSet</tt></a> 1496 needs to be calculated at a time. This also makes it possible to implement 1497 some <a 1498 href="#SMP">interesting enhancements</a> in the future.</p></li> 1499 1500 </ol> 1501 1502 <p>The effectiveness of the <tt>PassManager</tt> is influenced directly by how 1503 much information it has about the behaviors of the passes it is scheduling. For 1504 example, the "preserved" set is intentionally conservative in the face of an 1505 unimplemented <a href="#getAnalysisUsage"><tt>getAnalysisUsage</tt></a> method. 1506 Not implementing when it should be implemented will have the effect of not 1507 allowing any analysis results to live across the execution of your pass.</p> 1508 1509 <p>The <tt>PassManager</tt> class exposes a <tt>--debug-pass</tt> command line 1510 options that is useful for debugging pass execution, seeing how things work, and 1511 diagnosing when you should be preserving more analyses than you currently are 1512 (To get information about all of the variants of the <tt>--debug-pass</tt> 1513 option, just type '<tt>opt -help-hidden</tt>').</p> 1514 1515 <p>By using the <tt>--debug-pass=Structure</tt> option, for example, we can see 1516 how our <a href="#basiccode">Hello World</a> pass interacts with other passes. 1517 Lets try it out with the <tt>gcse</tt> and <tt>licm</tt> passes:</p> 1518 1519 <div class="doc_code"><pre> 1520 $ opt -load ../../../Debug+Asserts/lib/Hello.so -gcse -licm --debug-pass=Structure < hello.bc > /dev/null 1521 Module Pass Manager 1522 Function Pass Manager 1523 Dominator Set Construction 1524 Immediate Dominators Construction 1525 Global Common Subexpression Elimination 1526 -- Immediate Dominators Construction 1527 -- Global Common Subexpression Elimination 1528 Natural Loop Construction 1529 Loop Invariant Code Motion 1530 -- Natural Loop Construction 1531 -- Loop Invariant Code Motion 1532 Module Verifier 1533 -- Dominator Set Construction 1534 -- Module Verifier 1535 Bitcode Writer 1536 --Bitcode Writer 1537 </pre></div> 1538 1539 <p>This output shows us when passes are constructed and when the analysis 1540 results are known to be dead (prefixed with '<tt>--</tt>'). Here we see that 1541 GCSE uses dominator and immediate dominator information to do its job. The LICM 1542 pass uses natural loop information, which uses dominator sets, but not immediate 1543 dominators. Because immediate dominators are no longer useful after the GCSE 1544 pass, it is immediately destroyed. The dominator sets are then reused to 1545 compute natural loop information, which is then used by the LICM pass.</p> 1546 1547 <p>After the LICM pass, the module verifier runs (which is automatically added 1548 by the '<tt>opt</tt>' tool), which uses the dominator set to check that the 1549 resultant LLVM code is well formed. After it finishes, the dominator set 1550 information is destroyed, after being computed once, and shared by three 1551 passes.</p> 1552 1553 <p>Lets see how this changes when we run the <a href="#basiccode">Hello 1554 World</a> pass in between the two passes:</p> 1555 1556 <div class="doc_code"><pre> 1557 $ opt -load ../../../Debug+Asserts/lib/Hello.so -gcse -hello -licm --debug-pass=Structure < hello.bc > /dev/null 1558 Module Pass Manager 1559 Function Pass Manager 1560 Dominator Set Construction 1561 Immediate Dominators Construction 1562 Global Common Subexpression Elimination 1563 <b>-- Dominator Set Construction</b> 1564 -- Immediate Dominators Construction 1565 -- Global Common Subexpression Elimination 1566 <b> Hello World Pass 1567 -- Hello World Pass 1568 Dominator Set Construction</b> 1569 Natural Loop Construction 1570 Loop Invariant Code Motion 1571 -- Natural Loop Construction 1572 -- Loop Invariant Code Motion 1573 Module Verifier 1574 -- Dominator Set Construction 1575 -- Module Verifier 1576 Bitcode Writer 1577 --Bitcode Writer 1578 Hello: __main 1579 Hello: puts 1580 Hello: main 1581 </pre></div> 1582 1583 <p>Here we see that the <a href="#basiccode">Hello World</a> pass has killed the 1584 Dominator Set pass, even though it doesn't modify the code at all! To fix this, 1585 we need to add the following <a 1586 href="#getAnalysisUsage"><tt>getAnalysisUsage</tt></a> method to our pass:</p> 1587 1588 <div class="doc_code"><pre> 1589 <i>// We don't modify the program, so we preserve all analyses</i> 1590 <b>virtual void</b> getAnalysisUsage(AnalysisUsage &AU) <b>const</b> { 1591 AU.setPreservesAll(); 1592 } 1593 </pre></div> 1594 1595 <p>Now when we run our pass, we get this output:</p> 1596 1597 <div class="doc_code"><pre> 1598 $ opt -load ../../../Debug+Asserts/lib/Hello.so -gcse -hello -licm --debug-pass=Structure < hello.bc > /dev/null 1599 Pass Arguments: -gcse -hello -licm 1600 Module Pass Manager 1601 Function Pass Manager 1602 Dominator Set Construction 1603 Immediate Dominators Construction 1604 Global Common Subexpression Elimination 1605 -- Immediate Dominators Construction 1606 -- Global Common Subexpression Elimination 1607 Hello World Pass 1608 -- Hello World Pass 1609 Natural Loop Construction 1610 Loop Invariant Code Motion 1611 -- Loop Invariant Code Motion 1612 -- Natural Loop Construction 1613 Module Verifier 1614 -- Dominator Set Construction 1615 -- Module Verifier 1616 Bitcode Writer 1617 --Bitcode Writer 1618 Hello: __main 1619 Hello: puts 1620 Hello: main 1621 </pre></div> 1622 1623 <p>Which shows that we don't accidentally invalidate dominator information 1624 anymore, and therefore do not have to compute it twice.</p> 1625 1626 <!-- _______________________________________________________________________ --> 1627 <h4> 1628 <a name="releaseMemory">The <tt>releaseMemory</tt> method</a> 1629 </h4> 1630 1631 <div> 1632 1633 <div class="doc_code"><pre> 1634 <b>virtual void</b> releaseMemory(); 1635 </pre></div> 1636 1637 <p>The <tt>PassManager</tt> automatically determines when to compute analysis 1638 results, and how long to keep them around for. Because the lifetime of the pass 1639 object itself is effectively the entire duration of the compilation process, we 1640 need some way to free analysis results when they are no longer useful. The 1641 <tt>releaseMemory</tt> virtual method is the way to do this.</p> 1642 1643 <p>If you are writing an analysis or any other pass that retains a significant 1644 amount of state (for use by another pass which "requires" your pass and uses the 1645 <a href="#getAnalysis">getAnalysis</a> method) you should implement 1646 <tt>releaseMemory</tt> to, well, release the memory allocated to maintain this 1647 internal state. This method is called after the <tt>run*</tt> method for the 1648 class, before the next call of <tt>run*</tt> in your pass.</p> 1649 1650 </div> 1651 1652 </div> 1653 1654 <!-- *********************************************************************** --> 1655 <h2> 1656 <a name="registering">Registering dynamically loaded passes</a> 1657 </h2> 1658 <!-- *********************************************************************** --> 1659 1660 <div> 1661 1662 <p><i>Size matters</i> when constructing production quality tools using llvm, 1663 both for the purposes of distribution, and for regulating the resident code size 1664 when running on the target system. Therefore, it becomes desirable to 1665 selectively use some passes, while omitting others and maintain the flexibility 1666 to change configurations later on. You want to be able to do all this, and, 1667 provide feedback to the user. This is where pass registration comes into 1668 play.