1 <!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.01 Transitional//EN">
2 <html><head><title>Writing an LLVM Pass</title></head>
6 <table width="100%" bgcolor="#330077" border=0 cellpadding=4 cellspacing=0>
7 <tr><td> <font size=+3 color="#EEEEFF" face="Georgia,Palatino,Times,Roman"><b>Writing an LLVM Pass</b></font></td>
12 <li><a href="#introduction">Introduction - What is a pass?</a>
13 <li><a href="#quickstart">Quick Start - Writing hello world</a>
15 <li><a href="#makefile">Setting up the build environment</a>
16 <li><a href="#basiccode">Basic code required</a>
17 <li><a href="#running">Running a pass with <tt>opt</tt>
18 or <tt>analyze</tt></a>
20 <li><a href="#passtype">Pass classes and requirements</a>
22 <li><a href="#ImmutablePass">The <tt>ImmutablePass</tt> class</a>
23 <li><a href="#Pass">The <tt>Pass</tt> class</a>
25 <li><a href="#run">The <tt>run</tt> method</a>
27 <li><a href="#FunctionPass">The <tt>FunctionPass</tt> class</a>
29 <li><a href="#doInitialization_mod">The <tt>doInitialization(Module
30 &)</tt> method</a>
31 <li><a href="#runOnFunction">The <tt>runOnFunction</tt> method</a>
32 <li><a href="#doFinalization_mod">The <tt>doFinalization(Module
33 &)</tt> method</a>
35 <li><a href="#BasicBlockPass">The <tt>BasicBlockPass</tt> class</a>
37 <li><a href="#doInitialization_fn">The <tt>doInitialization(Function
38 &)</tt> method</a>
39 <li><a href="#runOnBasicBlock">The <tt>runOnBasicBlock</tt> method</a>
40 <li><a href="#doFinalization_fn">The <tt>doFinalization(Function
41 &)</tt> method</a>
44 <li><a href="#registration">Pass Registration</a>
46 <li><a href="#print">The <tt>print</tt> method</a>
48 <li><a href="#interaction">Specifying interactions between passes</a>
50 <li><a href="#getAnalysisUsage">The <tt>getAnalysisUsage</tt> method</a>
51 <li><a href="#getAnalysis">The <tt>getAnalysis</tt> method</a>
53 <li><a href="#analysisgroup">Implementing Analysis Groups</a>
55 <li><a href="#agconcepts">Analysis Group Concepts</a>
56 <li><a href="#registerag">Using <tt>RegisterAnalysisGroup</tt></a>
58 <li><a href="#passmanager">What PassManager does</a>
60 <li><a href="#releaseMemory">The <tt>releaseMemory</tt> method</a>
62 <li><a href="#debughints">Using GDB with dynamically loaded passes</a>
64 <li><a href="#breakpoint">Setting a breakpoint in your pass
65 <li><a href="#debugmisc">Miscellaneous Problems
67 <li><a href="#future">Future extensions planned</a>
69 <li><a href="#SMP">Multithreaded LLVM</a>
70 <li><a href="#ModuleSource">A new <tt>ModuleSource</tt> interface</a>
71 <li><a href="#PassFunctionPass"><tt>Pass</tt>'s requiring <tt>FunctionPass</tt>'s</a>
74 <p><b>Written by <a href="mailto:sabre@nondot.org">Chris Lattner</a></b><p>
79 <!-- *********************************************************************** -->
80 <table width="100%" bgcolor="#330077" border=0 cellpadding=4 cellspacing=0>
81 <tr><td align=center><font color="#EEEEFF" size=+2 face="Georgia,Palatino"><b>
82 <a name="introduction">Introduction - What is a pass?
83 </b></font></td></tr></table><ul>
84 <!-- *********************************************************************** -->
86 The LLVM Pass Framework is an important part of the LLVM system, because LLVM
87 passes are where the interesting parts of the compiler exist. Passes perform
88 the transformations and optimizations that make up the compiler, they build
89 the analysis results that are used by these transformations, and they are, above
90 all, a structuring technique for compiler code.<p>
92 All LLVM passes are subclasses of the <tt><a
93 href="http://llvm.cs.uiuc.edu/doxygen/classPass.html">Pass</a></tt> class, which
94 implement functionality by overriding virtual methods inherited from
95 <tt>Pass</tt>. Depending on how your pass works, you may be able to inherit
97 href="http://llvm.cs.uiuc.edu/doxygen/structFunctionPass.html">FunctionPass</a></tt>
99 href="http://llvm.cs.uiuc.edu/doxygen/structBasicBlockPass.html">BasicBlockPass</a></tt>,
100 which gives the system more information about what your pass does, and how it
101 can be combined with other passes. One of the main features of the LLVM Pass
102 Framework is that it schedules passes to run in an efficient way based on the
103 constraints that your pass has.<p>
105 We start by showing you how to construct a pass, everything from setting up the
106 code, to compiling, loading, and executing it. After the basics are down, more
107 advanced features are discussed.<p>
110 <!-- *********************************************************************** -->
111 </ul><table width="100%" bgcolor="#330077" border=0 cellpadding=4 cellspacing=0>
112 <tr><td align=center><font color="#EEEEFF" size=+2 face="Georgia,Palatino"><b>
113 <a name="quickstart">Quick Start - Writing hello world
114 </b></font></td></tr></table><ul>
115 <!-- *********************************************************************** -->
117 Here we describe how to write the "hello world" of passes. The "Hello" pass is
118 designed to simply print out the name of non-external functions that exist in
119 the program being compiled. It does not modify the program at all, just
120 inspects it. The source code and files for this pass are available in the LLVM
121 source tree in the <tt>lib/Transforms/Hello</tt> directory.<p>
124 <!-- ======================================================================= -->
125 </ul><table width="100%" bgcolor="#441188" border=0 cellpadding=4 cellspacing=0>
126 <tr><td> </td><td width="100%">
127 <font color="#EEEEFF" face="Georgia,Palatino"><b>
128 <a name="makefile">Setting up the build environment
129 </b></font></td></tr></table><ul>
131 First thing you need to do is create a new directory somewhere in the LLVM
132 source base. For this example, we'll assume that you made
133 "<tt>lib/Transforms/Hello</tt>". The first thing you must do is set up a build
134 script (Makefile) that will compile the source code for the new pass. To do
135 this, copy this into "<tt>Makefile</tt>":<p>
138 # Makefile for hello pass
140 # Path to top level of LLVM heirarchy
143 # Name of the library to build
146 # Build a dynamically loadable shared object
149 # Include the makefile implementation stuff
150 include $(LEVEL)/Makefile.common
151 </pre></ul><hr><ul><p>
153 This makefile specifies that all of the <tt>.cpp</tt> files in the current
154 directory are to be compiled and linked together into a
155 <tt>lib/Debug/libhello.so</tt> shared object that can be dynamically loaded by
156 the <tt>opt</tt> or <tt>analyze</tt> tools.<p>
158 Now that we have the build scripts set up, we just need to write the code for
162 <!-- ======================================================================= -->
163 </ul><table width="100%" bgcolor="#441188" border=0 cellpadding=4 cellspacing=0>
164 <tr><td> </td><td width="100%">
165 <font color="#EEEEFF" face="Georgia,Palatino"><b>
166 <a name="basiccode">Basic code required
167 </b></font></td></tr></table><ul>
169 Now that we have a way to compile our new pass, we just have to write it. Start
173 <b>#include</b> "<a href="http://llvm.cs.uiuc.edu/doxygen/Pass_8h-source.html">llvm/Pass.h</a>"
174 <b>#include</b> "<a href="http://llvm.cs.uiuc.edu/doxygen/Function_8h-source.html">llvm/Function.h</a>"
177 Which are needed because we are writing a <tt><a
178 href="http://llvm.cs.uiuc.edu/doxygen/classPass.html">Pass</a></tt>, and we are
180 href="http://llvm.cs.uiuc.edu/doxygen/classFunction.html">Function</a></tt>'s.<p>
188 ... which starts out an anonymous namespace. Anonymous namespaces are to C++
189 what the "<tt>static</tt>" keyword is to C (at global scope). It makes the
190 things declared inside of the anonymous namespace only visible to the current
191 file. If you're not familiar with them, consult a decent C++ book for more
194 Next, we declare our pass itself:<p>
197 <b>struct</b> Hello : <b>public</b> <a href="#FunctionPass">FunctionPass</a> {
200 This declares a "<tt>Hello</tt>" class that is a subclass of <tt><a
201 href="http://llvm.cs.uiuc.edu/doxygen/structFunctionPass.html">FunctionPass</a></tt>.
