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="#AU::addRequired">The <tt>AnalysisUsage::addRequired<></tt> and <tt>AnalysisUsage::addRequiredTransitive<></tt> methods</a></li>
63 <li><a href="#AU::addPreserved">The <tt>AnalysisUsage::addPreserved<></tt> method</a></li>
64 <li><a href="#AU::examples">Example implementations of <tt>getAnalysisUsage</tt></a></li>
65 <li><a href="#getAnalysis">The <tt>getAnalysis<></tt> and <tt>getAnalysisToUpdate<></tt> methods</a></li>
67 <li><a href="#analysisgroup">Implementing Analysis Groups</a>
69 <li><a href="#agconcepts">Analysis Group Concepts</a></li>
70 <li><a href="#registerag">Using <tt>RegisterAnalysisGroup</tt></a></li>
72 <li><a href="#passmanager">What PassManager does</a>
74 <li><a href="#releaseMemory">The <tt>releaseMemory</tt> method</a></li>
76 <li><a href="#debughints">Using GDB with dynamically loaded passes</a>
78 <li><a href="#breakpoint">Setting a breakpoint in your pass</a></li>
79 <li><a href="#debugmisc">Miscellaneous Problems</a></li>
81 <li><a href="#future">Future extensions planned</a>
83 <li><a href="#SMP">Multithreaded LLVM</a></li>
84 <li><a href="#ModuleSource">A new <tt>ModuleSource</tt> interface</a></li>
85 <li><a href="#PassFunctionPass"><tt>Pass</tt>es requiring
86 <tt>FunctionPass</tt>es</a></li>
90 <div class="doc_author">
91 <p>Written by <a href="mailto:sabre@nondot.org">Chris Lattner</a></p>
94 <!-- *********************************************************************** -->
95 <div class="doc_section">
96 <a name="introduction">Introduction - What is a pass?</a>
98 <!-- *********************************************************************** -->
100 <div class="doc_text">
102 <p>The LLVM Pass Framework is an important part of the LLVM system, because LLVM
103 passes are where the interesting parts of the compiler exist. Passes perform
104 the transformations and optimizations that make up the compiler, they build
105 the analysis results that are used by these transformations, and they are, above
106 all, a structuring technique for compiler code.</p>
108 <p>All LLVM passes are subclasses of the <tt><a
109 href="http://llvm.cs.uiuc.edu/doxygen/classllvm_1_1Pass.html">Pass</a></tt>
110 class, which implement functionality by overriding virtual methods inherited
111 from <tt>Pass</tt>. Depending on how your pass works, you may be able to
112 inherit from the <tt><a
113 href="http://llvm.cs.uiuc.edu/doxygen/structllvm_1_1FunctionPass.html">FunctionPass</a></tt>
115 href="http://llvm.cs.uiuc.edu/doxygen/structllvm_1_1BasicBlockPass.html">BasicBlockPass</a></tt>,
116 which gives the system more information about what your pass does, and how it
117 can be combined with other passes. One of the main features of the LLVM Pass
118 Framework is that it schedules passes to run in an efficient way based on the
119 constraints that your pass has.</p>
121 <p>We start by showing you how to construct a pass, everything from setting up
122 the code, to compiling, loading, and executing it. After the basics are down,
123 more advanced features are discussed.</p>
127 <!-- *********************************************************************** -->
128 <div class="doc_section">
129 <a name="quickstart">Quick Start - Writing hello world</a>
131 <!-- *********************************************************************** -->
133 <div class="doc_text">
135 <p>Here we describe how to write the "hello world" of passes. The "Hello" pass
136 is designed to simply print out the name of non-external functions that exist in
137 the program being compiled. It does not modify the program at all, just
138 inspects it. The source code and files for this pass are available in the LLVM
139 source tree in the <tt>lib/Transforms/Hello</tt> directory.</p>
143 <!-- ======================================================================= -->
144 <div class="doc_subsection">
145 <a name="makefile">Setting up the build environment</a>
148 <div class="doc_text">
150 <p>First thing you need to do is create a new directory somewhere in the LLVM
151 source base. For this example, we'll assume that you made
152 "<tt>lib/Transforms/Hello</tt>". The first thing you must do is set up a build
153 script (Makefile) that will compile the source code for the new pass. To do
154 this, copy this into "<tt>Makefile</tt>":</p>
159 # Makefile for hello pass
161 # Path to top level of LLVM heirarchy
164 # Name of the library to build
167 # Build a dynamically loadable shared object
170 # Include the makefile implementation stuff
171 include $(LEVEL)/Makefile.common
174 <p>This makefile specifies that all of the <tt>.cpp</tt> files in the current
175 directory are to be compiled and linked together into a
176 <tt>lib/Debug/libhello.so</tt> shared object that can be dynamically loaded by
177 the <tt>opt</tt> or <tt>analyze</tt> tools. If your operating system uses a
178 suffix other than .so (such as windows of Mac OS/X), the appropriate extension
181 <p>Now that we have the build scripts set up, we just need to write the code for
186 <!-- ======================================================================= -->
187 <div class="doc_subsection">
188 <a name="basiccode">Basic code required</a>
191 <div class="doc_text">
193 <p>Now that we have a way to compile our new pass, we just have to write it.
197 <b>#include</b> "<a href="http://llvm.cs.uiuc.edu/doxygen/Pass_8h-source.html">llvm/Pass.h</a>"
198 <b>#include</b> "<a href="http://llvm.cs.uiuc.edu/doxygen/Function_8h-source.html">llvm/Function.h</a>"
201 <p>Which are needed because we are writing a <tt><a
202 href="http://llvm.cs.uiuc.edu/doxygen/classllvm_1_1Pass.html">Pass</a></tt>, and
203 we are operating on <tt><a
204 href="http://llvm.cs.uiuc.edu/doxygen/classllvm_1_1Function.html">Function</a></tt>'s.</p>
208 <b>using namespace llvm;</b>
210 <p>... which is required because the functions from the include files
211 live in the llvm namespace.
220 <p>... which starts out an anonymous namespace. Anonymous namespaces are to C++
221 what the "<tt>static</tt>" keyword is to C (at global scope). It makes the
222 things declared inside of the anonymous namespace only visible to the current
223 file. If you're not familiar with them, consult a decent C++ book for more
226 <p>Next, we declare our pass itself:</p>
229 <b>struct</b> Hello : <b>public</b> <a href="#FunctionPass">FunctionPass</a> {
232 <p>This declares a "<tt>Hello</tt>" class that is a subclass of <tt><a
233 href="http://llvm.cs.uiuc.edu/doxygen/structllvm_1_1FunctionPass.html">FunctionPass</a></tt>.
