1 <!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.01//EN"
2 "http://www.w3.org/TR/html4/strict.dtd">
5 <title>Writing an LLVM Pass</title>
6 <link rel="stylesheet" href="llvm.css" type="text/css">
10 <div class="doc_title">
15 <li><a href="#introduction">Introduction - What is a pass?</a></li>
16 <li><a href="#quickstart">Quick Start - Writing hello world</a>
18 <li><a href="#makefile">Setting up the build environment</a></li>
19 <li><a href="#basiccode">Basic code required</a></li>
20 <li><a href="#running">Running a pass with <tt>opt</tt>
21 or <tt>analyze</tt></a></li>
23 <li><a href="#passtype">Pass classes and requirements</a>
25 <li><a href="#ImmutablePass">The <tt>ImmutablePass</tt> class</a></li>
26 <li><a href="#Pass">The <tt>Pass</tt> class</a>
28 <li><a href="#run">The <tt>run</tt> method</a></li>
30 <li><a href="#FunctionPass">The <tt>FunctionPass</tt> class</a>
32 <li><a href="#doInitialization_mod">The <tt>doInitialization(Module
33 &)</tt> method</a></li>
34 <li><a href="#runOnFunction">The <tt>runOnFunction</tt> method</a></li>
35 <li><a href="#doFinalization_mod">The <tt>doFinalization(Module
36 &)</tt> method</a></li>
38 <li><a href="#BasicBlockPass">The <tt>BasicBlockPass</tt> class</a>
40 <li><a href="#doInitialization_fn">The <tt>doInitialization(Function
41 &)</tt> method</a></li>
42 <li><a href="#runOnBasicBlock">The <tt>runOnBasicBlock</tt>
44 <li><a href="#doFinalization_fn">The <tt>doFinalization(Function
45 &)</tt> method</a></li>
47 <li><a href="#MachineFunctionPass">The <tt>MachineFunctionPass</tt>
50 <li><a href="#runOnMachineFunction">The
51 <tt>runOnMachineFunction(MachineFunction &)</tt> method</a></li>
54 <li><a href="#registration">Pass Registration</a>
56 <li><a href="#print">The <tt>print</tt> method</a></li>
58 <li><a href="#interaction">Specifying interactions between passes</a>
60 <li><a href="#getAnalysisUsage">The <tt>getAnalysisUsage</tt>
62 <li><a href="#getAnalysis">The <tt>getAnalysis</tt> method</a></li>
64 <li><a href="#analysisgroup">Implementing Analysis Groups</a>
66 <li><a href="#agconcepts">Analysis Group Concepts</a></li>
67 <li><a href="#registerag">Using <tt>RegisterAnalysisGroup</tt></a></li>
69 <li><a href="#passmanager">What PassManager does</a>
71 <li><a href="#releaseMemory">The <tt>releaseMemory</tt> method</a></li>
73 <li><a href="#debughints">Using GDB with dynamically loaded passes</a>
75 <li><a href="#breakpoint">Setting a breakpoint in your pass</a></li>
76 <li><a href="#debugmisc">Miscellaneous Problems</a></li>
78 <li><a href="#future">Future extensions planned</a>
80 <li><a href="#SMP">Multithreaded LLVM</a></li>
81 <li><a href="#ModuleSource">A new <tt>ModuleSource</tt> interface</a></li>
82 <li><a href="#PassFunctionPass"><tt>Pass</tt>es requiring
83 <tt>FunctionPass</tt>es</a></li>
87 <div class="doc_text">
88 <p><b>Written by <a href="mailto:sabre@nondot.org">Chris Lattner</a></b><p>
91 <!-- *********************************************************************** -->
92 <div class="doc_section">
93 <a name="introduction">Introduction - What is a pass?</a>
95 <!-- *********************************************************************** -->
97 <div class="doc_text">
99 <p>The LLVM Pass Framework is an important part of the LLVM system, because LLVM
100 passes are where the interesting parts of the compiler exist. Passes perform
101 the transformations and optimizations that make up the compiler, they build
102 the analysis results that are used by these transformations, and they are, above
103 all, a structuring technique for compiler code.</p>
105 <p>All LLVM passes are subclasses of the <tt><a
106 href="http://llvm.cs.uiuc.edu/doxygen/classllvm_1_1Pass.html">Pass</a></tt>
107 class, which implement functionality by overriding virtual methods inherited
108 from <tt>Pass</tt>. Depending on how your pass works, you may be able to
109 inherit from the <tt><a
110 href="http://llvm.cs.uiuc.edu/doxygen/structllvm_1_1FunctionPass.html">FunctionPass</a></tt>
112 href="http://llvm.cs.uiuc.edu/doxygen/structllvm_1_1BasicBlockPass.html">BasicBlockPass</a></tt>,
113 which gives the system more information about what your pass does, and how it
114 can be combined with other passes. One of the main features of the LLVM Pass
115 Framework is that it schedules passes to run in an efficient way based on the
116 constraints that your pass has.</p>
118 <p>We start by showing you how to construct a pass, everything from setting up
119 the code, to compiling, loading, and executing it. After the basics are down,
120 more advanced features are discussed.</p>
124 <!-- *********************************************************************** -->
125 <div class="doc_section">
126 <a name="quickstart">Quick Start - Writing hello world</a>
128 <!-- *********************************************************************** -->
130 <div class="doc_text">
132 <p>Here we describe how to write the "hello world" of passes. The "Hello" pass
133 is designed to simply print out the name of non-external functions that exist in
134 the program being compiled. It does not modify the program at all, just
135 inspects it. The source code and files for this pass are available in the LLVM
136 source tree in the <tt>lib/Transforms/Hello</tt> directory.</p>
140 <!-- ======================================================================= -->
141 <div class="doc_subsection">
142 <a name="makefile">Setting up the build environment</a>
145 <div class="doc_text">
147 <p>First thing you need to do is create a new directory somewhere in the LLVM
148 source base. For this example, we'll assume that you made
149 "<tt>lib/Transforms/Hello</tt>". The first thing you must do is set up a build
150 script (Makefile) that will compile the source code for the new pass. To do
151 this, copy this into "<tt>Makefile</tt>":</p>
156 # Makefile for hello pass
158 # Path to top level of LLVM heirarchy
161 # Name of the library to build
164 # Build a dynamically loadable shared object
167 # Include the makefile implementation stuff
168 include $(LEVEL)/Makefile.common
171 <p>This makefile specifies that all of the <tt>.cpp</tt> files in the current
172 directory are to be compiled and linked together into a
173 <tt>lib/Debug/libhello.so</tt> shared object that can be dynamically loaded by
174 the <tt>opt</tt> or <tt>analyze</tt> tools.</p>
176 <p>Now that we have the build scripts set up, we just need to write the code for
181 <!-- ======================================================================= -->
182 <div class="doc_subsection">
183 <a name="basiccode">Basic code required</a>
186 <div class="doc_text">
188 <p>Now that we have a way to compile our new pass, we just have to write it.