202 The different builtin pass subclasses are described in detail <a
203 href="#passtype">later</a>, but for now, know that <a href="#FunctionPass"><tt>FunctionPass</tt></a>'s
204 operate a function at a time.<p>
207 <b>virtual bool</b> <a href="#runOnFunction">runOnFunction</a>(Function &F) {
208 std::cerr << "<i>Hello: </i>" << F.getName() << "\n";
211 }; <i>// end of struct Hello</i>
214 We declare a "<a href="#runOnFunction"><tt>runOnFunction</tt></a>" method, which
215 overloads an abstract virtual method inherited from <a
216 href="#FunctionPass"><tt>FunctionPass</tt></a>. This is where we are supposed
217 to do our thing, so we just print out our message with the name of each
221 RegisterOpt<Hello> X("<i>hello</i>", "<i>Hello World Pass</i>");
222 } <i>// end of anonymous namespace</i>
225 Lastly, we register our class <tt>Hello</tt>, giving it a command line argument
226 "<tt>hello</tt>", and a name "<tt>Hello World Pass</tt>". There are several
227 different ways of <a href="#registration">registering your pass</a>, depending
228 on what it is to be used for. For "optimizations" we use the
229 <tt>RegisterOpt</tt> template.<p>
231 As a whole, the <tt>.cpp</tt> file looks like:<p>
234 <b>#include</b> "<a href="http://llvm.cs.uiuc.edu/doxygen/Pass_8h-source.html">llvm/Pass.h</a>"
235 <b>#include</b> "<a href="http://llvm.cs.uiuc.edu/doxygen/Function_8h-source.html">llvm/Function.h</a>"
238 <b>struct Hello</b> : <b>public</b> <a href="#FunctionPass">FunctionPass</a> {
239 <b>virtual bool</b> <a href="#runOnFunction">runOnFunction</a>(Function &F) {
240 std::cerr << "<i>Hello: </i>" << F.getName() << "\n";
245 RegisterOpt<Hello> X("<i>hello</i>", "<i>Hello World Pass</i>");
249 Now that it's all together, compile the file with a simple "<tt>gmake</tt>"
250 command in the local directory and you should get a new
251 "<tt>lib/Debug/libhello.so</tt> file. Note that everything in this file is
252 contained in an anonymous namespace: this reflects the fact that passes are self
253 contained units that do not need external interfaces (although they can have
254 them) to be useful.<p>
257 <!-- ======================================================================= -->
258 </ul><table width="100%" bgcolor="#441188" border=0 cellpadding=4 cellspacing=0>
259 <tr><td> </td><td width="100%">
260 <font color="#EEEEFF" face="Georgia,Palatino"><b>
261 <a name="running">Running a pass with <tt>opt</tt> or <tt>analyze</tt>
262 </b></font></td></tr></table><ul>
264 Now that you have a brand new shiny <tt>.so</tt> file, we can use the
265 <tt>opt</tt> command to run an LLVM program through your pass. Because you
266 registered your pass with the <tt>RegisterOpt</tt> template, you will be able to
267 use the <tt>opt</tt> tool to access it, once loaded.<p>
269 To test it, follow the example at the end of the <a
270 href="GettingStarted.html">Getting Started Guide</a> to compile "Hello World" to
271 LLVM. We can now run the bytecode file (<tt>hello.bc</tt>) for the program
272 through our transformation like this (or course, any bytecode file will
276 $ opt -load ../../../lib/Debug/libhello.so -hello < hello.bc > /dev/null
282 The '<tt>-load</tt>' option specifies that '<tt>opt</tt>' should load your pass
283 as a shared object, which makes '<tt>-hello</tt>' a valid command line argument
284 (which is one reason you need to <a href="#registration">register your
285 pass</a>). Because the hello pass does not modify the program in any
286 interesting way, we just throw away the result of <tt>opt</tt> (sending it to
287 <tt>/dev/null</tt>).<p>
289 To see what happened to the other string you registered, try running
290 <tt>opt</tt> with the <tt>--help</tt> option:<p>
293 $ opt -load ../../../lib/Debug/libhello.so --help
294 OVERVIEW: llvm .bc -> .bc modular optimizer
296 USAGE: opt [options] <input bytecode>
299 Optimizations available:
301 -funcresolve - Resolve Functions
302 -gcse - Global Common Subexpression Elimination
303 -globaldce - Dead Global Elimination
304 <b>-hello - Hello World Pass</b>
305 -indvars - Cannonicalize Induction Variables
306 -inline - Function Integration/Inlining
307 -instcombine - Combine redundant instructions
311 The pass name get added as the information string for your pass, giving some
312 documentation to users of <tt>opt</tt>. Now that you have a working pass, you
313 would go ahead and make it do the cool transformations you want. Once you get
314 it all working and tested, it may become useful to find out how fast your pass
315 is. The <a href="#passManager"><tt>PassManager</tt></a> provides a nice command
316 line option (<tt>--time-passes</tt>) that allows you to get information about
317 the execution time of your pass along with the other passes you queue up. For
321 $ opt -load ../../../lib/Debug/libhello.so -hello -time-passes < hello.bc > /dev/null
325 ===============================================================================
326 ... Pass execution timing report ...