</p> 1669 1670 <p>The fundamental mechanisms for pass registration are the 1671 <tt>MachinePassRegistry</tt> class and subclasses of 1672 <tt>MachinePassRegistryNode</tt>.</p> 1673 1674 <p>An instance of <tt>MachinePassRegistry</tt> is used to maintain a list of 1675 <tt>MachinePassRegistryNode</tt> objects. This instance maintains the list and 1676 communicates additions and deletions to the command line interface.</p> 1677 1678 <p>An instance of <tt>MachinePassRegistryNode</tt> subclass is used to maintain 1679 information provided about a particular pass. This information includes the 1680 command line name, the command help string and the address of the function used 1681 to create an instance of the pass. A global static constructor of one of these 1682 instances <i>registers</i> with a corresponding <tt>MachinePassRegistry</tt>, 1683 the static destructor <i>unregisters</i>. Thus a pass that is statically linked 1684 in the tool will be registered at start up. A dynamically loaded pass will 1685 register on load and unregister at unload.</p> 1686 1687 <!-- _______________________________________________________________________ --> 1688 <h3> 1689 <a name="registering_existing">Using existing registries</a> 1690 </h3> 1691 1692 <div> 1693 1694 <p>There are predefined registries to track instruction scheduling 1695 (<tt>RegisterScheduler</tt>) and register allocation (<tt>RegisterRegAlloc</tt>) 1696 machine passes. Here we will describe how to <i>register</i> a register 1697 allocator machine pass.</p> 1698 1699 <p>Implement your register allocator machine pass. In your register allocator 1700 .cpp file add the following include;</p> 1701 1702 <div class="doc_code"><pre> 1703 #include "llvm/CodeGen/RegAllocRegistry.h" 1704 </pre></div> 1705 1706 <p>Also in your register allocator .cpp file, define a creator function in the 1707 form; </p> 1708 1709 <div class="doc_code"><pre> 1710 FunctionPass *createMyRegisterAllocator() { 1711 return new MyRegisterAllocator(); 1712 } 1713 </pre></div> 1714 1715 <p>Note that the signature of this function should match the type of 1716 <tt>RegisterRegAlloc::FunctionPassCtor</tt>. In the same file add the 1717 "installing" declaration, in the form;</p> 1718 1719 <div class="doc_code"><pre> 1720 static RegisterRegAlloc myRegAlloc("myregalloc", 1721 " my register allocator help string", 1722 createMyRegisterAllocator); 1723 </pre></div> 1724 1725 <p>Note the two spaces prior to the help string produces a tidy result on the 1726 -help query.</p> 1727 1728 <div class="doc_code"><pre> 1729 $ llc -help 1730 ... 1731 -regalloc - Register allocator to use (default=linearscan) 1732 =linearscan - linear scan register allocator 1733 =local - local register allocator 1734 =simple - simple register allocator 1735 =myregalloc - my register allocator help string 1736 ... 1737 </pre></div> 1738 1739 <p>And that's it. The user is now free to use <tt>-regalloc=myregalloc</tt> as 1740 an option. Registering instruction schedulers is similar except use the 1741 <tt>RegisterScheduler</tt> class. Note that the 1742 <tt>RegisterScheduler::FunctionPassCtor</tt> is significantly different from 1743 <tt>RegisterRegAlloc::FunctionPassCtor</tt>.</p> 1744 1745 <p>To force the load/linking of your register allocator into the llc/lli tools, 1746 add your creator function's global declaration to "Passes.h" and add a "pseudo" 1747 call line to <tt>llvm/Codegen/LinkAllCodegenComponents.h</tt>.</p> 1748 1749 </div> 1750 1751 1752 <!-- _______________________________________________________________________ --> 1753 <h3> 1754 <a name="registering_new">Creating new registries</a> 1755 </h3> 1756 1757 <div> 1758 1759 <p>The easiest way to get started is to clone one of the existing registries; we 1760 recommend <tt>llvm/CodeGen/RegAllocRegistry.h</tt>. The key things to modify 1761 are the class name and the <tt>FunctionPassCtor</tt> type.