192 <b>#include</b> "<a href="http://llvm.cs.uiuc.edu/doxygen/Pass_8h-source.html">llvm/Pass.h</a>"
193 <b>#include</b> "<a href="http://llvm.cs.uiuc.edu/doxygen/Function_8h-source.html">llvm/Function.h</a>"
196 <p>Which are needed because we are writing a <tt><a
197 href="http://llvm.cs.uiuc.edu/doxygen/classllvm_1_1Pass.html">Pass</a></tt>, and
198 we are operating on <tt><a
199 href="http://llvm.cs.uiuc.edu/doxygen/classllvm_1_1Function.html">Function</a></tt>'s.</p>
207 <p>... which starts out an anonymous namespace. Anonymous namespaces are to C++
208 what the "<tt>static</tt>" keyword is to C (at global scope). It makes the
209 things declared inside of the anonymous namespace only visible to the current
210 file. If you're not familiar with them, consult a decent C++ book for more
213 <p>Next, we declare our pass itself:</p>
216 <b>struct</b> Hello : <b>public</b> <a href="#FunctionPass">FunctionPass</a> {
219 <p>This declares a "<tt>Hello</tt>" class that is a subclass of <tt><a
220 href="http://llvm.cs.uiuc.edu/doxygen/structllvm_1_1FunctionPass.html">FunctionPass</a></tt>.
221 The different builtin pass subclasses are described in detail <a
222 href="#passtype">later</a>, but for now, know that <a
223 href="#FunctionPass"><tt>FunctionPass</tt></a>'s operate a function at a
227 <b>virtual bool</b> <a href="#runOnFunction">runOnFunction</a>(Function &F) {
228 std::cerr << "<i>Hello: </i>" << F.getName() << "\n";
231 }; <i>// end of struct Hello</i>
234 <p>We declare a "<a href="#runOnFunction"><tt>runOnFunction</tt></a>" method,
235 which overloads an abstract virtual method inherited from <a
236 href="#FunctionPass"><tt>FunctionPass</tt></a>. This is where we are supposed
237 to do our thing, so we just print out our message with the name of each
241 RegisterOpt<Hello> X("<i>hello</i>", "<i>Hello World Pass</i>");
242 } <i>// end of anonymous namespace</i>
245 <p>Lastly, we register our class <tt>Hello</tt>, giving it a command line
246 argument "<tt>hello</tt>", and a name "<tt>Hello World Pass</tt>". There are
247 several different ways of <a href="#registration">registering your pass</a>,
248 depending on what it is to be used for. For "optimizations" we use the
249 <tt>RegisterOpt</tt> template.</p>
251 <p>As a whole, the <tt>.cpp</tt> file looks like:</p>
254 <b>#include</b> "<a href="http://llvm.cs.uiuc.edu/doxygen/Pass_8h-source.html">llvm/Pass.h</a>"
255 <b>#include</b> "<a href="http://llvm.cs.uiuc.edu/doxygen/Function_8h-source.html">llvm/Function.h</a>"
258 <b>struct Hello</b> : <b>public</b> <a href="#FunctionPass">FunctionPass</a> {
259 <b>virtual bool</b> <a href="#runOnFunction">runOnFunction</a>(Function &F) {
260 std::cerr << "<i>Hello: </i>" << F.getName() << "\n";
265 RegisterOpt<Hello> X("<i>hello</i>", "<i>Hello World Pass</i>");
269 <p>Now that it's all together, compile the file with a simple "<tt>gmake</tt>"
270 command in the local directory and you should get a new
271 "<tt>lib/Debug/libhello.so</tt> file. Note that everything in this file is
272 contained in an anonymous namespace: this reflects the fact that passes are self
273 contained units that do not need external interfaces (although they can have
274 them) to be useful.</p>
278 <!-- ======================================================================= -->
279 <div class="doc_subsection">
280 <a name="running">Running a pass with <tt>opt</tt> or <tt>analyze</tt></a>
283 <div class="doc_text">
285 <p>Now that you have a brand new shiny <tt>.so</tt> file, we can use the
286 <tt>opt</tt> command to run an LLVM program through your pass. Because you
287 registered your pass with the <tt>RegisterOpt</tt> template, you will be able to
288 use the <tt>opt</tt> tool to access it, once loaded.</p>
290 <p>To test it, follow the example at the end of the <a
291 href="GettingStarted.html">Getting Started Guide</a> to compile "Hello World" to
292 LLVM. We can now run the bytecode file (<tt>hello.bc</tt>) for the program
293 through our transformation like this (or course, any bytecode file will
297 $ opt -load ../../../lib/Debug/libhello.so -hello < hello.bc > /dev/null
303 <p>The '<tt>-load</tt>' option specifies that '<tt>opt</tt>' should load your
304 pass as a shared object, which makes '<tt>-hello</tt>' a valid command line
305 argument (which is one reason you need to <a href="#registration">register your
306 pass</a>). Because the hello pass does not modify the program in any
307 interesting way, we just throw away the result of <tt>opt</tt> (sending it to
308 <tt>/dev/null</tt>).</p>
310 <p>To see what happened to the other string you registered, try running
311 <tt>opt</tt> with the <tt>--help</tt> option:</p>
314 $ opt -load ../../../lib/Debug/libhello.so --help
315 OVERVIEW: llvm .bc -> .bc modular optimizer
317 USAGE: opt [options] <input bytecode>
320 Optimizations available:
322 -funcresolve - Resolve Functions
323 -gcse - Global Common Subexpression Elimination
324 -globaldce - Dead Global Elimination
325 <b>-hello - Hello World Pass</b>
326 -indvars - Canonicalize Induction Variables
327 -inline - Function Integration/Inlining
328 -instcombine - Combine redundant instructions
332 <p>The pass name get added as the information string for your pass, giving some
333 documentation to users of <tt>opt</tt>. Now that you have a working pass, you
334 would go ahead and make it do the cool transformations you want. Once you get
335 it all working and tested, it may become useful to find out how fast your pass
336 is. The <a href="#passManager"><tt>PassManager</tt></a> provides a nice command
337 line option (<tt>--time-passes</tt>) that allows you to get information about
338 the execution time of your pass along with the other passes you queue up. For
342 $ opt -load ../../../lib/Debug/libhello.so -hello -time-passes < hello.bc > /dev/null
346 ===============================================================================
347 ... Pass execution timing report ...