234 The different builtin pass subclasses are described in detail <a
235 href="#passtype">later</a>, but for now, know that <a
236 href="#FunctionPass"><tt>FunctionPass</tt></a>'s operate a function at a
240 <b>virtual bool</b> <a href="#runOnFunction">runOnFunction</a>(Function &F) {
241 std::cerr << "<i>Hello: </i>" << F.getName() << "\n";
244 }; <i>// end of struct Hello</i>
247 <p>We declare a "<a href="#runOnFunction"><tt>runOnFunction</tt></a>" method,
248 which overloads an abstract virtual method inherited from <a
249 href="#FunctionPass"><tt>FunctionPass</tt></a>. This is where we are supposed
250 to do our thing, so we just print out our message with the name of each
254 RegisterOpt<Hello> X("<i>hello</i>", "<i>Hello World Pass</i>");
255 } <i>// end of anonymous namespace</i>
258 <p>Lastly, we register our class <tt>Hello</tt>, giving it a command line
259 argument "<tt>hello</tt>", and a name "<tt>Hello World Pass</tt>". There are
260 several different ways of <a href="#registration">registering your pass</a>,
261 depending on what it is to be used for. For "optimizations" we use the
262 <tt>RegisterOpt</tt> template.</p>
264 <p>As a whole, the <tt>.cpp</tt> file looks like:</p>
267 <b>#include</b> "<a href="http://llvm.cs.uiuc.edu/doxygen/Pass_8h-source.html">llvm/Pass.h</a>"
268 <b>#include</b> "<a href="http://llvm.cs.uiuc.edu/doxygen/Function_8h-source.html">llvm/Function.h</a>"
270 <b>using namespace llvm;</b>
273 <b>struct Hello</b> : <b>public</b> <a href="#FunctionPass">FunctionPass</a> {
274 <b>virtual bool</b> <a href="#runOnFunction">runOnFunction</a>(Function &F) {
275 std::cerr << "<i>Hello: </i>" << F.getName() << "\n";
280 RegisterOpt<Hello> X("<i>hello</i>", "<i>Hello World Pass</i>");
284 <p>Now that it's all together, compile the file with a simple "<tt>gmake</tt>"
285 command in the local directory and you should get a new
286 "<tt>lib/Debug/libhello.so</tt> file. Note that everything in this file is
287 contained in an anonymous namespace: this reflects the fact that passes are self
288 contained units that do not need external interfaces (although they can have
289 them) to be useful.</p>
293 <!-- ======================================================================= -->
294 <div class="doc_subsection">
295 <a name="running">Running a pass with <tt>opt</tt> or <tt>analyze</tt></a>
298 <div class="doc_text">
300 <p>Now that you have a brand new shiny shared object file, we can use the
301 <tt>opt</tt> command to run an LLVM program through your pass. Because you
302 registered your pass with the <tt>RegisterOpt</tt> template, you will be able to
303 use the <tt>opt</tt> tool to access it, once loaded.</p>
305 <p>To test it, follow the example at the end of the <a
306 href="GettingStarted.html">Getting Started Guide</a> to compile "Hello World" to
307 LLVM. We can now run the bytecode file (<tt>hello.bc</tt>) for the program
308 through our transformation like this (or course, any bytecode file will
312 $ opt -load ../../../lib/Debug/libhello.so -hello < hello.bc > /dev/null
318 <p>The '<tt>-load</tt>' option specifies that '<tt>opt</tt>' should load your
319 pass as a shared object, which makes '<tt>-hello</tt>' a valid command line
320 argument (which is one reason you need to <a href="#registration">register your
321 pass</a>). Because the hello pass does not modify the program in any
322 interesting way, we just throw away the result of <tt>opt</tt> (sending it to
323 <tt>/dev/null</tt>).</p>
325 <p>To see what happened to the other string you registered, try running
326 <tt>opt</tt> with the <tt>--help</tt> option:</p>
329 $ opt -load ../../../lib/Debug/libhello.so --help
330 OVERVIEW: llvm .bc -> .bc modular optimizer
332 USAGE: opt [options] <input bytecode>
335 Optimizations available:
337 -funcresolve - Resolve Functions
338 -gcse - Global Common Subexpression Elimination
339 -globaldce - Dead Global Elimination
340 <b>-hello - Hello World Pass</b>
341 -indvars - Canonicalize Induction Variables
342 -inline - Function Integration/Inlining
343 -instcombine - Combine redundant instructions
347 <p>The pass name get added as the information string for your pass, giving some
348 documentation to users of <tt>opt</tt>. Now that you have a working pass, you
349 would go ahead and make it do the cool transformations you want. Once you get
350 it all working and tested, it may become useful to find out how fast your pass
351 is. The <a href="#passManager"><tt>PassManager</tt></a> provides a nice command
352 line option (<tt>--time-passes</tt>) that allows you to get information about
353 the execution time of your pass along with the other passes you queue up. For
357 $ opt -load ../../../lib/Debug/libhello.so -hello -time-passes < hello.bc > /dev/null
361 ===============================================================================
362 ... Pass execution timing report ...
363 ===============================================================================
364 Total Execution Time: 0.02 seconds (0.0479059 wall clock)
366 ---User Time--- --System Time-- --User+System-- ---Wall Time--- --- Pass Name ---
367 0.0100 (100.0%) 0.0000 ( 0.0%) 0.0100 ( 50.0%) 0.0402 ( 84.0%) Bytecode Writer
368 0.0000 ( 0.0%) 0.0100 (100.0%) 0.0100 ( 50.0%) 0.0031 ( 6.4%) Dominator Set Construction
369 0.0000 ( 0.0%) 0.0000 ( 0.0%) 0.0000 ( 0.0%) 0.0013 ( 2.7%) Module Verifier
370 <b> 0.0000 ( 0.0%) 0.0000 ( 0.0%) 0.0000 ( 0.0%) 0.0033 ( 6.9%) Hello World Pass</b>
371 0.0100 (100.0%) 0.0100 (100.0%) 0.0200 (100.0%) 0.0479 (100.0%) TOTAL
374 <p>As you can see, our implementation above is pretty fast :). The additional
375 passes listed are automatically inserted by the '<tt>opt</tt>' tool to verify
376 that the LLVM emitted by your pass is still valid and well formed LLVM, which
377 hasn't been broken somehow.</p>
379 <p>Now that you have seen the basics of the mechanics behind passes, we can talk
380 about some more details of how they work and how to use them.</p>
384 <!-- *********************************************************************** -->
385 <div class="doc_section">
386 <a name="passtype">Pass classes and requirements</a>
388 <!-- *********************************************************************** -->
390 <div class="doc_text">
392 <p>One of the first things that you should do when designing a new pass is to
393 decide what class you should subclass for your pass. The <a
394 href="#basiccode">Hello World</a> example uses the <tt><a
395 href="#FunctionPass">FunctionPass</a></tt> class for its implementation, but we
396 did not discuss why or when this should occur. Here we talk about the classes
397 available, from the most general to the most specific.</p>
399 <p>When choosing a superclass for your Pass, you should choose the <b>most
400 specific</b> class possible, while still being able to meet the requirements
401 listed. This gives the LLVM Pass Infrastructure information necessary to
402 optimize how passes are run, so that the resultant compiler isn't unneccesarily
407 <!-- ======================================================================= -->
408 <div class="doc_subsection">
409 <a name="ImmutablePass">The <tt>ImmutablePass</tt> class</a>
412 <div class="doc_text">
414 <p>The most plain and boring type of pass is the "<tt><a
415 href="http://llvm.cs.uiuc.edu/doxygen/structllvm_1_1ImmutablePass.html">ImmutablePass</a></tt>"
416 class. This pass type is used for passes that do not have to be run, do not
417 change state, and never need to be updated. This is not a normal type of
418 transformation or analysis, but can provide information about the current
419 compiler configuration.</p>
421 <p>Although this pass class is very infrequently used, it is important for
422 providing information about the current target machine being compiled for, and
423 other static information that can affect the various transformations.</p>
425 <p><tt>ImmutablePass</tt>es never invalidate other transformations, are never
426 invalidated, and are never "run".</p>
430 <!-- ======================================================================= -->
431 <div class="doc_subsection">
432 <a name="Pass">The <tt>Pass</tt> class</a>
435 <div class="doc_text">
438 href="http://llvm.cs.uiuc.edu/doxygen/classllvm_1_1Pass.html">Pass</a></tt>"
439 class is the most general of all superclasses that you can use. Deriving from
440 <tt>Pass</tt> indicates that your pass uses the entire program as a unit,
441 refering to function bodies in no predictable order, or adding and removing
442 functions. Because nothing is known about the behavior of direct <tt>Pass</tt>
443 subclasses, no optimization can be done for their execution.</p>
445 <p>To write a correct <tt>Pass</tt> subclass, derive from <tt>Pass</tt> and
446 overload the <tt>run</tt> method with the following signature:</p>
450 <!-- _______________________________________________________________________ -->
451 <div class="doc_subsubsection">
452 <a name="run">The <tt>run</tt> method</a>
455 <div class="doc_text">
458 <b>virtual bool</b> run(Module &M) = 0;
461 <p>The <tt>run</tt> method performs the interesting work of the pass, and should
462 return true if the module was modified by the transformation, false
467 <!-- ======================================================================= -->
468 <div class="doc_subsection">
469 <a name="FunctionPass">The <tt>FunctionPass</tt> class</a>
472 <div class="doc_text">
474 <p>In contrast to direct <tt>Pass</tt> subclasses, direct <tt><a
475 href="http://llvm.cs.uiuc.edu/doxygen/classllvm_1_1Pass.html">FunctionPass</a></tt>
476 subclasses do have a predictable, local behavior that can be expected by the
477 system. All <tt>FunctionPass</tt> execute on each function in the program
478 independent of all of the other functions in the program.