327 ===============================================================================
328 Total Execution Time: 0.02 seconds (0.0479059 wall clock)
330 ---User Time--- --System Time-- --User+System-- ---Wall Time--- --- Pass Name ---
331 0.0100 (100.0%) 0.0000 ( 0.0%) 0.0100 ( 50.0%) 0.0402 ( 84.0%) Bytecode Writer
332 0.0000 ( 0.0%) 0.0100 (100.0%) 0.0100 ( 50.0%) 0.0031 ( 6.4%) Dominator Set Construction
333 0.0000 ( 0.0%) 0.0000 ( 0.0%) 0.0000 ( 0.0%) 0.0013 ( 2.7%) Module Verifier
334 <b> 0.0000 ( 0.0%) 0.0000 ( 0.0%) 0.0000 ( 0.0%) 0.0033 ( 6.9%) Hello World Pass</b>
335 0.0100 (100.0%) 0.0100 (100.0%) 0.0200 (100.0%) 0.0479 (100.0%) TOTAL
338 As you can see, our implementation above is pretty fast :). The additional
339 passes listed are automatically inserted by the '<tt>opt</tt>' tool to verify
340 that the LLVM emitted by your pass is still valid and well formed LLVM, which
341 hasn't been broken somehow.
343 Now that you have seen the basics of the mechanics behind passes, we can talk
344 about some more details of how they work and how to use them.<p>
348 <!-- *********************************************************************** -->
349 </ul><table width="100%" bgcolor="#330077" border=0 cellpadding=4 cellspacing=0>
350 <tr><td align=center><font color="#EEEEFF" size=+2 face="Georgia,Palatino"><b>
351 <a name="passtype">Pass classes and requirements
352 </b></font></td></tr></table><ul>
353 <!-- *********************************************************************** -->
355 One of the first things that you should do when designing a new pass is to
356 decide what class you should subclass for your pass. The <a
357 href="#basiccode">Hello World</a> example uses the <tt><a
358 href="#FunctionPass">FunctionPass</a></tt> class for its implementation, but we
359 did not discuss why or when this should occur. Here we talk about the classes
360 available, from the most general to the most specific.<p>
362 When choosing a superclass for your Pass, you should choose the <b>most
363 specific</b> class possible, while still being able to meet the requirements
364 listed. This gives the LLVM Pass Infrastructure information neccesary to
365 optimize how passes are run, so that the resultant compiler isn't unneccesarily
369 <!-- ======================================================================= -->
370 </ul><table width="100%" bgcolor="#441188" border=0 cellpadding=4 cellspacing=0>
371 <tr><td> </td><td width="100%">
372 <font color="#EEEEFF" face="Georgia,Palatino"><b>
373 <a name="ImmutablePass">The <tt>ImmutablePass</tt> class
374 </b></font></td></tr></table><ul>
376 The most plain and boring type of pass is the "<tt><a
377 href="http://llvm.cs.uiuc.edu/doxygen/structImmutablePass.html">ImmutablePass</a></tt>"
378 class. This pass type is used for passes that do not have to be run, do not
379 change state, and never need to be updated. This is not a normal type of
380 transformation or analysis, but can provide information about the current
381 compiler configuration.<p>
383 Although this pass class is very infrequently used, it is important for
384 providing information about the current target machine being compiled for, and
385 other static information that can affect the various transformations.<p>
387 <tt>ImmutablePass</tt>'s never invalidate other transformations, are never
388 invalidated, and are never "run".<p>
391 <!-- ======================================================================= -->
392 </ul><table width="100%" bgcolor="#441188" border=0 cellpadding=4 cellspacing=0>
393 <tr><td> </td><td width="100%">
394 <font color="#EEEEFF" face="Georgia,Palatino"><b>
395 <a name="Pass">The <tt>Pass</tt> class
396 </b></font></td></tr></table><ul>
398 The "<tt><a href="http://llvm.cs.uiuc.edu/doxygen/classPass.html">Pass</a></tt>"
399 class is the most general of all superclasses that you can use. Deriving from
400 <tt>Pass</tt> indicates that your pass uses the entire program as a unit,
401 refering to function bodies in no predictable order, or adding and removing
402 functions. Because nothing is known about the behavior of direct <tt>Pass</tt>
403 subclasses, no optimization can be done for their execution.<p>
405 To write a correct <tt>Pass</tt> subclass, derive from <tt>Pass</tt> and
406 overload the <tt>run</tt> method with the following signature:<p>
408 <!-- _______________________________________________________________________ -->
409 </ul><h4><a name="run"><hr size=0>The <tt>run</tt> method</h4><ul>
413 <b>virtual bool</b> run(Module &M) = 0;
416 The <tt>run</tt> method performs the interesting work of the pass, and should
417 return true if the module was modified by the transformation, false
422 <!-- ======================================================================= -->
423 </ul><table width="100%" bgcolor="#441188" border=0 cellpadding=4 cellspacing=0>
424 <tr><td> </td><td width="100%">
425 <font color="#EEEEFF" face="Georgia,Palatino"><b>
426 <a name="FunctionPass">The <tt>FunctionPass</tt> class
427 </b></font></td></tr></table><ul>
429 In contrast to direct <tt>Pass</tt> subclasses, direct <tt><a
430 href="http://llvm.cs.uiuc.edu/doxygen/classPass.html">FunctionPass</a></tt>
431 subclasses do have a predictable, local behavior that can be expected by the
432 system. All <tt>FunctionPass</tt> execute on each function in the program
433 independant of all of the other functions in the program.
434 <tt>FunctionPass</tt>'s do not require that they are executed in a particular
435 order, and <tt>FunctionPass</tt>'s do not modify external functions.<p>
437 To be explicit, <tt>FunctionPass</tt> subclasses are not allowed to:<p>
440 <li>Modify a Function other than the one currently being processed.
441 <li>Add or remove Function's from the current Module.
442 <li>Add or remove global variables from the current Module.
443 <li>Maintain state across invocations of
444 <a href="#runOnFunction"><tt>runOnFunction</tt></a> (including global data)
447 Implementing a <tt>FunctionPass</tt> is usually straightforward (See the <a
448 href="#basiccode">Hello World</a> pass for example). <tt>FunctionPass</tt>'s
449 may overload three virtual methods to do their work. All of these methods
450 should return true if they modified the program, or false if they didn't.<p>
452 <!-- _______________________________________________________________________ -->
453 </ul><h4><a name="doInitialization_mod"><hr size=0>The
454 <tt>doInitialization(Module &)</tt> method</h4><ul>
457 <b>virtual bool</b> doInitialization(Module &M);
460 The <tt>doIninitialize</tt> method is allowed to do most of the things that
461 <tt>FunctionPass</tt>'s are not allowed to do. They can add and remove
462 functions, get pointers to functions, etc. The <tt>doInitialization</tt> method
463 is designed to do simple initialization type of stuff that does not depend on
464 the functions being processed. The <tt>doInitialization</tt> method call is not
465 scheduled to overlap with any other pass executions (thus it should be very
468 A good example of how this method should be used is the <a
469 href="http://llvm.cs.uiuc.edu/doxygen/LowerAllocations_8cpp-source.html">LowerAllocations</a>
470 pass. This pass converts <tt>malloc</tt> and <tt>free</tt> instructions into
471 platform dependant <tt>malloc()</tt> and <tt>free()</tt> function calls. It
472 uses the <tt>doInitialization</tt> method to get a reference to the malloc and
473 free functions that it needs, adding prototypes to the module if neccesary.<p>
475 <!-- _______________________________________________________________________ -->
476 </ul><h4><a name="runOnFunction"><hr size=0>The <tt>runOnFunction</tt> method</h4><ul>
479 <b>virtual bool</b> runOnFunction(Function &F) = 0;
482 The <tt>runOnFunction</tt> method must be implemented by your subclass to do the
483 transformation or analysis work of your pass. As usual, a true value should be
484 returned if the function is modified.<p>
486 <!-- _______________________________________________________________________ -->
487 </ul><h4><a name="doFinalization_mod"><hr size=0>The <tt>doFinalization(Module &)</tt> method</h4><ul>
490 <b>virtual bool</b> doFinalization(Module &M);
493 The <tt>doFinalization</tt> method is an infrequently used method that is called
494 when the pass framework has finished calling <a
495 href="#runOnFunction"><tt>runOnFunction</tt></a> for every function in the
496 program being compiled.<p>
500 <!-- ======================================================================= -->
501 </ul><table width="100%" bgcolor="#441188" border=0 cellpadding=4 cellspacing=0>
502 <tr><td> </td><td width="100%">
503 <font color="#EEEEFF" face="Georgia,Palatino"><b>
504 <a name="BasicBlockPass">The <tt>BasicBlockPass</tt> class</a>
505 </b></font></td></tr></table><ul>
507 <tt>BasicBlockPass</tt>'s are just like <a
508 href="#FunctionPass"><tt>FunctionPass</tt></a>'s, except that they must limit
509 their scope of inspection and modification to a single basic block at a time.