</p> 1762 1763 <p>Then you need to declare the registry. Example: if your pass registry is 1764 <tt>RegisterMyPasses</tt> then define;</p> 1765 1766 <div class="doc_code"><pre> 1767 MachinePassRegistry RegisterMyPasses::Registry; 1768 </pre></div> 1769 1770 <p>And finally, declare the command line option for your passes. Example:</p> 1771 1772 <div class="doc_code"><pre> 1773 cl::opt<RegisterMyPasses::FunctionPassCtor, false, 1774 RegisterPassParser<RegisterMyPasses> > 1775 MyPassOpt("mypass", 1776 cl::init(&createDefaultMyPass), 1777 cl::desc("my pass option help")); 1778 </pre></div> 1779 1780 <p>Here the command option is "mypass", with createDefaultMyPass as the default 1781 creator.</p> 1782 1783 </div> 1784 1785 </div> 1786 1787 <!-- *********************************************************************** --> 1788 <h2> 1789 <a name="debughints">Using GDB with dynamically loaded passes</a> 1790 </h2> 1791 <!-- *********************************************************************** --> 1792 1793 <div> 1794 1795 <p>Unfortunately, using GDB with dynamically loaded passes is not as easy as it 1796 should be. First of all, you can't set a breakpoint in a shared object that has 1797 not been loaded yet, and second of all there are problems with inlined functions 1798 in shared objects. Here are some suggestions to debugging your pass with 1799 GDB.</p> 1800 1801 <p>For sake of discussion, I'm going to assume that you are debugging a 1802 transformation invoked by <tt>opt</tt>, although nothing described here depends 1803 on that.</p> 1804 1805 <!-- _______________________________________________________________________ --> 1806 <h4> 1807 <a name="breakpoint">Setting a breakpoint in your pass</a> 1808 </h4> 1809 1810 <div> 1811 1812 <p>First thing you do is start <tt>gdb</tt> on the <tt>opt</tt> process:</p> 1813 1814 <div class="doc_code"><pre> 1815 $ <b>gdb opt</b> 1816 GNU gdb 5.0 1817 Copyright 2000 Free Software Foundation, Inc. 1818 GDB is free software, covered by the GNU General Public License, and you are 1819 welcome to change it and/or distribute copies of it under certain conditions. 1820 Type "show copying" to see the conditions. 1821 There is absolutely no warranty for GDB. Type "show warranty" for details. 1822 This GDB was configured as "sparc-sun-solaris2.6"... 1823 (gdb) 1824 </pre></div> 1825 1826 <p>Note that <tt>opt</tt> has a lot of debugging information in it, so it takes 1827 time to load. Be patient. Since we cannot set a breakpoint in our pass yet 1828 (the shared object isn't loaded until runtime), we must execute the process, and 1829 have it stop before it invokes our pass, but after it has loaded the shared 1830 object. The most foolproof way of doing this is to set a breakpoint in 1831 <tt>PassManager::run</tt> and then run the process with the arguments you 1832 want:</p> 1833 1834 <div class="doc_code"><pre> 1835 (gdb) <b>break llvm::PassManager::run</b> 1836 Breakpoint 1 at 0x2413bc: file Pass.cpp, line 70. 1837 (gdb) <b>run test.bc -load $(LLVMTOP)/llvm/Debug+Asserts/lib/[libname].so -[passoption]</b> 1838 Starting program: opt test.bc -load $(LLVMTOP)/llvm/Debug+Asserts/lib/[libname].so -[passoption] 1839 Breakpoint 1, PassManager::run (this=0xffbef174, M=@0x70b298) at Pass.cpp:70 1840 70 bool PassManager::run(Module &M) { return PM->run(M); } 1841 (gdb) 1842 </pre></div> 1843 1844 <p>Once the <tt>opt</tt> stops in the <tt>PassManager::run</tt> method you are 1845 now free to set breakpoints in your pass so that you can trace through execution 1846 or do other standard debugging stuff.</p> 1847 1848 </div> 1849 1850 <!