479 <tt>FunctionPass</tt>'s do not require that they are executed in a particular
480 order, and <tt>FunctionPass</tt>'s do not modify external functions.</p>
482 <p>To be explicit, <tt>FunctionPass</tt> subclasses are not allowed to:</p>
485 <li>Modify a Function other than the one currently being processed.</li>
486 <li>Add or remove Function's from the current Module.</li>
487 <li>Add or remove global variables from the current Module.</li>
488 <li>Maintain state across invocations of
489 <a href="#runOnFunction"><tt>runOnFunction</tt></a> (including global data)</li>
492 <p>Implementing a <tt>FunctionPass</tt> is usually straightforward (See the <a
493 href="#basiccode">Hello World</a> pass for example). <tt>FunctionPass</tt>'s
494 may overload three virtual methods to do their work. All of these methods
495 should return true if they modified the program, or false if they didn't.</p>
499 <!-- _______________________________________________________________________ -->
500 <div class="doc_subsubsection">
501 <a name="doInitialization_mod">The <tt>doInitialization(Module &)</tt>
505 <div class="doc_text">
508 <b>virtual bool</b> doInitialization(Module &M);
511 <p>The <tt>doIninitialize</tt> method is allowed to do most of the things that
512 <tt>FunctionPass</tt>'s are not allowed to do. They can add and remove
513 functions, get pointers to functions, etc. The <tt>doInitialization</tt> method
514 is designed to do simple initialization type of stuff that does not depend on
515 the functions being processed. The <tt>doInitialization</tt> method call is not
516 scheduled to overlap with any other pass executions (thus it should be very
519 <p>A good example of how this method should be used is the <a
520 href="http://llvm.cs.uiuc.edu/doxygen/LowerAllocations_8cpp-source.html">LowerAllocations</a>
521 pass. This pass converts <tt>malloc</tt> and <tt>free</tt> instructions into
522 platform dependent <tt>malloc()</tt> and <tt>free()</tt> function calls. It
523 uses the <tt>doInitialization</tt> method to get a reference to the malloc and
524 free functions that it needs, adding prototypes to the module if necessary.</p>
528 <!-- _______________________________________________________________________ -->
529 <div class="doc_subsubsection">
530 <a name="runOnFunction">The <tt>runOnFunction</tt> method</a>
533 <div class="doc_text">
536 <b>virtual bool</b> runOnFunction(Function &F) = 0;
539 <p>The <tt>runOnFunction</tt> method must be implemented by your subclass to do
540 the transformation or analysis work of your pass. As usual, a true value should
541 be returned if the function is modified.</p>
545 <!-- _______________________________________________________________________ -->
546 <div class="doc_subsubsection">
547 <a name="doFinalization_mod">The <tt>doFinalization(Module
548 &)</tt> method</a>
551 <div class="doc_text">
554 <b>virtual bool</b> doFinalization(Module &M);
557 <p>The <tt>doFinalization</tt> method is an infrequently used method that is
558 called when the pass framework has finished calling <a
559 href="#runOnFunction"><tt>runOnFunction</tt></a> for every function in the
560 program being compiled.</p>
564 <!-- ======================================================================= -->
565 <div class="doc_subsection">
566 <a name="BasicBlockPass">The <tt>BasicBlockPass</tt> class</a>
569 <div class="doc_text">
571 <p><tt>BasicBlockPass</tt>'s are just like <a
572 href="#FunctionPass"><tt>FunctionPass</tt></a>'s, except that they must limit
573 their scope of inspection and modification to a single basic block at a time.
574 As such, they are <b>not</b> allowed to do any of the following:</p>
577 <li>Modify or inspect any basic blocks outside of the current one</li>
578 <li>Maintain state across invocations of
579 <a href="#runOnBasicBlock"><tt>runOnBasicBlock</tt></a></li>
580 <li>Modify the constrol flow graph (by altering terminator instructions)</li>
581 <li>Any of the things verboten for
582 <a href="#FunctionPass"><tt>FunctionPass</tt></a>es.</li>
585 <p><tt>BasicBlockPass</tt>es are useful for traditional local and "peephole"
586 optimizations. They may override the same <a
587 href="#doInitialization_mod"><tt>doInitialization(Module &)</tt></a> and <a
588 href="#doFinalization_mod"><tt>doFinalization(Module &)</tt></a> methods that <a
589 href="#FunctionPass"><tt>FunctionPass</tt></a>'s have, but also have the following virtual methods that may also be implemented:</p>
593 <!-- _______________________________________________________________________ -->
594 <div class="doc_subsubsection">
595 <a name="doInitialization_fn">The <tt>doInitialization(Function
596 &)</tt> method</a>
599 <div class="doc_text">
602 <b>virtual bool</b> doInitialization(Function &F);
605 <p>The <tt>doIninitialize</tt> method is allowed to do most of the things that
606 <tt>BasicBlockPass</tt>'s are not allowed to do, but that
607 <tt>FunctionPass</tt>'s can. The <tt>doInitialization</tt> method is designed
608 to do simple initialization type of stuff that does not depend on the
609 BasicBlocks being processed. The <tt>doInitialization</tt> method call is not
610 scheduled to overlap with any other pass executions (thus it should be very
615 <!-- _______________________________________________________________________ -->
616 <div class="doc_subsubsection">
617 <a name="runOnBasicBlock">The <tt>runOnBasicBlock</tt> method</a>
620 <div class="doc_text">
623 <b>virtual bool</b> runOnBasicBlock(BasicBlock &BB) = 0;
626 <p>Override this function to do the work of the <tt>BasicBlockPass</tt>. This
627 function is not allowed to inspect or modify basic blocks other than the
628 parameter, and are not allowed to modify the CFG. A true value must be returned
629 if the basic block is modified.</p>
633 <!-- _______________________________________________________________________ -->
634 <div class="doc_subsubsection">
635 <a name="doFinalization_fn">The <tt>doFinalization(Function &)</tt>
639 <div class="doc_text">
642 <b>virtual bool</b> doFinalization(Function &F);
645 <p>The <tt>doFinalization</tt> method is an infrequently used method that is
646 called when the pass framework has finished calling <a
647 href="#runOnBasicBlock"><tt>runOnBasicBlock</tt></a> for every BasicBlock in the
648 program being compiled. This can be used to perform per-function
653 <!-- ======================================================================= -->
654 <div class="doc_subsection">
655 <a name="MachineFunctionPass">The <tt>MachineFunctionPass</tt> class</a>
658 <div class="doc_text">
660 <p>A <tt>MachineFunctionPass</tt> is a part of the LLVM code generator that
661 executes on the machine-dependent representation of each LLVM function in the
662 program. A <tt>MachineFunctionPass</tt> is also a <tt>FunctionPass</tt>, so all
663 the restrictions that apply to a <tt>FunctionPass</tt> also apply to it.