510 As such, they are <b>not</b> allowed to do any of the following:<p>
513 <li>Modify or inspect any basic blocks outside of the current one
514 <li>Maintain state across invocations of
515 <a href="#runOnBasicBlock"><tt>runOnBasicBlock</tt></a>
516 <li>Modify the constrol flow graph (by altering terminator instructions)
517 <li>Any of the things verboten for
518 <a href="#FunctionPass"><tt>FunctionPass</tt></a>'s.
521 <tt>BasicBlockPass</tt>'s are useful for traditional local and "peephole"
522 optimizations. They may override the same <a
523 href="#doInitialization_mod"><tt>doInitialization(Module &)</tt></a> and <a
524 href="#doFinalization_mod"><tt>doFinalization(Module &)</tt></a> methods that <a
525 href="#FunctionPass"><tt>FunctionPass</tt></a>'s have, but also have the following virtual methods that may also be implemented:<p>
527 <!-- _______________________________________________________________________ -->
528 </ul><h4><a name="doInitialization_fn"><hr size=0>The
529 <tt>doInitialization(Function &)</tt> method</h4><ul>
532 <b>virtual bool</b> doInitialization(Function &F);
535 The <tt>doIninitialize</tt> method is allowed to do most of the things that
536 <tt>BasicBlockPass</tt>'s are not allowed to do, but that
537 <tt>FunctionPass</tt>'s can. The <tt>doInitialization</tt> method is designed
538 to do simple initialization type of stuff that does not depend on the
539 BasicBlocks being processed. The <tt>doInitialization</tt> method call is not
540 scheduled to overlap with any other pass executions (thus it should be very
544 <!-- _______________________________________________________________________ -->
545 </ul><h4><a name="runOnBasicBlock"><hr size=0>The <tt>runOnBasicBlock</tt> method</h4><ul>
548 <b>virtual bool</b> runOnBasicBlock(BasicBlock &BB) = 0;
551 Override this function to do the work of the <tt>BasicBlockPass</tt>. This
552 function is not allowed to inspect or modify basic blocks other than the
553 parameter, and are not allowed to modify the CFG. A true value must be returned
554 if the basic block is modified.<p>
557 <!-- _______________________________________________________________________ -->
558 </ul><h4><a name="doFinalization_fn"><hr size=0>The <tt>doFinalization(Function
559 &)</tt> method</h4><ul>
562 <b>virtual bool</b> doFinalization(Function &F);
565 The <tt>doFinalization</tt> method is an infrequently used method that is called
566 when the pass framework has finished calling <a
567 href="#runOnBasicBlock"><tt>runOnBasicBlock</tt></a> for every BasicBlock in the
568 program being compiled. This can be used to perform per-function
573 <!-- *********************************************************************** -->
574 </ul><table width="100%" bgcolor="#330077" border=0 cellpadding=4 cellspacing=0>
575 <tr><td align=center><font color="#EEEEFF" size=+2 face="Georgia,Palatino"><b>
576 <a name="registration">Pass registration
577 </b></font></td></tr></table><ul>
578 <!-- *********************************************************************** -->
580 In the <a href="#basiccode">Hello World</a> example pass we illustrated how pass
581 registration works, and discussed some of the reasons that it is used and what
582 it does. Here we discuss how and why passes are registered.<p>
584 Passes can be registered in several different ways. Depending on the general
585 classification of the pass, you should use one of the following templates to
586 register the pass:<p>
589 <li><b><tt>RegisterOpt</tt></b> - This template should be used when you are
590 registering a pass that logically should be available for use in the
591 '<tt>opt</tt>' utility.<p>
593 <li><b><tt>RegisterAnalysis</tt></b> - This template should be used when you are
594 registering a pass that logically should be available for use in the
595 '<tt>analysis</tt>' utility.<p>
597 <li><b><tt>RegisterLLC</tt></b> - This template should be used when you are
598 registering a pass that logically should be available for use in the
599 '<tt>llc</tt>' utility.<p>
601 <li><b><tt>RegisterPass</tt></b> - This is the generic form of the
602 <tt>Register*</tt> templates that should be used if you want your pass listed by
603 multiple or no utilities. This template takes an extra third argument that
604 specifies which tools it should be listed in. See the <a
605 href="http://llvm.cs.uiuc.edu/doxygen/PassSupport_8h-source.html">PassSupport.h</a>
606 file for more information.<p>
609 Regardless of how you register your pass, you must specify at least two
610 parameters. The first parameter is the name of the pass that is to be used on
611 the command line to specify that the pass should be added to a program (for
612 example <tt>opt</tt> or <tt>analyze</tt>). The second argument is the name of
613 the pass, which is to be used for the <tt>--help</tt> output of programs, as
614 well as for debug output generated by the <tt>--debug-pass</tt> option.<p>
616 If you pass is constructed by its default constructor, you only ever have to
617 pass these two arguments. If, on the other hand, you require other information
618 (like target specific information), you must pass an additional argument. This
619 argument is a pointer to a function used to create the pass. For an example of
620 how this works, look at the <a
621 href="http://llvm.cs.uiuc.edu/doxygen/LowerAllocations_8cpp-source.html">LowerAllocations.cpp</a>
624 If a pass is registered to be used by the <tt>analyze</tt> utility, you should
625 implement the virtual <tt>print</tt> method:<p>
627 <!-- _______________________________________________________________________ -->
628 </ul><h4><a name="print"><hr size=0>The <tt>print</tt> method</h4><ul>
631 <b>virtual void</b> print(std::ostream &O, <b>const</b> Module *M) <b>const</b>;
634 The <tt>print</tt> method must be implemented by "analyses" in order to print a
635 human readable version of the analysis results. This is useful for debugging an
636 analysis itself, as well as for other people to figure out how an analysis
637 works. The <tt>analyze</tt> tool uses this method to generate its output.<p>
639 The <tt>ostream</tt> parameter specifies the stream to write the results on, and
640 the <tt>Module</tt> parameter gives a pointer to the top level module of the
641 program that has been analyzed. Note however that this pointer may be null in
642 certain circumstances (such as calling the <tt>Pass::dump()</tt> from a
643 debugger), so it should only be used to enhance debug output, it should not be
647 <!-- *********************************************************************** -->
648 </ul><table width="100%" bgcolor="#330077" border=0 cellpadding=4 cellspacing=0>
649 <tr><td align=center><font color="#EEEEFF" size=+2 face="Georgia,Palatino"><b>