348 ===============================================================================
349 Total Execution Time: 0.02 seconds (0.0479059 wall clock)
351 ---User Time--- --System Time-- --User+System-- ---Wall Time--- --- Pass Name ---
352 0.0100 (100.0%) 0.0000 ( 0.0%) 0.0100 ( 50.0%) 0.0402 ( 84.0%) Bytecode Writer
353 0.0000 ( 0.0%) 0.0100 (100.0%) 0.0100 ( 50.0%) 0.0031 ( 6.4%) Dominator Set Construction
354 0.0000 ( 0.0%) 0.0000 ( 0.0%) 0.0000 ( 0.0%) 0.0013 ( 2.7%) Module Verifier
355 <b> 0.0000 ( 0.0%) 0.0000 ( 0.0%) 0.0000 ( 0.0%) 0.0033 ( 6.9%) Hello World Pass</b>
356 0.0100 (100.0%) 0.0100 (100.0%) 0.0200 (100.0%) 0.0479 (100.0%) TOTAL
359 <p>As you can see, our implementation above is pretty fast :). The additional
360 passes listed are automatically inserted by the '<tt>opt</tt>' tool to verify
361 that the LLVM emitted by your pass is still valid and well formed LLVM, which
362 hasn't been broken somehow.</p>
364 <p>Now that you have seen the basics of the mechanics behind passes, we can talk
365 about some more details of how they work and how to use them.</p>
369 <!-- *********************************************************************** -->
370 <div class="doc_section">
371 <a name="passtype">Pass classes and requirements</a>
373 <!-- *********************************************************************** -->
375 <div class="doc_text">
377 <p>One of the first things that you should do when designing a new pass is to
378 decide what class you should subclass for your pass. The <a
379 href="#basiccode">Hello World</a> example uses the <tt><a
380 href="#FunctionPass">FunctionPass</a></tt> class for its implementation, but we
381 did not discuss why or when this should occur. Here we talk about the classes
382 available, from the most general to the most specific.</p>
384 <p>When choosing a superclass for your Pass, you should choose the <b>most
385 specific</b> class possible, while still being able to meet the requirements
386 listed. This gives the LLVM Pass Infrastructure information necessary to
387 optimize how passes are run, so that the resultant compiler isn't unneccesarily
392 <!-- ======================================================================= -->
393 <div class="doc_subsection">
394 <a name="ImmutablePass">The <tt>ImmutablePass</tt> class</a>
397 <div class="doc_text">
399 <p>The most plain and boring type of pass is the "<tt><a
400 href="http://llvm.cs.uiuc.edu/doxygen/structllvm_1_1ImmutablePass.html">ImmutablePass</a></tt>"
401 class. This pass type is used for passes that do not have to be run, do not
402 change state, and never need to be updated. This is not a normal type of
403 transformation or analysis, but can provide information about the current
404 compiler configuration.</p>
406 <p>Although this pass class is very infrequently used, it is important for
407 providing information about the current target machine being compiled for, and
408 other static information that can affect the various transformations.</p>
410 <p><tt>ImmutablePass</tt>es never invalidate other transformations, are never
411 invalidated, and are never "run".</p>
415 <!-- ======================================================================= -->
416 <div class="doc_subsection">
417 <a name="Pass">The <tt>Pass</tt> class</a>
420 <div class="doc_text">
423 href="http://llvm.cs.uiuc.edu/doxygen/classllvm_1_1Pass.html">Pass</a></tt>"
424 class is the most general of all superclasses that you can use. Deriving from
425 <tt>Pass</tt> indicates that your pass uses the entire program as a unit,
426 refering to function bodies in no predictable order, or adding and removing
427 functions. Because nothing is known about the behavior of direct <tt>Pass</tt>
428 subclasses, no optimization can be done for their execution.</p>
430 <p>To write a correct <tt>Pass</tt> subclass, derive from <tt>Pass</tt> and
431 overload the <tt>run</tt> method with the following signature:</p>
435 <!-- _______________________________________________________________________ -->
436 <div class="doc_subsubsection">
437 <a name="run">The <tt>run</tt> method</a>
440 <div class="doc_text">
443 <b>virtual bool</b> run(Module &M) = 0;
446 <p>The <tt>run</tt> method performs the interesting work of the pass, and should
447 return true if the module was modified by the transformation, false
452 <!-- ======================================================================= -->
453 <div class="doc_subsection">
454 <a name="FunctionPass">The <tt>FunctionPass</tt> class</a>
457 <div class="doc_text">
459 <p>In contrast to direct <tt>Pass</tt> subclasses, direct <tt><a
460 href="http://llvm.cs.uiuc.edu/doxygen/classllvm_1_1Pass.html">FunctionPass</a></tt>
461 subclasses do have a predictable, local behavior that can be expected by the
462 system. All <tt>FunctionPass</tt> execute on each function in the program
463 independent of all of the other functions in the program.
464 <tt>FunctionPass</tt>'s do not require that they are executed in a particular
465 order, and <tt>FunctionPass</tt>'s do not modify external functions.</p>
467 <p>To be explicit, <tt>FunctionPass</tt> subclasses are not allowed to:</p>
470 <li>Modify a Function other than the one currently being processed.</li>
471 <li>Add or remove Function's from the current Module.</li>
472 <li>Add or remove global variables from the current Module.</li>
473 <li>Maintain state across invocations of
474 <a href="#runOnFunction"><tt>runOnFunction</tt></a> (including global data)</li>
477 <p>Implementing a <tt>FunctionPass</tt> is usually straightforward (See the <a
478 href="#basiccode">Hello World</a> pass for example). <tt>FunctionPass</tt>'s
479 may overload three virtual methods to do their work. All of these methods
480 should return true if they modified the program, or false if they didn't.</p>
484 <!-- _______________________________________________________________________ -->
485 <div class="subsubsection">
486 <a name="doInitialization_mod">The <tt>doInitialization(Module &)</tt>
490 <div class="doc_text">
493 <b>virtual bool</b> doInitialization(Module &M);
496 <p>The <tt>doIninitialize</tt> method is allowed to do most of the things that
497 <tt>FunctionPass</tt>'s are not allowed to do. They can add and remove
498 functions, get pointers to functions, etc. The <tt>doInitialization</tt> method
499 is designed to do simple initialization type of stuff that does not depend on
500 the functions being processed. The <tt>doInitialization</tt> method call is not
501 scheduled to overlap with any other pass executions (thus it should be very
504 <p>A good example of how this method should be used is the <a
505 href="http://llvm.cs.uiuc.edu/doxygen/LowerAllocations_8cpp-source.html">LowerAllocations</a>
506 pass. This pass converts <tt>malloc</tt> and <tt>free</tt> instructions into
507 platform dependent <tt>malloc()</tt> and <tt>free()</tt> function calls. It
508 uses the <tt>doInitialization</tt> method to get a reference to the malloc and
509 free functions that it needs, adding prototypes to the module if necessary.</p>
513 <!-- _______________________________________________________________________ -->
514 <div class="subsubsection">
515 <a name="runOnFunction">The <tt>runOnFunction</tt> method</a>
518 <div class="doc_text">
521 <b>virtual bool</b> runOnFunction(Function &F) = 0;
524 <p>The <tt>runOnFunction</tt> method must be implemented by your subclass to do
525 the transformation or analysis work of your pass. As usual, a true value should
526 be returned if the function is modified.</p>
530 <!-- _______________________________________________________________________ -->
531 <div class="subsubsection">
532 <a name="doFinalization_mod">The <tt>doFinalization(Module
533 &)</tt> method</a>
536 <div class="doc_text">
539 <b>virtual bool</b> doFinalization(Module &M);
542 <p>The <tt>doFinalization</tt> method is an infrequently used method that is
543 called when the pass framework has finished calling <a
544 href="#runOnFunction"><tt>runOnFunction</tt></a> for every function in the
545 program being compiled.</p>
549 <!-- ======================================================================= -->
550 <div class="doc_subsection">
551 <a name="BasicBlockPass">The <tt>BasicBlockPass</tt> class</a>
554 <div class="doc_text">
556 <p><tt>BasicBlockPass</tt>'s are just like <a
557 href="#FunctionPass"><tt>FunctionPass</tt></a>'s, except that they must limit
558 their scope of inspection and modification to a single basic block at a time.