664 <tt>MachineFunctionPass</tt>es also have additional restrictions. In particular,
665 <tt>MachineFunctionPass</tt>es are not allowed to do any of the following:</p>
668 <li>Modify any LLVM Instructions, BasicBlocks or Functions.</li>
669 <li>Modify a MachineFunction other than the one currently being processed.</li>
670 <li>Add or remove MachineFunctions from the current Module.</li>
671 <li>Add or remove global variables from the current Module.</li>
672 <li>Maintain state across invocations of <a
673 href="#runOnMachineFunction"><tt>runOnMachineFunction</tt></a> (including global
679 <!-- _______________________________________________________________________ -->
680 <div class="doc_subsubsection">
681 <a name="runOnMachineFunction">The <tt>runOnMachineFunction(MachineFunction
682 &MF)</tt> method</a>
685 <div class="doc_text">
688 <b>virtual bool</b> runOnMachineFunction(MachineFunction &MF) = 0;
691 <p><tt>runOnMachineFunction</tt> can be considered the main entry point of a
692 <tt>MachineFunctionPass</tt>; that is, you should override this method to do the
693 work of your <tt>MachineFunctionPass</tt>.</p>
695 <p>The <tt>runOnMachineFunction</tt> method is called on every
696 <tt>MachineFunction</tt> in a <tt>Module</tt>, so that the
697 <tt>MachineFunctionPass</tt> may perform optimizations on the machine-dependent
698 representation of the function. If you want to get at the LLVM <tt>Function</tt>
699 for the <tt>MachineFunction</tt> you're working on, use
700 <tt>MachineFunction</tt>'s <tt>getFunction()</tt> accessor method -- but
701 remember, you may not modify the LLVM <tt>Function</tt> or its contents from a
702 <tt>MachineFunctionPass</tt>.</p>
706 <!-- *********************************************************************** -->
707 <div class="doc_section">
708 <a name="registration">Pass registration</a>
710 <!-- *********************************************************************** -->
712 <div class="doc_text">
714 <p>In the <a href="#basiccode">Hello World</a> example pass we illustrated how
715 pass registration works, and discussed some of the reasons that it is used and
716 what it does. Here we discuss how and why passes are registered.</p>
718 <p>Passes can be registered in several different ways. Depending on the general
719 classification of the pass, you should use one of the following templates to
720 register the pass:</p>
723 <li><b><tt>RegisterOpt</tt></b> - This template should be used when you are
724 registering a pass that logically should be available for use in the
725 '<tt>opt</tt>' utility.</li>
727 <li><b><tt>RegisterAnalysis</tt></b> - This template should be used when you are
728 registering a pass that logically should be available for use in the
729 '<tt>analyze</tt>' utility.</li>
731 <li><b><tt>RegisterPass</tt></b> - This is the generic form of the
732 <tt>Register*</tt> templates that should be used if you want your pass listed by
733 multiple or no utilities. This template takes an extra third argument that
734 specifies which tools it should be listed in. See the <a
735 href="http://llvm.cs.uiuc.edu/doxygen/PassSupport_8h-source.html">PassSupport.h</a>
736 file for more information.</li>
740 <p>Regardless of how you register your pass, you must specify at least two
741 parameters. The first parameter is the name of the pass that is to be used on
742 the command line to specify that the pass should be added to a program (for
743 example <tt>opt</tt> or <tt>analyze</tt>). The second argument is the name of
744 the pass, which is to be used for the <tt>--help</tt> output of programs, as
745 well as for debug output generated by the <tt>--debug-pass</tt> option.</p>
747 <p>If a pass is registered to be used by the <tt>analyze</tt> utility, you
748 should implement the virtual <tt>print</tt> method:</p>
752 <!-- _______________________________________________________________________ -->
753 <div class="doc_subsubsection">
754 <a name="print">The <tt>print</tt> method</a>
757 <div class="doc_text">
760 <b>virtual void</b> print(std::ostream &O, <b>const</b> Module *M) <b>const</b>;
763 <p>The <tt>print</tt> method must be implemented by "analyses" in order to print
764 a human readable version of the analysis results. This is useful for debugging
765 an analysis itself, as well as for other people to figure out how an analysis
766 works. The <tt>analyze</tt> tool uses this method to generate its output.</p>
768 <p>The <tt>ostream</tt> parameter specifies the stream to write the results on,
769 and the <tt>Module</tt> parameter gives a pointer to the top level module of the
770 program that has been analyzed. Note however that this pointer may be null in
771 certain circumstances (such as calling the <tt>Pass::dump()</tt> from a
772 debugger), so it should only be used to enhance debug output, it should not be
777 <!-- *********************************************************************** -->
778 <div class="doc_section">
779 <a name="interaction">Specifying interactions between passes</a>
781 <!-- *********************************************************************** -->
783 <div class="doc_text">
785 <p>One of the main responsibilities of the <tt>PassManager</tt> is the make sure
786 that passes interact with each other correctly. Because <tt>PassManager</tt>
787 tries to <a href="#passmanager">optimize the execution of passes</a> it must
788 know how the passes interact with each other and what dependencies exist between
789 the various passes. To track this, each pass can declare the set of passes that
790 are required to be executed before the current pass, and the passes which are
791 invalidated by the current pass.</p>
793 <p>Typically this functionality is used to require that analysis results are
794 computed before your pass is run. Running arbitrary transformation passes can
795 invalidate the computed analysis results, which is what the invalidation set
796 specifies. If a pass does not implement the <tt><a
797 href="#getAnalysisUsage">getAnalysisUsage</a></tt> method, it defaults to not
798 having any prerequisite passes, and invalidating <b>all</b> other passes.</p>
802 <!-- _______________________________________________________________________ -->
803 <div class="doc_subsubsection">
804 <a name="getAnalysisUsage">The <tt>getAnalysisUsage</tt> method</a>
807 <div class="doc_text">
810 <b>virtual void</b> getAnalysisUsage(AnalysisUsage &Info) <b>const</b>;
813 <p>By implementing the <tt>getAnalysisUsage</tt> method, the required and
814 invalidated sets may be specified for your transformation. The implementation
815 should fill in the <tt><a
816 href="http://llvm.cs.uiuc.edu/doxygen/classllvm_1_1AnalysisUsage.html">AnalysisUsage</a></tt>
817 object with information about which passes are required and not invalidated. To
818 do this, a pass may call any of the following methods on the AnalysisUsage
822 <!-- _______________________________________________________________________ -->
823 <div class="doc_subsubsection">
824 <a name="AU::addRequired">The <tt>AnalysisUsage::addRequired<></tt> and <tt>AnalysisUsage::addRequiredTransitive<></tt> methods</a>
827 <div class="doc_text">
829 If you pass requires a previous pass to be executed (an analysis for example),
830 it can use one of these methods to arrange for it to be run before your pass.