650 <a name="interaction">Specifying interactions between passes
651 </b></font></td></tr></table><ul>
652 <!-- *********************************************************************** -->
654 One of the main responsibilities of the <tt>PassManager</tt> is the make sure
655 that passes interact with each other correctly. Because <tt>PassManager</tt>
656 tries to <a href="#passmanager">optimize the execution of passes</a> it must
657 know how the passes interact with each other and what dependencies exist between
658 the various passes. To track this, each pass can declare the set of passes that
659 are required to be executed before the current pass, and the passes which are
660 invalidated by the current pass.<p>
662 Typically this functionality is used to require that analysis results are
663 computed before your pass is run. Running arbitrary transformation passes can
664 invalidate the computed analysis results, which is what the invalidation set
665 specifies. If a pass does not implement the <tt><a
666 href="#getAnalysisUsage">getAnalysisUsage</a></tt> method, it defaults to not
667 having any prerequisite passes, and invalidating <b>all</b> other passes.<p>
670 <!-- _______________________________________________________________________ -->
671 </ul><h4><a name="getAnalysisUsage"><hr size=0>The <tt>getAnalysisUsage</tt> method</h4><ul>
674 <b>virtual void</b> getAnalysisUsage(AnalysisUsage &Info) <b>const</b>;
677 By implementing the <tt>getAnalysisUsage</tt> method, the required and
678 invalidated sets may be specified for your transformation. The implementation
679 should fill in the <tt><a
680 href="http://llvm.cs.uiuc.edu/doxygen/classAnalysisUsage.html">AnalysisUsage</a></tt>
681 object with information about which passes are required and not invalidated. To do this, the following set methods are provided by the <tt><a
682 href="http://llvm.cs.uiuc.edu/doxygen/classAnalysisUsage.html">AnalysisUsage</a></tt> class:<p>
685 <i>// addRequires - Add the specified pass to the required set for your pass.</i>
686 <b>template</b><<b>class</b> PassClass>
687 AnalysisUsage &AnalysisUsage::addRequired();
689 <i>// addPreserved - Add the specified pass to the set of analyses preserved by
691 <b>template</b><<b>class</b> PassClass>
692 AnalysisUsage &AnalysisUsage::addPreserved();
694 <i>// setPreservesAll - Call this if the pass does not modify its input at all</i>
695 <b>void</b> AnalysisUsage::setPreservesAll();
697 <i>// preservesCFG - This function should be called by the pass, iff they do not:
699 // 1. Add or remove basic blocks from the function
700 // 2. Modify terminator instructions in any way.
702 // This is automatically implied for <a href="#BasicBlockPass">BasicBlockPass</a>'s
704 <b>void</b> AnalysisUsage::preservesCFG();
707 Some examples of how to use these methods are:<p>
710 <i>// This is an example implementation from an analysis, which does not modify
711 // the program at all, yet has a prerequisite.</i>
712 <b>void</b> <a href="http://llvm.cs.uiuc.edu/doxygen/structPostDominanceFrontier.html">PostDominanceFrontier</a>::getAnalysisUsage(AnalysisUsage &AU) <b>const</b> {
713 AU.setPreservesAll();
714 AU.addRequired<<a href="http://llvm.cs.uiuc.edu/doxygen/structPostDominatorTree.html">PostDominatorTree</a>>();
721 <i>// This example modifies the program, but does not modify the CFG</i>
722 <b>void</b> <a href="http://llvm.cs.uiuc.edu/doxygen/structLICM.html">LICM</a>::getAnalysisUsage(AnalysisUsage &AU) <b>const</b> {
724 AU.addRequired<<a href="http://llvm.cs.uiuc.edu/doxygen/classLoopInfo.html">LoopInfo</a>>();
728 <!-- _______________________________________________________________________ -->
729 </ul><h4><a name="getAnalysis"><hr size=0>The <tt>getAnalysis<></tt> method</h4><ul>
731 The <tt>Pass::getAnalysis<></tt> method is inherited by your class,
732 providing you with access to the passes that you declared that you required with
733 the <a href="#getAnalysisUsage"><tt>getAnalysisUsage</tt></a> method. It takes
734 a single template argument that specifies which pass class you want, and returns
735 a reference to that pass.<p>
738 <b>template</b><<b>typename</b> PassClass>
739 AnalysisType &getAnalysis();
742 This method call returns a reference to the pass desired. You may get a runtime
743 assertion failure if you attempt to get an analysis that you did not declare as
744 required in your <a href="#getAnalysisUsage"><tt>getAnalysisUsage</tt></a>
745 implementation. This method can be called by your <tt>run*</tt> method
746 implementation, or by any other local method invoked by your <tt>run*</tt>
749 <!-- *********************************************************************** -->
750 </ul><table width="100%" bgcolor="#330077" border=0 cellpadding=4 cellspacing=0>
751 <tr><td align=center><font color="#EEEEFF" size=+2 face="Georgia,Palatino"><b>
752 <a name="analysisgroup">Implementing Analysis Groups
753 </b></font></td></tr></table><ul>
754 <!-- *********************************************************************** -->
756 Now that we understand the basics of how passes are defined, how the are used,
757 and how they are required from other passes, it's time to get a little bit
758 fancier. All of the pass relationships that we have seen so far are very
759 simple: one pass depends on one other specific pass to be run before it can run.
760 For many applications, this is great, for others, more flexibility is
763 In particular, some analyses are defined such that there is a single simple
764 interface to the analysis results, but multiple ways of calculating them.
765 Consider alias analysis for example. The most trivial alias analysis returns
766 "may alias" for any alias query. The most sophisticated analysis a
767 flow-sensitive, context-sensitive interprocedural analysis that can take a
768 significant amount of time to execute (and obviously, there is a lot of room
769 between these two extremes for other implementations). To cleanly support
770 situations like this, the LLVM Pass Infrastructure supports the notion of
773 <!-- _______________________________________________________________________ -->
774 </ul><h4><a name="agconcepts"><hr size=0>Analysis Group Concepts</h4><ul>
776 An Analysis Group is a single simple interface that may be implemented by
777 multiple different passes. Analysis Groups can be given human readable names
778 just like passes, but unlike passes, they need not derive from the <tt>Pass</tt>
779 class. An analysis group may have one or more implementations, one of which is
780 the "default" implementation.<p>
782 Analysis groups are used by client passes just like other passes are: the
783 <tt>AnalysisUsage::addRequired()</tt> and <tt>Pass::getAnalysis()</tt> methods.