559 As such, they are <b>not</b> allowed to do any of the following:</p>
562 <li>Modify or inspect any basic blocks outside of the current one</li>
563 <li>Maintain state across invocations of
564 <a href="#runOnBasicBlock"><tt>runOnBasicBlock</tt></a></li>
565 <li>Modify the constrol flow graph (by altering terminator instructions)</li>
566 <li>Any of the things verboten for
567 <a href="#FunctionPass"><tt>FunctionPass</tt></a>es.</li>
570 <p><tt>BasicBlockPass</tt>es are useful for traditional local and "peephole"
571 optimizations. They may override the same <a
572 href="#doInitialization_mod"><tt>doInitialization(Module &)</tt></a> and <a
573 href="#doFinalization_mod"><tt>doFinalization(Module &)</tt></a> methods that <a
574 href="#FunctionPass"><tt>FunctionPass</tt></a>'s have, but also have the following virtual methods that may also be implemented:</p>
578 <!-- _______________________________________________________________________ -->
579 <div class="subsubsection">
580 <a name="doInitialization_fn">The <tt>doInitialization(Function
581 &)</tt> method</a>
584 <div class="doc_text">
587 <b>virtual bool</b> doInitialization(Function &F);
590 <p>The <tt>doIninitialize</tt> method is allowed to do most of the things that
591 <tt>BasicBlockPass</tt>'s are not allowed to do, but that
592 <tt>FunctionPass</tt>'s can. The <tt>doInitialization</tt> method is designed
593 to do simple initialization type of stuff that does not depend on the
594 BasicBlocks being processed. The <tt>doInitialization</tt> method call is not
595 scheduled to overlap with any other pass executions (thus it should be very
600 <!-- _______________________________________________________________________ -->
601 <div class="subsubsection">
602 <a name="runOnBasicBlock">The <tt>runOnBasicBlock</tt> method</a>
605 <div class="doc_text">
608 <b>virtual bool</b> runOnBasicBlock(BasicBlock &BB) = 0;
611 <p>Override this function to do the work of the <tt>BasicBlockPass</tt>. This
612 function is not allowed to inspect or modify basic blocks other than the
613 parameter, and are not allowed to modify the CFG. A true value must be returned
614 if the basic block is modified.</p>
618 <!-- _______________________________________________________________________ -->
619 <div class="subsubsection">
620 <a name="doFinalization_fn">The <tt>doFinalization(Function &)</tt>
624 <div class="doc_text">
627 <b>virtual bool</b> doFinalization(Function &F);
630 <p>The <tt>doFinalization</tt> method is an infrequently used method that is
631 called when the pass framework has finished calling <a
632 href="#runOnBasicBlock"><tt>runOnBasicBlock</tt></a> for every BasicBlock in the
633 program being compiled. This can be used to perform per-function
638 <!-- ======================================================================= -->
639 <div class="doc_subsection">
640 <a name="MachineFunctionPass">The <tt>MachineFunctionPass</tt> class</a>
643 <div class="doc_text">
645 <p>A <tt>MachineFunctionPass</tt> executes on the machine-dependent
646 representation of each LLVM function in the program,
647 independent of all of the other functions in the program.
648 A <tt>MachineFunctionPass</tt> is also a <tt>FunctionPass</tt>, so all
649 the restrictions that apply to a <tt>FunctionPass</tt> also apply to it.
650 <tt>MachineFunctionPass</tt>es also have additional restrictions. In
651 particular, <tt>MachineFunctionPass</tt>es are not allowed to do any of
655 <li>Modify any LLVM Instructions, BasicBlocks or Functions.</li>
656 <li>Modify a MachineFunction other than the one currently being processed.</li>
657 <li>Add or remove MachineFunctions from the current Module.</li>
658 <li>Add or remove global variables from the current Module.</li>
659 <li>Maintain state across invocations of <a
660 href="#runOnMachineFunction"><tt>runOnMachineFunction</tt></a> (including global
666 <!-- _______________________________________________________________________ -->
667 <div class="subsubsection">
668 <a name="runOnMachineFunction">The <tt>runOnMachineFunction(MachineFunction
669 &MF)</tt> method</a>
672 <div class="doc_text">
675 <b>virtual bool</b> runOnMachineFunction(MachineFunction &MF) = 0;
678 <p><tt>runOnMachineFunction</tt> can be considered the main entry point of a
679 <tt>MachineFunctionPass</tt>; that is, you should override this method to do the
680 work of your <tt>MachineFunctionPass</tt>.</p>
682 <p>The <tt>runOnMachineFunction</tt> method is called on every
683 <tt>MachineFunction</tt> in a <tt>Module</tt>, so that the
684 <tt>MachineFunctionPass</tt> may perform optimizations on the machine-dependent
685 representation of the function. If you want to get at the LLVM <tt>Function</tt>
686 for the <tt>MachineFunction</tt> you're working on, use
687 <tt>MachineFunction</tt>'s <tt>getFunction()</tt> accessor method -- but
688 remember, you may not modify the LLVM <tt>Function</tt> or its contents from a
689 <tt>MachineFunctionPass</tt>.</p>
693 <!-- *********************************************************************** -->
694 <div class="doc_section">
695 <a name="registration">Pass registration</a>
697 <!-- *********************************************************************** -->
699 <div class="doc_text">
701 <p>In the <a href="#basiccode">Hello World</a> example pass we illustrated how
702 pass registration works, and discussed some of the reasons that it is used and
703 what it does. Here we discuss how and why passes are registered.</p>
705 <p>Passes can be registered in several different ways. Depending on the general
706 classification of the pass, you should use one of the following templates to
707 register the pass:</p>
710 <li><b><tt>RegisterOpt</tt></b> - This template should be used when you are
711 registering a pass that logically should be available for use in the
712 '<tt>opt</tt>' utility.</li>
714 <li><b><tt>RegisterAnalysis</tt></b> - This template should be used when you are
715 registering a pass that logically should be available for use in the
716 '<tt>analysis</tt>' utility.</li>
718 <li><b><tt>RegisterLLC</tt></b> - This template should be used when you are
719 registering a pass that logically should be available for use in the
720 '<tt>llc</tt>' utility.</li>
722 <li><b><tt>RegisterPass</tt></b> - This is the generic form of the
723 <tt>Register*</tt> templates that should be used if you want your pass listed by
724 multiple or no utilities. This template takes an extra third argument that
725 specifies which tools it should be listed in. See the <a
726 href="http://llvm.cs.uiuc.edu/doxygen/PassSupport_8h-source.html">PassSupport.h</a>
727 file for more information.</li>
731 <p>Regardless of how you register your pass, you must specify at least two
732 parameters. The first parameter is the name of the pass that is to be used on
733 the command line to specify that the pass should be added to a program (for
734 example <tt>opt</tt> or <tt>analyze</tt>). The second argument is the name of
735 the pass, which is to be used for the <tt>--help</tt> output of programs, as
736 well as for debug output generated by the <tt>--debug-pass</tt> option.</p>
738 <p>If you pass is constructed by its default constructor, you only ever have to
739 pass these two arguments. If, on the other hand, you require other information
740 (like target specific information), you must pass an additional argument. This
741 argument is a pointer to a function used to create the pass. For an example of
742 how this works, look at the <a
743 href="http://llvm.cs.uiuc.edu/doxygen/LowerAllocations_8cpp-source.html">LowerAllocations.cpp</a>
746 <p>If a pass is registered to be used by the <tt>analyze</tt> utility, you
747 should implement the virtual <tt>print</tt> method:</p>
751 <!-- _______________________________________________________________________ -->