831 LLVM has many different types of analyses and passes that can be required,
832 spaning the range from <tt>DominatorSet</tt> to <tt>BreakCriticalEdges</tt>.
833 requiring <tt>BreakCriticalEdges</tt>, for example, guarantees that there will
834 be no critical edges in the CFG when your pass has been run.
838 Some analyses chain to other analyses to do their job. For example, an <a
839 href="AliasAnalysis.html">AliasAnalysis</a> implementation is required to <a
840 href="AliasAnalysis.html#chaining">chain</a> to other alias analysis passes. In
841 cases where analyses chain, the <tt>addRequiredTransitive</tt> method should be
842 used instead of the <tt>addRequired</tt> method. This informs the PassManager
843 that the transitively required pass should be alive as long as the requiring
848 <!-- _______________________________________________________________________ -->
849 <div class="doc_subsubsection">
850 <a name="AU::addPreserved">The <tt>AnalysisUsage::addPreserved<></tt> method</a>
853 <div class="doc_text">
855 One of the jobs of the PassManager is to optimize how and when analyses are run.
856 In particular, it attempts to avoid recomputing data unless it needs to. For
857 this reason, passes are allowed to declare that they preserve (i.e., they don't
858 invalidate) an existing analysis if it's available. For example, a simple
859 constant folding pass would not modify the CFG, so it can't possible effect the
860 results of dominator analysis. By default, all passes are assumed to invalidate
865 The <tt>AnalysisUsage</tt> class provides several methods which are useful in
866 certain circumstances that are related to <tt>addPreserved</tt>. In particular,
867 the <tt>setPreservesAll</tt> method can be called to indicate that the pass does
868 not modify the LLVM program at all (which is true for analyses), and the
869 <tt>setPreservesCFG</tt> method can be used by transformations that change
870 instructions in the program but do not modify the CFG or terminator instructions
871 (note that this property is implicitly set for <a
872 href="#BasicBlockPass">BasicBlockPass</a>'s).
876 <tt>addPreserved</tt> is particularly useful for transformations like
877 <tt>BreakCriticalEdges</tt>. This pass knows how to update a small set of loop
878 and dominator related analyses if they exist, so it can preserve them, despite
879 the fact that it hacks on the CFG.
883 <!-- _______________________________________________________________________ -->
884 <div class="doc_subsubsection">
885 <a name="AU::examples">Example implementations of <tt>getAnalysisUsage</tt></a>
888 <div class="doc_text">
891 <i>// This is an example implementation from an analysis, which does not modify
892 // the program at all, yet has a prerequisite.</i>
893 <b>void</b> <a href="http://llvm.cs.uiuc.edu/doxygen/structllvm_1_1PostDominanceFrontier.html">PostDominanceFrontier</a>::getAnalysisUsage(AnalysisUsage &AU) <b>const</b> {
894 AU.setPreservesAll();
895 AU.addRequired<<a href="http://llvm.cs.uiuc.edu/doxygen/structllvm_1_1PostDominatorTree.html">PostDominatorTree</a>>();
902 <i>// This example modifies the program, but does not modify the CFG</i>
903 <b>void</b> <a href="http://llvm.cs.uiuc.edu/doxygen/structLICM.html">LICM</a>::getAnalysisUsage(AnalysisUsage &AU) <b>const</b> {
904 AU.setPreservesCFG();
905 AU.addRequired<<a href="http://llvm.cs.uiuc.edu/doxygen/classllvm_1_1LoopInfo.html">LoopInfo</a>>();
911 <!-- _______________________________________________________________________ -->
912 <div class="doc_subsubsection">
913 <a name="getAnalysis">The <tt>getAnalysis<></tt> and <tt>getAnalysisToUpdate<></tt> methods</a>
916 <div class="doc_text">
918 <p>The <tt>Pass::getAnalysis<></tt> method is automatically inherited by
919 your class, providing you with access to the passes that you declared that you
920 required with the <a href="#getAnalysisUsage"><tt>getAnalysisUsage</tt></a>
921 method. It takes a single template argument that specifies which pass class you
922 want, and returns a reference to that pass. For example:</p>
925 bool LICM::runOnFunction(Function &F) {
926 LoopInfo &LI = getAnalysis<LoopInfo>();
931 <p>This method call returns a reference to the pass desired. You may get a
932 runtime assertion failure if you attempt to get an analysis that you did not
933 declare as required in your <a
934 href="#getAnalysisUsage"><tt>getAnalysisUsage</tt></a> implementation. This
935 method can be called by your <tt>run*</tt> method implementation, or by any
936 other local method invoked by your <tt>run*</tt> method.</p>
939 If your pass is capable of updating analyses if they exist (e.g.,
940 <tt>BreakCriticalEdges</tt>, as described above), you can use the
941 <tt>getAnalysisToUpdate</tt> method, which returns a pointer to the analysis if
942 it is active. For example:</p>
946 if (DominatorSet *DS = getAnalysisToUpdate<DominatorSet>()) {
947 <i>// A DominatorSet is active. This code will update it.</i>
954 <!-- *********************************************************************** -->
955 <div class="doc_section">
956 <a name="analysisgroup">Implementing Analysis Groups</a>
958 <!-- *********************************************************************** -->
960 <div class="doc_text">
962 <p>Now that we understand the basics of how passes are defined, how the are
963 used, and how they are required from other passes, it's time to get a little bit
964 fancier. All of the pass relationships that we have seen so far are very
965 simple: one pass depends on one other specific pass to be run before it can run.
966 For many applications, this is great, for others, more flexibility is
969 <p>In particular, some analyses are defined such that there is a single simple
970 interface to the analysis results, but multiple ways of calculating them.
971 Consider alias analysis for example. The most trivial alias analysis returns
972 "may alias" for any alias query. The most sophisticated analysis a
973 flow-sensitive, context-sensitive interprocedural analysis that can take a
974 significant amount of time to execute (and obviously, there is a lot of room
975 between these two extremes for other implementations). To cleanly support
976 situations like this, the LLVM Pass Infrastructure supports the notion of
981 <!-- _______________________________________________________________________ -->
982 <div class="doc_subsubsection">
983 <a name="agconcepts">Analysis Group Concepts</a>
986 <div class="doc_text">
988 <p>An Analysis Group is a single simple interface that may be implemented by
989 multiple different passes. Analysis Groups can be given human readable names
990 just like passes, but unlike passes, they need not derive from the <tt>Pass</tt>
991 class. An analysis group may have one or more implementations, one of which is
992 the "default" implementation.</p>
994 <p>Analysis groups are used by client passes just like other passes are: the
995 <tt>AnalysisUsage::addRequired()</tt> and <tt>Pass::getAnalysis()</tt> methods.