784 In order to resolve this requirement, the <a href="#passmanager">PassManager</a>
785 scans the available passes to see if any implementations of the analysis group
786 are available. If none is available, the default implementation is created for
787 the pass to use. All standard rules for <A href="#interaction">interaction
788 between passes</a> still apply.<p>
790 Although <a href="#registration">Pass Registration</a> is optional for normal
791 passes, all analysis group implementations must be registered, and must use the
792 <A href="#registerag"><tt>RegisterAnalysisGroup</tt></a> template to join the
793 implementation pool. Also, a default implementation of the interface
794 <b>must</b> be registered with <A
795 href="#registerag"><tt>RegisterAnalysisGroup</tt></a>.<p>
797 As a concrete example of an Analysis Group in action, consider the <a
798 href="http://llvm.cs.uiuc.edu/doxygen/structAliasAnalysis.html">AliasAnalysis</a>
799 analysis group. The default implementation of the alias analysis interface (the
801 href="http://llvm.cs.uiuc.edu/doxygen/structBasicAliasAnalysis.html">basicaa</a></tt>
802 pass) just does a few simple checks that don't require significant analysis to
803 compute (such as: two different globals can never alias each other, etc).
804 Passes that use the <tt><a
805 href="http://llvm.cs.uiuc.edu/doxygen/structAliasAnalysis.html">AliasAnalysis</a></tt>
806 interface (for example the <tt><a
807 href="http://llvm.cs.uiuc.edu/doxygen/classGCSE.html">gcse</a></tt> pass), do not care which implementation
808 of alias analysis is actually provided, they just use the designated
811 From the user's perspective, commands work just like normal. Issuing the
812 command '<tt>opt -gcse ...</tt>' will cause the <tt>basicaa</tt> class to be
813 instantiated and added to the pass sequence. Issuing the command '<tt>opt
814 -somefancyaa -gcse ...</tt>' will cause the <tt>gcse</tt> pass to use the
815 <tt>somefancyaa</tt> alias analysis (which doesn't actually exist, it's just a
816 hypothetical example) instead.<p>
819 <!-- _______________________________________________________________________ -->
820 </ul><h4><a name="registerag"><hr size=0>Using <tt>RegisterAnalysisGroup</tt></h4><ul>
822 The <tt>RegisterAnalysisGroup</tt> template is used to register the analysis
823 group itself as well as add pass implementations to the analysis group. First,
824 an analysis should be registered, with a human readable name provided for it.
825 Unlike registration of passes, there is no command line argument to be specified
826 for the Analysis Group Interface itself, because it is "abstract":<p>
829 <b>static</b> RegisterAnalysisGroup<<a href="http://llvm.cs.uiuc.edu/doxygen/structAliasAnalysis.html">AliasAnalysis</a>> A("<i>Alias Analysis</i>");
832 Once the analysis is registered, passes can declare that they are valid
833 implementations of the interface by using the following code:<p>
837 //<i> Analysis Group implementations <b>must</b> be registered normally...</i>
838 RegisterOpt<FancyAA>
839 B("<i>somefancyaa</i>", "<i>A more complex alias analysis implementation</i>");
841 //<i> Declare that we implement the AliasAnalysis interface</i>
842 RegisterAnalysisGroup<<a href="http://llvm.cs.uiuc.edu/doxygen/structAliasAnalysis.html">AliasAnalysis</a>, FancyAA> C;
846 This just shows a class <tt>FancyAA</tt> that is registered normally, then uses
847 the <tt>RegisterAnalysisGroup</tt> template to "join" the <tt><a
848 href="http://llvm.cs.uiuc.edu/doxygen/structAliasAnalysis.html">AliasAnalysis</a></tt>
849 analysis group. Every implementation of an analysis group should join using
850 this template. A single pass may join multiple different analysis groups with
855 //<i> Analysis Group implementations <b>must</b> be registered normally...</i>
856 RegisterOpt<<a href="http://llvm.cs.uiuc.edu/doxygen/structBasicAliasAnalysis.html">BasicAliasAnalysis</a>>
857 D("<i>basicaa</i>", "<i>Basic Alias Analysis (default AA impl)</i>");
859 //<i> Declare that we implement the AliasAnalysis interface</i>
860 RegisterAnalysisGroup<<a href="http://llvm.cs.uiuc.edu/doxygen/structAliasAnalysis.html">AliasAnalysis</a>, <a href="http://llvm.cs.uiuc.edu/doxygen/structBasicAliasAnalysis.html">BasicAliasAnalysis</a>, <b>true</b>> E;
864 Here we show how the default implementation is specified (using the extra
865 argument to the <tt>RegisterAnalysisGroup</tt> template). There must be exactly
866 one default implementation available at all times for an Analysis Group to be
867 used. Here we declare that the <tt><a
868 href="http://llvm.cs.uiuc.edu/doxygen/structBasicAliasAnalysis.html">BasicAliasAnalysis</a></tt>
869 pass is the default implementation for the interface.<p>
872 <!-- *********************************************************************** -->
873 </ul><table width="100%" bgcolor="#330077" border=0 cellpadding=4 cellspacing=0>
874 <tr><td align=center><font color="#EEEEFF" size=+2 face="Georgia,Palatino"><b>
875 <a name="passmanager">What PassManager does
876 </b></font></td></tr></table><ul>
877 <!-- *********************************************************************** -->
880 href="http://llvm.cs.uiuc.edu/doxygen/PassManager_8h-source.html"><tt>PassManager</tt></a>
881 <a href="http://llvm.cs.uiuc.edu/doxygen/classPassManager.html">class</a> takes
882 a list of passes, ensures their <a href="#interaction">prerequisites</a> are set
883 up correctly, and then schedules passes to run efficiently. All of the LLVM
884 tools that run passes use the <tt>PassManager</tt> for execution of these
887 The <tt>PassManager</tt> does two main things to try to reduce the execution
888 time of a series of passes:<p>
891 <li><b>Share analysis results</b> - The PassManager attempts to avoid
892 recomputing analysis results as much as possible. This means keeping track of
893 which analyses are available already, which analyses get invalidated, and which
894 analyses are needed to be run for a pass. An important part of work is that the
895 <tt>PassManager</tt> tracks the exact lifetime of all analysis results, allowing
896 it to <a href="#releaseMemory">free memory</a> allocated to holding analysis
897 results as soon as they are no longer needed.<p>
899 <li><b>Pipeline the execution of passes on the program</b> - The
900 <tt>PassManager</tt> attempts to get better cache and memory usage behavior out
901 of a series of passes by pipelining the passes together. This means that, given
902 a series of consequtive <a href="#FunctionPass"><tt>FunctionPass</tt></a>'s, it
903 will execute all of the <a href="#FunctionPass"><tt>FunctionPass</tt></a>'s on
904 the first function, then all of the <a
905 href="#FunctionPass"><tt>FunctionPass</tt></a>'s on the second function,
906 etc... until the entire program has been run through the passes.<p>
908 This improves the cache behavior of the compiler, because it is only touching
909 the LLVM program representation for a single function at a time, instead of
910 traversing the entire program. It reduces the memory consumption of compiler,
911 because, for example, only one <a
912 href="http://llvm.cs.uiuc.edu/doxygen/structDominatorSet.html"><tt>DominatorSet</tt></a>
913 needs to be calculated at a time. This also makes it possible some <a
914 href="#SMP">interesting enhancements</a> in the future.<p>
918 The effectiveness of the <tt>PassManager</tt> is influenced directly by how much
919 information it has about the behaviors of the passes it is scheduling. For
920 example, the "preserved" set is intentionally conservative in the face of an
921 unimplemented <a href="#getAnalysisUsage"><tt>getAnalysisUsage</tt></a> method.