752 <div class="doc_subsubsection">
753 <a name="print">The <tt>print</tt> method</a>
756 <div class="doc_text">
759 <b>virtual void</b> print(std::ostream &O, <b>const</b> Module *M) <b>const</b>;
762 <p>The <tt>print</tt> method must be implemented by "analyses" in order to print
763 a human readable version of the analysis results. This is useful for debugging
764 an analysis itself, as well as for other people to figure out how an analysis
765 works. The <tt>analyze</tt> tool uses this method to generate its output.</p>
767 <p>The <tt>ostream</tt> parameter specifies the stream to write the results on,
768 and the <tt>Module</tt> parameter gives a pointer to the top level module of the
769 program that has been analyzed. Note however that this pointer may be null in
770 certain circumstances (such as calling the <tt>Pass::dump()</tt> from a
771 debugger), so it should only be used to enhance debug output, it should not be
776 <!-- *********************************************************************** -->
777 <div class="doc_section">
778 <a name="interaction">Specifying interactions between passes</a>
780 <!-- *********************************************************************** -->
782 <div class="doc_text">
784 <p>One of the main responsibilities of the <tt>PassManager</tt> is the make sure
785 that passes interact with each other correctly. Because <tt>PassManager</tt>
786 tries to <a href="#passmanager">optimize the execution of passes</a> it must
787 know how the passes interact with each other and what dependencies exist between
788 the various passes. To track this, each pass can declare the set of passes that
789 are required to be executed before the current pass, and the passes which are
790 invalidated by the current pass.</p>
792 <p>Typically this functionality is used to require that analysis results are
793 computed before your pass is run. Running arbitrary transformation passes can
794 invalidate the computed analysis results, which is what the invalidation set
795 specifies. If a pass does not implement the <tt><a
796 href="#getAnalysisUsage">getAnalysisUsage</a></tt> method, it defaults to not
797 having any prerequisite passes, and invalidating <b>all</b> other passes.</p>
801 <!-- _______________________________________________________________________ -->
802 <div class="doc_subsubsection">
803 <a name="getAnalysisUsage">The <tt>getAnalysisUsage</tt> method</a>
806 <div class="doc_text">
809 <b>virtual void</b> getAnalysisUsage(AnalysisUsage &Info) <b>const</b>;
812 <p>By implementing the <tt>getAnalysisUsage</tt> method, the required and
813 invalidated sets may be specified for your transformation. The implementation
814 should fill in the <tt><a
815 href="http://llvm.cs.uiuc.edu/doxygen/classllvm_1_1AnalysisUsage.html">AnalysisUsage</a></tt>
816 object with information about which passes are required and not invalidated. To
817 do this, the following set methods are provided by the <tt><a
818 href="http://llvm.cs.uiuc.edu/doxygen/classllvm_1_1AnalysisUsage.html">AnalysisUsage</a></tt>
822 <i>// addRequires - Add the specified pass to the required set for your pass.</i>
823 <b>template</b><<b>class</b> PassClass>
824 AnalysisUsage &AnalysisUsage::addRequired();
826 <i>// addPreserved - Add the specified pass to the set of analyses preserved by
828 <b>template</b><<b>class</b> PassClass>
829 AnalysisUsage &AnalysisUsage::addPreserved();
831 <i>// setPreservesAll - Call this if the pass does not modify its input at all</i>
832 <b>void</b> AnalysisUsage::setPreservesAll();
834 <i>// setPreservesCFG - This function should be called by the pass, iff they do not:
836 // 1. Add or remove basic blocks from the function
837 // 2. Modify terminator instructions in any way.
839 // This is automatically implied for <a href="#BasicBlockPass">BasicBlockPass</a>'s
841 <b>void</b> AnalysisUsage::setPreservesCFG();
844 <p>Some examples of how to use these methods are:</p>
847 <i>// This is an example implementation from an analysis, which does not modify
848 // the program at all, yet has a prerequisite.</i>
849 <b>void</b> <a href="http://llvm.cs.uiuc.edu/doxygen/structllvm_1_1PostDominanceFrontier.html">PostDominanceFrontier</a>::getAnalysisUsage(AnalysisUsage &AU) <b>const</b> {
850 AU.setPreservesAll();
851 AU.addRequired<<a href="http://llvm.cs.uiuc.edu/doxygen/structllvm_1_1PostDominatorTree.html">PostDominatorTree</a>>();
858 <i>// This example modifies the program, but does not modify the CFG</i>
859 <b>void</b> <a href="http://llvm.cs.uiuc.edu/doxygen/structLICM.html">LICM</a>::getAnalysisUsage(AnalysisUsage &AU) <b>const</b> {
860 AU.setPreservesCFG();
861 AU.addRequired<<a href="http://llvm.cs.uiuc.edu/doxygen/classllvm_1_1LoopInfo.html">LoopInfo</a>>();
867 <!-- _______________________________________________________________________ -->
868 <div class="doc_subsubsection">
869 <a name="getAnalysis">The <tt>getAnalysis<></tt> method</a>
872 <div class="doc_text">
874 <p>The <tt>Pass::getAnalysis<></tt> method is inherited by your class,
875 providing you with access to the passes that you declared that you required with
876 the <a href="#getAnalysisUsage"><tt>getAnalysisUsage</tt></a> method. It takes
877 a single template argument that specifies which pass class you want, and returns
878 a reference to that pass.</p>
881 <b>template</b><<b>typename</b> PassClass>
882 AnalysisType &getAnalysis();
885 <p>This method call returns a reference to the pass desired. You may get a
886 runtime assertion failure if you attempt to get an analysis that you did not
887 declare as required in your <a
888 href="#getAnalysisUsage"><tt>getAnalysisUsage</tt></a> implementation. This
889 method can be called by your <tt>run*</tt> method implementation, or by any
890 other local method invoked by your <tt>run*</tt> method.</p>
894 <!-- *********************************************************************** -->
895 <div class="doc_section">
896 <a name="analysisgroup">Implementing Analysis Groups</a>
898 <!-- *********************************************************************** -->
900 <div class="doc_text">
902 <p>Now that we understand the basics of how passes are defined, how the are
903 used, and how they are required from other passes, it's time to get a little bit
904 fancier. All of the pass relationships that we have seen so far are very
905 simple: one pass depends on one other specific pass to be run before it can run.
906 For many applications, this is great, for others, more flexibility is
909 <p>In particular, some analyses are defined such that there is a single simple
910 interface to the analysis results, but multiple ways of calculating them.
911 Consider alias analysis for example. The most trivial alias analysis returns
912 "may alias" for any alias query. The most sophisticated analysis a
913 flow-sensitive, context-sensitive interprocedural analysis that can take a
914 significant amount of time to execute (and obviously, there is a lot of room
915 between these two extremes for other implementations). To cleanly support
916 situations like this, the LLVM Pass Infrastructure supports the notion of
921 <!-- _______________________________________________________________________ -->
922 <div class="subsubsection">
923 <a name="agconcepts">Analysis Group Concepts</a>
926 <div class="doc_text">
928 <p>An Analysis Group is a single simple interface that may be implemented by
929 multiple different passes. Analysis Groups can be given human readable names
930 just like passes, but unlike passes, they need not derive from the <tt>Pass</tt>
931 class. An analysis group may have one or more implementations, one of which is
932 the "default" implementation.</p>
934 <p>Analysis groups are used by client passes just like other passes are: the
935 <tt>AnalysisUsage::addRequired()</tt> and <tt>Pass::getAnalysis()</tt> methods.