996 In order to resolve this requirement, the <a href="#passmanager">PassManager</a>
997 scans the available passes to see if any implementations of the analysis group
998 are available. If none is available, the default implementation is created for
999 the pass to use. All standard rules for <A href="#interaction">interaction
1000 between passes</a> still apply.</p>
1002 <p>Although <a href="#registration">Pass Registration</a> is optional for normal
1003 passes, all analysis group implementations must be registered, and must use the
1004 <A href="#registerag"><tt>RegisterAnalysisGroup</tt></a> template to join the
1005 implementation pool. Also, a default implementation of the interface
1006 <b>must</b> be registered with <A
1007 href="#registerag"><tt>RegisterAnalysisGroup</tt></a>.</p>
1009 <p>As a concrete example of an Analysis Group in action, consider the <a
1010 href="http://llvm.cs.uiuc.edu/doxygen/classllvm_1_1AliasAnalysis.html">AliasAnalysis</a>
1011 analysis group. The default implementation of the alias analysis interface (the
1013 href="http://llvm.cs.uiuc.edu/doxygen/structBasicAliasAnalysis.html">basicaa</a></tt>
1014 pass) just does a few simple checks that don't require significant analysis to
1015 compute (such as: two different globals can never alias each other, etc).
1016 Passes that use the <tt><a
1017 href="http://llvm.cs.uiuc.edu/doxygen/classllvm_1_1AliasAnalysis.html">AliasAnalysis</a></tt>
1018 interface (for example the <tt><a
1019 href="http://llvm.cs.uiuc.edu/doxygen/structGCSE.html">gcse</a></tt> pass), do
1020 not care which implementation of alias analysis is actually provided, they just
1021 use the designated interface.</p>
1023 <p>From the user's perspective, commands work just like normal. Issuing the
1024 command '<tt>opt -gcse ...</tt>' will cause the <tt>basicaa</tt> class to be
1025 instantiated and added to the pass sequence. Issuing the command '<tt>opt
1026 -somefancyaa -gcse ...</tt>' will cause the <tt>gcse</tt> pass to use the
1027 <tt>somefancyaa</tt> alias analysis (which doesn't actually exist, it's just a
1028 hypothetical example) instead.</p>
1032 <!-- _______________________________________________________________________ -->
1033 <div class="doc_subsubsection">
1034 <a name="registerag">Using <tt>RegisterAnalysisGroup</tt></a>
1037 <div class="doc_text">
1039 <p>The <tt>RegisterAnalysisGroup</tt> template is used to register the analysis
1040 group itself as well as add pass implementations to the analysis group. First,
1041 an analysis should be registered, with a human readable name provided for it.
1042 Unlike registration of passes, there is no command line argument to be specified
1043 for the Analysis Group Interface itself, because it is "abstract":</p>
1046 <b>static</b> RegisterAnalysisGroup<<a href="http://llvm.cs.uiuc.edu/doxygen/classllvm_1_1AliasAnalysis.html">AliasAnalysis</a>> A("<i>Alias Analysis</i>");
1049 <p>Once the analysis is registered, passes can declare that they are valid
1050 implementations of the interface by using the following code:</p>
1054 //<i> Analysis Group implementations <b>must</b> be registered normally...</i>
1055 RegisterOpt<FancyAA>
1056 B("<i>somefancyaa</i>", "<i>A more complex alias analysis implementation</i>");
1058 //<i> Declare that we implement the AliasAnalysis interface</i>
1059 RegisterAnalysisGroup<<a href="http://llvm.cs.uiuc.edu/doxygen/classllvm_1_1AliasAnalysis.html">AliasAnalysis</a>, FancyAA> C;
1063 <p>This just shows a class <tt>FancyAA</tt> that is registered normally, then
1064 uses the <tt>RegisterAnalysisGroup</tt> template to "join" the <tt><a
1065 href="http://llvm.cs.uiuc.edu/doxygen/classllvm_1_1AliasAnalysis.html">AliasAnalysis</a></tt>
1066 analysis group. Every implementation of an analysis group should join using
1067 this template. A single pass may join multiple different analysis groups with
1072 //<i> Analysis Group implementations <b>must</b> be registered normally...</i>
1073 RegisterOpt<<a href="http://llvm.cs.uiuc.edu/doxygen/structBasicAliasAnalysis.html">BasicAliasAnalysis</a>>
1074 D("<i>basicaa</i>", "<i>Basic Alias Analysis (default AA impl)</i>");
1076 //<i> Declare that we implement the AliasAnalysis interface</i>
1077 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;
1081 <p>Here we show how the default implementation is specified (using the extra
1082 argument to the <tt>RegisterAnalysisGroup</tt> template). There must be exactly
1083 one default implementation available at all times for an Analysis Group to be
1084 used. Here we declare that the <tt><a
1085 href="http://llvm.cs.uiuc.edu/doxygen/structBasicAliasAnalysis.html">BasicAliasAnalysis</a></tt>
1086 pass is the default implementation for the interface.</p>
1090 <!-- *********************************************************************** -->
1091 <div class="doc_section">
1092 <a name="passmanager">What PassManager does</a>
1094 <!-- *********************************************************************** -->
1096 <div class="doc_text">
1099 href="http://llvm.cs.uiuc.edu/doxygen/PassManager_8h-source.html"><tt>PassManager</tt></a>
1101 href="http://llvm.cs.uiuc.edu/doxygen/classllvm_1_1PassManager.html">class</a>
1102 takes a list of passes, ensures their <a href="#interaction">prerequisites</a>
1103 are set up correctly, and then schedules passes to run efficiently. All of the
1104 LLVM tools that run passes use the <tt>PassManager</tt> for execution of these
1107 <p>The <tt>PassManager</tt> does two main things to try to reduce the execution
1108 time of a series of passes:</p>
1111 <li><b>Share analysis results</b> - The PassManager attempts to avoid
1112 recomputing analysis results as much as possible. This means keeping track of
1113 which analyses are available already, which analyses get invalidated, and which
1114 analyses are needed to be run for a pass. An important part of work is that the
1115 <tt>PassManager</tt> tracks the exact lifetime of all analysis results, allowing
1116 it to <a href="#releaseMemory">free memory</a> allocated to holding analysis
1117 results as soon as they are no longer needed.</li>
1119 <li><b>Pipeline the execution of passes on the program</b> - The
1120 <tt>PassManager</tt> attempts to get better cache and memory usage behavior out
1121 of a series of passes by pipelining the passes together. This means that, given
1122 a series of consequtive <a href="#FunctionPass"><tt>FunctionPass</tt></a>'s, it
1123 will execute all of the <a href="#FunctionPass"><tt>FunctionPass</tt></a>'s on
1124 the first function, then all of the <a
1125 href="#FunctionPass"><tt>FunctionPass</tt></a>es on the second function,
1126 etc... until the entire program has been run through the passes.