922 Not implementing when it should be implemented will have the effect of not
923 allowing any analysis results to live across the execution of your pass.<p>
925 The <tt>PassManager</tt> class exposes a <tt>--debug-pass</tt> command line
926 options that is useful for debugging pass execution, seeing how things work, and
927 diagnosing when you should be preserving more analyses than you currently are
928 (To get information about all of the variants of the <tt>--debug-pass</tt>
929 option, just type '<tt>opt --help-hidden</tt>').<p>
931 By using the <tt>--debug-pass=Structure</tt> option, for example, we can see how
932 our <a href="#basiccode">Hello World</a> pass interacts with other passes. Lets
933 try it out with the <tt>gcse</tt> and <tt>licm</tt> passes:<p>
936 $ opt -load ../../../lib/Debug/libhello.so -gcse -licm --debug-pass=Structure < hello.bc > /dev/null
938 Function Pass Manager
939 Dominator Set Construction
940 Immediate Dominators Construction
941 Global Common Subexpression Elimination
942 -- Immediate Dominators Construction
943 -- Global Common Subexpression Elimination
944 Natural Loop Construction
945 Loop Invariant Code Motion
946 -- Natural Loop Construction
947 -- Loop Invariant Code Motion
949 -- Dominator Set Construction
955 This output shows us when passes are constructed and when the analysis results
956 are known to be dead (prefixed with '<tt>--</tt>'). Here we see that GCSE uses
957 dominator and immediate dominator information to do its job. The LICM pass uses
958 natural loop information, which uses dominator sets, but not immediate
959 dominators. Because immediate dominators are no longer useful after the GCSE
960 pass, it is immediately destroyed. The dominator sets are then reused to
961 compute natural loop information, which is then used by the LICM pass.<p>
963 After the LICM pass, the module verifier runs (which is automatically added by
964 the '<tt>opt</tt>' tool), which uses the dominator set to check that the
965 resultant LLVM code is well formed. After it finishes, the dominator set
966 information is destroyed, after being computed once, and shared by three
969 Lets see how this changes when we run the <a href="#basiccode">Hello World</a>
970 pass in between the two passes:<p>
973 $ opt -load ../../../lib/Debug/libhello.so -gcse -hello -licm --debug-pass=Structure < hello.bc > /dev/null
975 Function Pass Manager
976 Dominator Set Construction
977 Immediate Dominators Construction
978 Global Common Subexpression Elimination
979 <b>-- Dominator Set Construction</b>
980 -- Immediate Dominators Construction
981 -- Global Common Subexpression Elimination
984 Dominator Set Construction</b>
985 Natural Loop Construction
986 Loop Invariant Code Motion
987 -- Natural Loop Construction
988 -- Loop Invariant Code Motion
990 -- Dominator Set Construction
999 Here we see that the <a href="#basiccode">Hello World</a> pass has killed the
1000 Dominator Set pass, even though it doesn't modify the code at all! To fix this,
1001 we need to add the following <a
1002 href="#getAnalysisUsage"><tt>getAnalysisUsage</tt></a> method to our pass:<p>
1005 <i>// We don't modify the program, so we preserve all analyses</i>
1006 <b>virtual void</b> getAnalysisUsage(AnalysisUsage &AU) <b>const</b> {
1007 AU.setPreservesAll();
1011 Now when we run our pass, we get this output:<p>
1014 $ opt -load ../../../lib/Debug/libhello.so -gcse -hello -licm --debug-pass=Structure < hello.bc > /dev/null
1015 Pass Arguments: -gcse -hello -licm
1017 Function Pass Manager
1018 Dominator Set Construction
1019 Immediate Dominators Construction
1020 Global Common Subexpression Elimination
1021 -- Immediate Dominators Construction
1022 -- Global Common Subexpression Elimination
1025 Natural Loop Construction
1026 Loop Invariant Code Motion
1027 -- Loop Invariant Code Motion
1028 -- Natural Loop Construction
1030 -- Dominator Set Construction
1039 Which shows that we don't accidentally invalidate dominator information
1040 anymore, and therefore do not have to compute it twice.<p>
1043 <!-- _______________________________________________________________________ -->
1044 </ul><h4><a name="releaseMemory"><hr size=0>The <tt>releaseMemory</tt> method</h4><ul>
1047 <b>virtual void</b> releaseMemory();
1050 The <tt>PassManager</tt> automatically determines when to compute analysis
1051 results, and how long to keep them around for. Because the lifetime of the pass
1052 object itself is effectively the entire duration of the compilation process, we
1053 need some way to free analysis results when they are no longer useful. The
1054 <tt>releaseMemory</tt> virtual method is the way to do this.<p>
1056 If you are writing an analysis or any other pass that retains a significant
1057 amount of state (for use by another pass which "requires" your pass and uses the
1058 <a href="#getAnalysis">getAnalysis</a> method) you should implement
1059 <tt>releaseMEmory</tt> to, well, release the memory allocated to maintain this
1060 internal state. This method is called after the <tt>run*</tt> method for the
1061 class, before the next call of <tt>run*</tt> in your pass.<p>
1064 <!-- *********************************************************************** -->
1065 </ul><table width="100%" bgcolor="#330077" border=0 cellpadding=4 cellspacing=0>
1066 <tr><td align=center><font color="#EEEEFF" size=+2 face="Georgia,Palatino"><b>
1067 <a name="debughints">Using GDB with dynamically loaded passes
1068 </b></font></td></tr></table><ul>
1069 <!-- *********************************************************************** -->
1071 Unfortunately, using GDB with dynamically loaded passes is not as easy as it
1072 should be. First of all, you can't set a breakpoint in a shared object that has
1073 not been loaded yet, and second of all there are problems with inlined functions
1074 in shared objects. Here are some suggestions to debugging your pass with
1077 For sake of discussion, I'm going to assume that you are debugging a
1078 transformation invoked by <tt>opt</tt>, although nothing described here depends
1081 <!-- _______________________________________________________________________ -->
1082 </ul><h4><a name="breakpoint"><hr size=0>Setting a breakpoint in your pass</h4><ul>
1084 First thing you do is start <tt>gdb</tt> on the <tt>opt</tt> process:<p>
1089 Copyright 2000 Free Software Foundation, Inc.
1090 GDB is free software, covered by the GNU General Public License, and you are
1091 welcome to change it and/or distribute copies of it under certain conditions.