936 In order to resolve this requirement, the <a href="#passmanager">PassManager</a>
937 scans the available passes to see if any implementations of the analysis group
938 are available. If none is available, the default implementation is created for
939 the pass to use. All standard rules for <A href="#interaction">interaction
940 between passes</a> still apply.</p>
942 <p>Although <a href="#registration">Pass Registration</a> is optional for normal
943 passes, all analysis group implementations must be registered, and must use the
944 <A href="#registerag"><tt>RegisterAnalysisGroup</tt></a> template to join the
945 implementation pool. Also, a default implementation of the interface
946 <b>must</b> be registered with <A
947 href="#registerag"><tt>RegisterAnalysisGroup</tt></a>.</p>
949 <p>As a concrete example of an Analysis Group in action, consider the <a
950 href="http://llvm.cs.uiuc.edu/doxygen/classllvm_1_1AliasAnalysis.html">AliasAnalysis</a>
951 analysis group. The default implementation of the alias analysis interface (the
953 href="http://llvm.cs.uiuc.edu/doxygen/structBasicAliasAnalysis.html">basicaa</a></tt>
954 pass) just does a few simple checks that don't require significant analysis to
955 compute (such as: two different globals can never alias each other, etc).
956 Passes that use the <tt><a
957 href="http://llvm.cs.uiuc.edu/doxygen/classllvm_1_1AliasAnalysis.html">AliasAnalysis</a></tt>
958 interface (for example the <tt><a
959 href="http://llvm.cs.uiuc.edu/doxygen/classGCSE.html">gcse</a></tt> pass), do
960 not care which implementation of alias analysis is actually provided, they just
961 use the designated interface.</p>
963 <p>From the user's perspective, commands work just like normal. Issuing the
964 command '<tt>opt -gcse ...</tt>' will cause the <tt>basicaa</tt> class to be
965 instantiated and added to the pass sequence. Issuing the command '<tt>opt
966 -somefancyaa -gcse ...</tt>' will cause the <tt>gcse</tt> pass to use the
967 <tt>somefancyaa</tt> alias analysis (which doesn't actually exist, it's just a
968 hypothetical example) instead.</p>
972 <!-- _______________________________________________________________________ -->
973 <div class="subsubsection">
974 <a name="registerag">Using <tt>RegisterAnalysisGroup</tt></a>
977 <div class="doc_text">
979 <p>The <tt>RegisterAnalysisGroup</tt> template is used to register the analysis
980 group itself as well as add pass implementations to the analysis group. First,
981 an analysis should be registered, with a human readable name provided for it.
982 Unlike registration of passes, there is no command line argument to be specified
983 for the Analysis Group Interface itself, because it is "abstract":</p>
986 <b>static</b> RegisterAnalysisGroup<<a href="http://llvm.cs.uiuc.edu/doxygen/classllvm_1_1AliasAnalysis.html">AliasAnalysis</a>> A("<i>Alias Analysis</i>");
989 <p>Once the analysis is registered, passes can declare that they are valid
990 implementations of the interface by using the following code:</p>
994 //<i> Analysis Group implementations <b>must</b> be registered normally...</i>
995 RegisterOpt<FancyAA>
996 B("<i>somefancyaa</i>", "<i>A more complex alias analysis implementation</i>");
998 //<i> Declare that we implement the AliasAnalysis interface</i>
999 RegisterAnalysisGroup<<a href="http://llvm.cs.uiuc.edu/doxygen/classllvm_1_1AliasAnalysis.html">AliasAnalysis</a>, FancyAA> C;
1003 <p>This just shows a class <tt>FancyAA</tt> that is registered normally, then
1004 uses the <tt>RegisterAnalysisGroup</tt> template to "join" the <tt><a
1005 href="http://llvm.cs.uiuc.edu/doxygen/classllvm_1_1AliasAnalysis.html">AliasAnalysis</a></tt>
1006 analysis group. Every implementation of an analysis group should join using
1007 this template. A single pass may join multiple different analysis groups with
1012 //<i> Analysis Group implementations <b>must</b> be registered normally...</i>
1013 RegisterOpt<<a href="http://llvm.cs.uiuc.edu/doxygen/structBasicAliasAnalysis.html">BasicAliasAnalysis</a>>
1014 D("<i>basicaa</i>", "<i>Basic Alias Analysis (default AA impl)</i>");
1016 //<i> Declare that we implement the AliasAnalysis interface</i>
1017 RegisterAnalysisGroup<<a href="http://llvm.cs.uiuc.edu/doxygen/classllvm_1_1AliasAnalysis.html">AliasAnalysis</a>, <a href="http://llvm.cs.uiuc.edu/doxygen/structBasicAliasAnalysis.html">BasicAliasAnalysis</a>, <b>true</b>> E;
1021 <p>Here we show how the default implementation is specified (using the extra
1022 argument to the <tt>RegisterAnalysisGroup</tt> template). There must be exactly
1023 one default implementation available at all times for an Analysis Group to be
1024 used. Here we declare that the <tt><a
1025 href="http://llvm.cs.uiuc.edu/doxygen/structBasicAliasAnalysis.html">BasicAliasAnalysis</a></tt>
1026 pass is the default implementation for the interface.</p>
1030 <!-- *********************************************************************** -->
1031 <div class="doc_section">
1032 <a name="passmanager">What PassManager does</a>
1034 <!-- *********************************************************************** -->
1036 <div class="doc_text">
1039 href="http://llvm.cs.uiuc.edu/doxygen/PassManager_8h-source.html"><tt>PassManager</tt></a>
1041 href="http://llvm.cs.uiuc.edu/doxygen/classllvm_1_1PassManager.html">class</a>
1042 takes a list of passes, ensures their <a href="#interaction">prerequisites</a>
1043 are set up correctly, and then schedules passes to run efficiently. All of the
1044 LLVM tools that run passes use the <tt>PassManager</tt> for execution of these
1047 <p>The <tt>PassManager</tt> does two main things to try to reduce the execution
1048 time of a series of passes:</p>
1051 <li><b>Share analysis results</b> - The PassManager attempts to avoid
1052 recomputing analysis results as much as possible. This means keeping track of
1053 which analyses are available already, which analyses get invalidated, and which
1054 analyses are needed to be run for a pass. An important part of work is that the
1055 <tt>PassManager</tt> tracks the exact lifetime of all analysis results, allowing
1056 it to <a href="#releaseMemory">free memory</a> allocated to holding analysis
1057 results as soon as they are no longer needed.</li>
1059 <li><b>Pipeline the execution of passes on the program</b> - The
1060 <tt>PassManager</tt> attempts to get better cache and memory usage behavior out
1061 of a series of passes by pipelining the passes together. This means that, given
1062 a series of consequtive <a href="#FunctionPass"><tt>FunctionPass</tt></a>'s, it
1063 will execute all of the <a href="#FunctionPass"><tt>FunctionPass</tt></a>'s on
1064 the first function, then all of the <a
1065 href="#FunctionPass"><tt>FunctionPass</tt></a>es on the second function,
1066 etc... until the entire program has been run through the passes.