-- _______________________________________________________________________ --> 1851 <h4> 1852 <a name="debugmisc">Miscellaneous Problems</a> 1853 </h4> 1854 1855 <div> 1856 1857 <p>Once you have the basics down, there are a couple of problems that GDB has, 1858 some with solutions, some without.</p> 1859 1860 <ul> 1861 <li>Inline functions have bogus stack information. In general, GDB does a 1862 pretty good job getting stack traces and stepping through inline functions. 1863 When a pass is dynamically loaded however, it somehow completely loses this 1864 capability. The only solution I know of is to de-inline a function (move it 1865 from the body of a class to a .cpp file).</li> 1866 1867 <li>Restarting the program breaks breakpoints. After following the information 1868 above, you have succeeded in getting some breakpoints planted in your pass. Nex 1869 thing you know, you restart the program (i.e., you type '<tt>run</tt>' again), 1870 and you start getting errors about breakpoints being unsettable. The only way I 1871 have found to "fix" this problem is to <tt>delete</tt> the breakpoints that are 1872 already set in your pass, run the program, and re-set the breakpoints once 1873 execution stops in <tt>PassManager::run</tt>.</li> 1874 1875 </ul> 1876 1877 <p>Hopefully these tips will help with common case debugging situations. If 1878 you'd like to contribute some tips of your own, just contact <a 1879 href="mailto:sabre (a] nondot.org">Chris</a>.</p> 1880 1881 </div> 1882 1883 </div> 1884 1885 <!-- *********************************************************************** --> 1886 <h2> 1887 <a name="future">Future extensions planned</a> 1888 </h2> 1889 <!-- *********************************************************************** --> 1890 1891 <div> 1892 1893 <p>Although the LLVM Pass Infrastructure is very capable as it stands, and does 1894 some nifty stuff, there are things we'd like to add in the future. Here is 1895 where we are going:</p> 1896 1897 <!-- _______________________________________________________________________ --> 1898 <h4> 1899 <a name="SMP">Multithreaded LLVM</a> 1900 </h4> 1901 1902 <div> 1903 1904 <p>Multiple CPU machines are becoming more common and compilation can never be 1905 fast enough: obviously we should allow for a multithreaded compiler. Because of 1906 the semantics defined for passes above (specifically they cannot maintain state 1907 across invocations of their <tt>run*</tt> methods), a nice clean way to 1908 implement a multithreaded compiler would be for the <tt>PassManager</tt> class 1909 to create multiple instances of each pass object, and allow the separate 1910 instances to be hacking on different parts of the program at the same time.</p> 1911 1912 <p>This implementation would prevent each of the passes from having to implement 1913 multithreaded constructs, requiring only the LLVM core to have locking in a few 1914 places (for global resources). Although this is a simple extension, we simply 1915 haven't had time (or multiprocessor machines, thus a reason) to implement this. 1916 Despite that, we have kept the LLVM passes SMP ready, and you should too.</p> 1917 1918 </div> 1919 1920 </div> 1921 1922 <!-- *********************************************************************** --> 1923 <hr> 1924 <address> 1925 <a href="http://jigsaw.w3.org/css-validator/check/referer"><img 1926 src="http://jigsaw.w3.org/css-validator/images/vcss-blue" alt="Valid CSS"></a> 1927 <a href="http://validator.w3.org/check/referer"><img 1928 src="http://www.w3.org/Icons/valid-html401-blue" alt="Valid HTML 4.01"></a> 1929 1930 <a href="mailto:sabre (a] nondot.org">Chris Lattner</a><br> 1931 <a href="http://llvm.org/">The LLVM Compiler Infrastructure</a><br> 1932 Last modified: $Date$ 1933 </address> 1934 1935 </body> 1936 </html> 1937