1092 Type "show copying" to see the conditions.
1093 There is absolutely no warranty for GDB. Type "show warranty" for details.
1094 This GDB was configured as "sparc-sun-solaris2.6"...
1098 Note that <tt>opt</tt> has a lot of debugging information in it, so it takes
1099 time to load. Be patient. Since we cannot set a breakpoint in our pass yet
1100 (the shared object isn't loaded until runtime), we must execute the process, and
1101 have it stop before it invokes our pass, but after it has loaded the shared
1102 object. The most foolproof way of doing this is to set a breakpoint in
1103 <tt>PassManager::run</tt> and then run the process with the arguments you
1107 (gdb) <b>break PassManager::run</b>
1108 Breakpoint 1 at 0x2413bc: file Pass.cpp, line 70.
1109 (gdb) <b>run test.bc -load /shared/lattner/cvs/llvm/lib/Debug/[libname].so -[passoption]</b>
1110 Starting program: /shared/lattner/cvs/llvm/tools/Debug/opt test.bc
1111 -load /shared/lattner/cvs/llvm/lib/Debug/[libname].so -[passoption]
1112 Breakpoint 1, PassManager::run (this=0xffbef174, M=@0x70b298) at Pass.cpp:70
1113 70 bool PassManager::run(Module &M) { return PM->run(M); }
1117 Once the <tt>opt</tt> stops in the <tt>PassManager::run</tt> method you are now
1118 free to set breakpoints in your pass so that you can trace through execution or
1119 do other standard debugging stuff.<p>
1122 <!-- _______________________________________________________________________ -->
1123 </ul><h4><a name="debugmisc"><hr size=0>Miscellaneous Problems</h4><ul>
1125 Once you have the basics down, there are a couple of problems that GDB has, some
1126 with solutions, some without.<p>
1129 <li>Inline functions have bogus stack information. In general, GDB does a
1130 pretty good job getting stack traces and stepping through inline functions.
1131 When a pass is dynamically loaded however, it somehow completely loses this
1132 capability. The only solution I know of is to de-inline a function (move it
1133 from the body of a class to a .cpp file).<p>
1135 <li>Restarting the program breaks breakpoints. After following the information
1136 above, you have succeeded in getting some breakpoints planted in your pass. Nex
1137 thing you know, you restart the program (i.e., you type '<tt>run</tt>' again),
1138 and you start getting errors about breakpoints being unsettable. The only way I
1139 have found to "fix" this problem is to <tt>delete</tt> the breakpoints that are
1140 already set in your pass, run the program, and re-set the breakpoints once
1141 execution stops in <tt>PassManager::run</tt>.<p>
1145 Hopefully these tips will help with common case debugging situations. If you'd
1146 like to contribute some tips of your own, just contact <a
1147 href="mailto:sabre@nondot.org">Chris</a>.<p>
1150 <!-- *********************************************************************** -->
1151 </ul><table width="100%" bgcolor="#330077" border=0 cellpadding=4 cellspacing=0>
1152 <tr><td align=center><font color="#EEEEFF" size=+2 face="Georgia,Palatino"><b>
1153 <a name="future">Future extensions planned
1154 </b></font></td></tr></table><ul>
1155 <!-- *********************************************************************** -->
1157 Although the LLVM Pass Infrastructure is very capable as it stands, and does
1158 some nifty stuff, there are things we'd like to add in the future. Here is
1159 where we are going:<p>
1161 <!-- _______________________________________________________________________ -->
1162 </ul><h4><a name="SMP"><hr size=0>Multithreaded LLVM</h4><ul>
1164 Multiple CPU machines are becoming more common and compilation can never be
1165 fast enough: obviously we should allow for a multithreaded compiler. Because of
1166 the semantics defined for passes above (specifically they cannot maintain state
1167 across invocations of their <tt>run*</tt> methods), a nice clean way to
1168 implement a multithreaded compiler would be for the <tt>PassManager</tt> class
1169 to create multiple instances of each pass object, and allow the seperate
1170 instances to be hacking on different parts of the program at the same time.<p>
1172 This implementation would prevent each of the passes from having to implement
1173 multithreaded constructs, requiring only the LLVM core to have locking in a few
1174 places (for global resources). Although this is a simple extension, we simply
1175 haven't had time (or multiprocessor machines, thus a reason) to implement this.
1176 Despite that, we have kept the LLVM passes SMP ready, and you should too.<p>
1179 <!-- _______________________________________________________________________ -->
1180 </ul><h4><a name="ModuleSource"><hr size=0>A new <tt>ModuleSource</tt> interface</h4><ul>
1182 Currently, the <tt>PassManager</tt>'s <tt>run</tt> method takes a <tt><a
1183 href="http://llvm.cs.uiuc.edu/doxygen/classModule.html">Module</a></tt> as
1184 input, and runs all of the passes on this module. The problem with this
1185 approach is that none of the <tt>PassManager</tt> features can be used for
1186 timing and debugging the actual <b>loading</b> of the module from disk or
1189 To solve this problem, eventually the <tt>PassManger</tt> class will accept a
1190 <tt>ModuleSource</tt> object instead of a Module itself. When complete, this
1191 will also allow for streaming of functions out of the bytecode representation,
1192 allowing us to avoid holding the entire program in memory at once if we only are
1193 dealing with <a href="#FunctionPass">FunctionPass</a>'s.<p>
1195 As part of a different issue, eventually the bytecode loader will be extended to
1196 allow on-demand loading of functions from the bytecode representation, in order
1197 to better support the runtime reoptimizer. The bytecode format is already
1198 capable of this, the loader just needs to be reworked a bit.<p>
1201 <!-- _______________________________________________________________________ -->
1202 </ul><h4><a name="PassFunctionPass"><hr size=0><tt>Pass</tt>'s requiring <tt>FunctionPass</tt>'s</h4><ul>
1204 Currently it is illegal for a <a href="#Pass"><tt>Pass</tt></a> to require a <a
1205 href="#FunctionPass"><tt>FunctionPass</tt></a>. This is because there is only
1206 one instance of the <a href="#FunctionPass"><tt>FunctionPass</tt></a> object
1207 ever created, thus nowhere to store information for all of the functions in the
1208 program at the same time. Although this has come up a couple of times before,
1209 this has always been worked around by factoring one big complicated pass into a
1210 global and an interprocedural part, both of which are distinct. In the future,
1211 it would be nice to have this though.<p>
1213 Note that it is no problem for a <a
1214 href="#FunctionPass"><tt>FunctionPass</tt></a> to require the results of a <a
1215 href="#Pass"><tt>Pass</tt></a>, only the other way around.<p>
1218 <!-- *********************************************************************** -->
1220 <!-- *********************************************************************** -->
1223 <address><a href="mailto:sabre@nondot.org">Chris Lattner</a></address>
1224 <!-- Created: Tue Aug 6 15:00:33 CDT 2002 -->
1225 <!-- hhmts start -->
1226 Last modified: Wed Sep 25 17:20:10 CDT 2002
1228 </font></body></html>