1068 <p>This improves the cache behavior of the compiler, because it is only touching
1069 the LLVM program representation for a single function at a time, instead of
1070 traversing the entire program. It reduces the memory consumption of compiler,
1071 because, for example, only one <a
1072 href="http://llvm.cs.uiuc.edu/doxygen/structllvm_1_1DominatorSet.html"><tt>DominatorSet</tt></a>
1073 needs to be calculated at a time. This also makes it possible some <a
1074 href="#SMP">interesting enhancements</a> in the future.</p></li>
1078 <p>The effectiveness of the <tt>PassManager</tt> is influenced directly by how
1079 much information it has about the behaviors of the passes it is scheduling. For
1080 example, the "preserved" set is intentionally conservative in the face of an
1081 unimplemented <a href="#getAnalysisUsage"><tt>getAnalysisUsage</tt></a> method.
1082 Not implementing when it should be implemented will have the effect of not
1083 allowing any analysis results to live across the execution of your pass.</p>
1085 <p>The <tt>PassManager</tt> class exposes a <tt>--debug-pass</tt> command line
1086 options that is useful for debugging pass execution, seeing how things work, and
1087 diagnosing when you should be preserving more analyses than you currently are
1088 (To get information about all of the variants of the <tt>--debug-pass</tt>
1089 option, just type '<tt>opt --help-hidden</tt>').</p>
1091 <p>By using the <tt>--debug-pass=Structure</tt> option, for example, we can see
1092 how our <a href="#basiccode">Hello World</a> pass interacts with other passes.
1093 Lets try it out with the <tt>gcse</tt> and <tt>licm</tt> passes:</p>
1096 $ opt -load ../../../lib/Debug/libhello.so -gcse -licm --debug-pass=Structure < hello.bc > /dev/null
1098 Function Pass Manager
1099 Dominator Set Construction
1100 Immediate Dominators Construction
1101 Global Common Subexpression Elimination
1102 -- Immediate Dominators Construction
1103 -- Global Common Subexpression Elimination
1104 Natural Loop Construction
1105 Loop Invariant Code Motion
1106 -- Natural Loop Construction
1107 -- Loop Invariant Code Motion
1109 -- Dominator Set Construction
1115 <p>This output shows us when passes are constructed and when the analysis
1116 results are known to be dead (prefixed with '<tt>--</tt>'). Here we see that
1117 GCSE uses dominator and immediate dominator information to do its job. The LICM
1118 pass uses natural loop information, which uses dominator sets, but not immediate
1119 dominators. Because immediate dominators are no longer useful after the GCSE
1120 pass, it is immediately destroyed. The dominator sets are then reused to
1121 compute natural loop information, which is then used by the LICM pass.</p>
1123 <p>After the LICM pass, the module verifier runs (which is automatically added
1124 by the '<tt>opt</tt>' tool), which uses the dominator set to check that the
1125 resultant LLVM code is well formed. After it finishes, the dominator set
1126 information is destroyed, after being computed once, and shared by three
1129 <p>Lets see how this changes when we run the <a href="#basiccode">Hello
1130 World</a> pass in between the two passes:</p>
1133 $ opt -load ../../../lib/Debug/libhello.so -gcse -hello -licm --debug-pass=Structure < hello.bc > /dev/null
1135 Function Pass Manager
1136 Dominator Set Construction
1137 Immediate Dominators Construction
1138 Global Common Subexpression Elimination
1139 <b>-- Dominator Set Construction</b>
1140 -- Immediate Dominators Construction
1141 -- Global Common Subexpression Elimination
1142 <b> Hello World Pass
1144 Dominator Set Construction</b>
1145 Natural Loop Construction
1146 Loop Invariant Code Motion
1147 -- Natural Loop Construction
1148 -- Loop Invariant Code Motion
1150 -- Dominator Set Construction
1159 <p>Here we see that the <a href="#basiccode">Hello World</a> pass has killed the
1160 Dominator Set pass, even though it doesn't modify the code at all! To fix this,
1161 we need to add the following <a
1162 href="#getAnalysisUsage"><tt>getAnalysisUsage</tt></a> method to our pass:</p>
1165 <i>// We don't modify the program, so we preserve all analyses</i>
1166 <b>virtual void</b> getAnalysisUsage(AnalysisUsage &AU) <b>const</b> {
1167 AU.setPreservesAll();
1171 <p>Now when we run our pass, we get this output:</p>
1174 $ opt -load ../../../lib/Debug/libhello.so -gcse -hello -licm --debug-pass=Structure < hello.bc > /dev/null
1175 Pass Arguments: -gcse -hello -licm
1177 Function Pass Manager
1178 Dominator Set Construction
1179 Immediate Dominators Construction
1180 Global Common Subexpression Elimination
1181 -- Immediate Dominators Construction
1182 -- Global Common Subexpression Elimination
1185 Natural Loop Construction
1186 Loop Invariant Code Motion
1187 -- Loop Invariant Code Motion
1188 -- Natural Loop Construction
1190 -- Dominator Set Construction
1199 <p>Which shows that we don't accidentally invalidate dominator information
1200 anymore, and therefore do not have to compute it twice.</p>
1204 <!-- _______________________________________________________________________ -->
1205 <div class="doc_subsubsection">
1206 <a name="releaseMemory">The <tt>releaseMemory</tt> method</a>
1209 <div class="doc_text">
1212 <b>virtual void</b> releaseMemory();
1215 <p>The <tt>PassManager</tt> automatically determines when to compute analysis
1216 results, and how long to keep them around for. Because the lifetime of the pass
1217 object itself is effectively the entire duration of the compilation process, we
1218 need some way to free analysis results when they are no longer useful. The
1219 <tt>releaseMemory</tt> virtual method is the way to do this.</p>
1221 <p>If you are writing an analysis or any other pass that retains a significant
1222 amount of state (for use by another pass which "requires" your pass and uses the
1223 <a href="#getAnalysis">getAnalysis</a> method) you should implement
1224 <tt>releaseMEmory</tt> to, well, release the memory allocated to maintain this
1225 internal state. This method is called after the <tt>run*</tt> method for the
1226 class, before the next call of <tt>run*</tt> in your pass.</p>
1230 <!-- *********************************************************************** -->
1231 <div class="doc_section">
1232 <a name="debughints">Using GDB with dynamically loaded passes</a>
1234 <!-- *********************************************************************** -->
1236 <div class="doc_text">
1238 <p>Unfortunately, using GDB with dynamically loaded passes is not as easy as it
1239 should be. First of all, you can't set a breakpoint in a shared object that has
1240 not been loaded yet, and second of all there are problems with inlined functions
1241 in shared objects. Here are some suggestions to debugging your pass with
1244 <p>For sake of discussion, I'm going to assume that you are debugging a
1245 transformation invoked by <tt>opt</tt>, although nothing described here depends
1250 <!-- _______________________________________________________________________ -->
1251 <div class="doc_subsubsection">
1252 <a name="breakpoint">Setting a breakpoint in your pass</a>
1255 <div class="doc_text">
1257 <p>First thing you do is start <tt>gdb</tt> on the <tt>opt</tt> process:</p>
1262 Copyright 2000 Free Software Foundation, Inc.
1263 GDB is free software, covered by the GNU General Public License, and you are
1264 welcome to change it and/or distribute copies of it under certain conditions.