1128 <p>This improves the cache behavior of the compiler, because it is only touching
1129 the LLVM program representation for a single function at a time, instead of
1130 traversing the entire program. It reduces the memory consumption of compiler,
1131 because, for example, only one <a
1132 href="http://llvm.cs.uiuc.edu/doxygen/structllvm_1_1DominatorSet.html"><tt>DominatorSet</tt></a>
1133 needs to be calculated at a time. This also makes it possible some <a
1134 href="#SMP">interesting enhancements</a> in the future.</p></li>
1138 <p>The effectiveness of the <tt>PassManager</tt> is influenced directly by how
1139 much information it has about the behaviors of the passes it is scheduling. For
1140 example, the "preserved" set is intentionally conservative in the face of an
1141 unimplemented <a href="#getAnalysisUsage"><tt>getAnalysisUsage</tt></a> method.
1142 Not implementing when it should be implemented will have the effect of not
1143 allowing any analysis results to live across the execution of your pass.</p>
1145 <p>The <tt>PassManager</tt> class exposes a <tt>--debug-pass</tt> command line
1146 options that is useful for debugging pass execution, seeing how things work, and
1147 diagnosing when you should be preserving more analyses than you currently are
1148 (To get information about all of the variants of the <tt>--debug-pass</tt>
1149 option, just type '<tt>opt --help-hidden</tt>').</p>
1151 <p>By using the <tt>--debug-pass=Structure</tt> option, for example, we can see
1152 how our <a href="#basiccode">Hello World</a> pass interacts with other passes.
1153 Lets try it out with the <tt>gcse</tt> and <tt>licm</tt> passes:</p>
1156 $ opt -load ../../../lib/Debug/libhello.so -gcse -licm --debug-pass=Structure < hello.bc > /dev/null
1158 Function Pass Manager
1159 Dominator Set Construction
1160 Immediate Dominators Construction
1161 Global Common Subexpression Elimination
1162 -- Immediate Dominators Construction
1163 -- Global Common Subexpression Elimination
1164 Natural Loop Construction
1165 Loop Invariant Code Motion
1166 -- Natural Loop Construction
1167 -- Loop Invariant Code Motion
1169 -- Dominator Set Construction
1175 <p>This output shows us when passes are constructed and when the analysis
1176 results are known to be dead (prefixed with '<tt>--</tt>'). Here we see that
1177 GCSE uses dominator and immediate dominator information to do its job. The LICM
1178 pass uses natural loop information, which uses dominator sets, but not immediate
1179 dominators. Because immediate dominators are no longer useful after the GCSE
1180 pass, it is immediately destroyed. The dominator sets are then reused to
1181 compute natural loop information, which is then used by the LICM pass.</p>
1183 <p>After the LICM pass, the module verifier runs (which is automatically added
1184 by the '<tt>opt</tt>' tool), which uses the dominator set to check that the
1185 resultant LLVM code is well formed. After it finishes, the dominator set
1186 information is destroyed, after being computed once, and shared by three
1189 <p>Lets see how this changes when we run the <a href="#basiccode">Hello
1190 World</a> pass in between the two passes:</p>
1193 $ opt -load ../../../lib/Debug/libhello.so -gcse -hello -licm --debug-pass=Structure < hello.bc > /dev/null
1195 Function Pass Manager
1196 Dominator Set Construction
1197 Immediate Dominators Construction
1198 Global Common Subexpression Elimination
1199 <b>-- Dominator Set Construction</b>
1200 -- Immediate Dominators Construction
1201 -- Global Common Subexpression Elimination
1202 <b> Hello World Pass
1204 Dominator Set Construction</b>
1205 Natural Loop Construction
1206 Loop Invariant Code Motion
1207 -- Natural Loop Construction
1208 -- Loop Invariant Code Motion
1210 -- Dominator Set Construction
1219 <p>Here we see that the <a href="#basiccode">Hello World</a> pass has killed the
1220 Dominator Set pass, even though it doesn't modify the code at all! To fix this,
1221 we need to add the following <a
1222 href="#getAnalysisUsage"><tt>getAnalysisUsage</tt></a> method to our pass:</p>
1225 <i>// We don't modify the program, so we preserve all analyses</i>
1226 <b>virtual void</b> getAnalysisUsage(AnalysisUsage &AU) <b>const</b> {
1227 AU.setPreservesAll();
1231 <p>Now when we run our pass, we get this output:</p>
1234 $ opt -load ../../../lib/Debug/libhello.so -gcse -hello -licm --debug-pass=Structure < hello.bc > /dev/null
1235 Pass Arguments: -gcse -hello -licm
1237 Function Pass Manager
1238 Dominator Set Construction
1239 Immediate Dominators Construction
1240 Global Common Subexpression Elimination
1241 -- Immediate Dominators Construction
1242 -- Global Common Subexpression Elimination
1245 Natural Loop Construction
1246 Loop Invariant Code Motion
1247 -- Loop Invariant Code Motion
1248 -- Natural Loop Construction
1250 -- Dominator Set Construction
1259 <p>Which shows that we don't accidentally invalidate dominator information
1260 anymore, and therefore do not have to compute it twice.</p>
1264 <!-- _______________________________________________________________________ -->
1265 <div class="doc_subsubsection">
1266 <a name="releaseMemory">The <tt>releaseMemory</tt> method</a>
1269 <div class="doc_text">
1272 <b>virtual void</b> releaseMemory();
1275 <p>The <tt>PassManager</tt> automatically determines when to compute analysis
1276 results, and how long to keep them around for. Because the lifetime of the pass
1277 object itself is effectively the entire duration of the compilation process, we
1278 need some way to free analysis results when they are no longer useful. The
1279 <tt>releaseMemory</tt> virtual method is the way to do this.</p>
1281 <p>If you are writing an analysis or any other pass that retains a significant
1282 amount of state (for use by another pass which "requires" your pass and uses the
1283 <a href="#getAnalysis">getAnalysis</a> method) you should implement
1284 <tt>releaseMEmory</tt> to, well, release the memory allocated to maintain this
1285 internal state. This method is called after the <tt>run*</tt> method for the
1286 class, before the next call of <tt>run*</tt> in your pass.</p>
1290 <!-- *********************************************************************** -->
1291 <div class="doc_section">
1292 <a name="debughints">Using GDB with dynamically loaded passes</a>
1294 <!-- *********************************************************************** -->
1296 <div class="doc_text">
1298 <p>Unfortunately, using GDB with dynamically loaded passes is not as easy as it
1299 should be. First of all, you can't set a breakpoint in a shared object that has
1300 not been loaded yet, and second of all there are problems with inlined functions
1301 in shared objects. Here are some suggestions to debugging your pass with
1304 <p>For sake of discussion, I'm going to assume that you are debugging a
1305 transformation invoked by <tt>opt</tt>, although nothing described here depends
1310 <!-- _______________________________________________________________________ -->
1311 <div class="doc_subsubsection">
1312 <a name="breakpoint">Setting a breakpoint in your pass</a>
1315 <div class="doc_text">
1317 <p>First thing you do is start <tt>gdb</tt> on the <tt>opt</tt> process:</p>
1322 Copyright 2000 Free Software Foundation, Inc.
1323 GDB is free software, covered by the GNU General Public License, and you are
1324 welcome to change it and/or distribute copies of it under certain conditions.