1265 Type "show copying" to see the conditions.
1266 There is absolutely no warranty for GDB. Type "show warranty" for details.
1267 This GDB was configured as "sparc-sun-solaris2.6"...
1271 <p>Note that <tt>opt</tt> has a lot of debugging information in it, so it takes
1272 time to load. Be patient. Since we cannot set a breakpoint in our pass yet
1273 (the shared object isn't loaded until runtime), we must execute the process, and
1274 have it stop before it invokes our pass, but after it has loaded the shared
1275 object. The most foolproof way of doing this is to set a breakpoint in
1276 <tt>PassManager::run</tt> and then run the process with the arguments you
1280 (gdb) <b>break PassManager::run</b>
1281 Breakpoint 1 at 0x2413bc: file Pass.cpp, line 70.
1282 (gdb) <b>run test.bc -load $(LLVMTOP)/llvm/lib/Debug/[libname].so -[passoption]</b>
1283 Starting program: opt test.bc -load $(LLVMTOP)/llvm/lib/Debug/[libname].so -[passoption]
1284 Breakpoint 1, PassManager::run (this=0xffbef174, M=@0x70b298) at Pass.cpp:70
1285 70 bool PassManager::run(Module &M) { return PM->run(M); }
1289 <p>Once the <tt>opt</tt> stops in the <tt>PassManager::run</tt> method you are
1290 now free to set breakpoints in your pass so that you can trace through execution
1291 or do other standard debugging stuff.</p>
1295 <!-- _______________________________________________________________________ -->
1296 <div class="doc_subsubsection">
1297 <a name="debugmisc">Miscellaneous Problems</a>
1300 <div class="doc_text">
1302 <p>Once you have the basics down, there are a couple of problems that GDB has,
1303 some with solutions, some without.</p>
1306 <li>Inline functions have bogus stack information. In general, GDB does a
1307 pretty good job getting stack traces and stepping through inline functions.
1308 When a pass is dynamically loaded however, it somehow completely loses this
1309 capability. The only solution I know of is to de-inline a function (move it
1310 from the body of a class to a .cpp file).</li>
1312 <li>Restarting the program breaks breakpoints. After following the information
1313 above, you have succeeded in getting some breakpoints planted in your pass. Nex
1314 thing you know, you restart the program (i.e., you type '<tt>run</tt>' again),
1315 and you start getting errors about breakpoints being unsettable. The only way I
1316 have found to "fix" this problem is to <tt>delete</tt> the breakpoints that are
1317 already set in your pass, run the program, and re-set the breakpoints once
1318 execution stops in <tt>PassManager::run</tt>.</li>
1322 <p>Hopefully these tips will help with common case debugging situations. If
1323 you'd like to contribute some tips of your own, just contact <a
1324 href="mailto:sabre@nondot.org">Chris</a>.</p>
1328 <!-- *********************************************************************** -->
1329 <div class="doc_section">
1330 <a name="future">Future extensions planned</a>
1332 <!-- *********************************************************************** -->
1334 <div class="doc_text">
1336 <p>Although the LLVM Pass Infrastructure is very capable as it stands, and does
1337 some nifty stuff, there are things we'd like to add in the future. Here is
1338 where we are going:</p>
1342 <!-- _______________________________________________________________________ -->
1343 <div class="subsubsection">
1344 <a name="SMP">Multithreaded LLVM</a>
1347 <div class="doc_text">
1349 <p>Multiple CPU machines are becoming more common and compilation can never be
1350 fast enough: obviously we should allow for a multithreaded compiler. Because of
1351 the semantics defined for passes above (specifically they cannot maintain state
1352 across invocations of their <tt>run*</tt> methods), a nice clean way to
1353 implement a multithreaded compiler would be for the <tt>PassManager</tt> class
1354 to create multiple instances of each pass object, and allow the separate
1355 instances to be hacking on different parts of the program at the same time.</p>
1357 <p>This implementation would prevent each of the passes from having to implement
1358 multithreaded constructs, requiring only the LLVM core to have locking in a few
1359 places (for global resources). Although this is a simple extension, we simply
1360 haven't had time (or multiprocessor machines, thus a reason) to implement this.
1361 Despite that, we have kept the LLVM passes SMP ready, and you should too.</p>
1365 <!-- _______________________________________________________________________ -->
1366 <div class="subsubsection">
1367 <a name="ModuleSource">A new <tt>ModuleSource</tt> interface</a>
1370 <div class="doc_text">
1372 <p>Currently, the <tt>PassManager</tt>'s <tt>run</tt> method takes a <tt><a
1373 href="http://llvm.cs.uiuc.edu/doxygen/structllvm_1_1Module.html">Module</a></tt>
1374 as input, and runs all of the passes on this module. The problem with this
1375 approach is that none of the <tt>PassManager</tt> features can be used for
1376 timing and debugging the actual <b>loading</b> of the module from disk or
1379 <p>To solve this problem, eventually the <tt>PassManger</tt> class will accept a
1380 <tt>ModuleSource</tt> object instead of a Module itself. When complete, this
1381 will also allow for streaming of functions out of the bytecode representation,
1382 allowing us to avoid holding the entire program in memory at once if we only are
1383 dealing with <a href="#FunctionPass">FunctionPass</a>es.</p>
1385 <p>As part of a different issue, eventually the bytecode loader will be extended
1386 to allow on-demand loading of functions from the bytecode representation, in
1387 order to better support the runtime reoptimizer. The bytecode format is already
1388 capable of this, the loader just needs to be reworked a bit.</p>
1392 <!-- _______________________________________________________________________ -->
1393 <div class="subsubsection">
1394 <a name="PassFunctionPass"><tt>Pass</tt>'s requiring <tt>FunctionPass</tt>'s</a>
1397 <div class="doc_text">
1399 <p>Currently it is illegal for a <a href="#Pass"><tt>Pass</tt></a> to require a
1400 <a href="#FunctionPass"><tt>FunctionPass</tt></a>. This is because there is
1401 only one instance of the <a href="#FunctionPass"><tt>FunctionPass</tt></a>
1402 object ever created, thus nowhere to store information for all of the functions
1403 in the program at the same time. Although this has come up a couple of times
1404 before, this has always been worked around by factoring one big complicated pass
1405 into a global and an interprocedural part, both of which are distinct. In the
1406 future, it would be nice to have this though.</p>
1408 <p>Note that it is no problem for a <a
1409 href="#FunctionPass"><tt>FunctionPass</tt></a> to require the results of a <a
1410 href="#Pass"><tt>Pass</tt></a>, only the other way around.</p>
1414 <!-- *********************************************************************** -->
1417 <a href="http://jigsaw.w3.org/css-validator/check/referer"><img
1418 src="http://jigsaw.w3.org/css-validator/images/vcss" alt="Valid CSS!"></a>
1419 <a href="http://validator.w3.org/check/referer"><img
1420 src="http://www.w3.org/Icons/valid-html401" alt="Valid HTML 4.01!" /></a>
1422 <a href="mailto:sabre@nondot.org">Chris Lattner</a><br>
1423 <a href="http://llvm.cs.uiuc.edu">The LLVM Compiler Infrastructure</a><br>
1424 Last modified: $Date$