1325 Type "show copying" to see the conditions.
1326 There is absolutely no warranty for GDB. Type "show warranty" for details.
1327 This GDB was configured as "sparc-sun-solaris2.6"...
1331 <p>Note that <tt>opt</tt> has a lot of debugging information in it, so it takes
1332 time to load. Be patient. Since we cannot set a breakpoint in our pass yet
1333 (the shared object isn't loaded until runtime), we must execute the process, and
1334 have it stop before it invokes our pass, but after it has loaded the shared
1335 object. The most foolproof way of doing this is to set a breakpoint in
1336 <tt>PassManager::run</tt> and then run the process with the arguments you
1340 (gdb) <b>break PassManager::run</b>
1341 Breakpoint 1 at 0x2413bc: file Pass.cpp, line 70.
1342 (gdb) <b>run test.bc -load $(LLVMTOP)/llvm/lib/Debug/[libname].so -[passoption]</b>
1343 Starting program: opt test.bc -load $(LLVMTOP)/llvm/lib/Debug/[libname].so -[passoption]
1344 Breakpoint 1, PassManager::run (this=0xffbef174, M=@0x70b298) at Pass.cpp:70
1345 70 bool PassManager::run(Module &M) { return PM->run(M); }
1349 <p>Once the <tt>opt</tt> stops in the <tt>PassManager::run</tt> method you are
1350 now free to set breakpoints in your pass so that you can trace through execution
1351 or do other standard debugging stuff.</p>
1355 <!-- _______________________________________________________________________ -->
1356 <div class="doc_subsubsection">
1357 <a name="debugmisc">Miscellaneous Problems</a>
1360 <div class="doc_text">
1362 <p>Once you have the basics down, there are a couple of problems that GDB has,
1363 some with solutions, some without.</p>
1366 <li>Inline functions have bogus stack information. In general, GDB does a
1367 pretty good job getting stack traces and stepping through inline functions.
1368 When a pass is dynamically loaded however, it somehow completely loses this
1369 capability. The only solution I know of is to de-inline a function (move it
1370 from the body of a class to a .cpp file).</li>
1372 <li>Restarting the program breaks breakpoints. After following the information
1373 above, you have succeeded in getting some breakpoints planted in your pass. Nex
1374 thing you know, you restart the program (i.e., you type '<tt>run</tt>' again),
1375 and you start getting errors about breakpoints being unsettable. The only way I
1376 have found to "fix" this problem is to <tt>delete</tt> the breakpoints that are
1377 already set in your pass, run the program, and re-set the breakpoints once
1378 execution stops in <tt>PassManager::run</tt>.</li>
1382 <p>Hopefully these tips will help with common case debugging situations. If
1383 you'd like to contribute some tips of your own, just contact <a
1384 href="mailto:sabre@nondot.org">Chris</a>.</p>
1388 <!-- *********************************************************************** -->
1389 <div class="doc_section">
1390 <a name="future">Future extensions planned</a>
1392 <!-- *********************************************************************** -->
1394 <div class="doc_text">
1396 <p>Although the LLVM Pass Infrastructure is very capable as it stands, and does
1397 some nifty stuff, there are things we'd like to add in the future. Here is
1398 where we are going:</p>
1402 <!-- _______________________________________________________________________ -->
1403 <div class="doc_subsubsection">
1404 <a name="SMP">Multithreaded LLVM</a>
1407 <div class="doc_text">
1409 <p>Multiple CPU machines are becoming more common and compilation can never be
1410 fast enough: obviously we should allow for a multithreaded compiler. Because of
1411 the semantics defined for passes above (specifically they cannot maintain state
1412 across invocations of their <tt>run*</tt> methods), a nice clean way to
1413 implement a multithreaded compiler would be for the <tt>PassManager</tt> class
1414 to create multiple instances of each pass object, and allow the separate
1415 instances to be hacking on different parts of the program at the same time.</p>
1417 <p>This implementation would prevent each of the passes from having to implement
1418 multithreaded constructs, requiring only the LLVM core to have locking in a few
1419 places (for global resources). Although this is a simple extension, we simply
1420 haven't had time (or multiprocessor machines, thus a reason) to implement this.
1421 Despite that, we have kept the LLVM passes SMP ready, and you should too.</p>
1425 <!-- _______________________________________________________________________ -->
1426 <div class="doc_subsubsection">
1427 <a name="ModuleSource">A new <tt>ModuleSource</tt> interface</a>
1430 <div class="doc_text">
1432 <p>Currently, the <tt>PassManager</tt>'s <tt>run</tt> method takes a <tt><a
1433 href="http://llvm.cs.uiuc.edu/doxygen/classllvm_1_1Module.html">Module</a></tt>
1434 as input, and runs all of the passes on this module. The problem with this
1435 approach is that none of the <tt>PassManager</tt> features can be used for
1436 timing and debugging the actual <b>loading</b> of the module from disk or
1439 <p>To solve this problem, eventually the <tt>PassManager</tt> class will accept
1440 a <tt>ModuleSource</tt> object instead of a Module itself. When complete, this
1441 will also allow for streaming of functions out of the bytecode representation,
1442 allowing us to avoid holding the entire program in memory at once if we only are
1443 dealing with <a href="#FunctionPass">FunctionPass</a>es.</p>
1445 <p>As part of a different issue, eventually the bytecode loader will be extended
1446 to allow on-demand loading of functions from the bytecode representation, in
1447 order to better support the runtime reoptimizer. The bytecode format is already
1448 capable of this, the loader just needs to be reworked a bit.</p>
1452 <!-- _______________________________________________________________________ -->
1453 <div class="doc_subsubsection">
1454 <a name="PassFunctionPass"><tt>Pass</tt>es requiring <tt>FunctionPass</tt>es</a>
1457 <div class="doc_text">
1459 <p>Currently it is illegal for a <a href="#Pass"><tt>Pass</tt></a> to require a
1460 <a href="#FunctionPass"><tt>FunctionPass</tt></a>. This is because there is
1461 only one instance of the <a href="#FunctionPass"><tt>FunctionPass</tt></a>
1462 object ever created, thus nowhere to store information for all of the functions
1463 in the program at the same time. Although this has come up a couple of times
1464 before, this has always been worked around by factoring one big complicated pass
1465 into a global and an interprocedural part, both of which are distinct. In the
1466 future, it would be nice to have this though.</p>
1468 <p>Note that it is no problem for a <a
1469 href="#FunctionPass"><tt>FunctionPass</tt></a> to require the results of a <a
1470 href="#Pass"><tt>Pass</tt></a>, only the other way around.</p>
1474 <!-- *********************************************************************** -->
1477 <a href="http://jigsaw.w3.org/css-validator/check/referer"><img
1478 src="http://jigsaw.w3.org/css-validator/images/vcss" alt="Valid CSS!"></a>
1479 <a href="http://validator.w3.org/check/referer"><img
1480 src="http://www.w3.org/Icons/valid-html401" alt="Valid HTML 4.01!" /></a>
1482 <a href="mailto:sabre@nondot.org">Chris Lattner</a><br>
1483 <a href="http://llvm.cs.uiuc.edu">The LLVM Compiler Infrastructure</a><br>
1484 Last modified: $Date$