| blob | 4ae039d75396d58327ff7a3a9e857445993c435b |
3 <html>
4 <head>
8 </head>
9 <body>
12 Writing an LLVM Pass
13 </div>
15 <ol>
18 <ul>
22 </ul></li>
24 <ul>
27 <ul>
29 </ul></li>
31 <ul>
37 </ul></li>
39 <ul>
45 </ul></li>
47 <ul>
53 </ul></li>
55 <ul>
59 method</a></li>
62 </ul></li>
64 class</a>
65 <ul>
68 </ul></li>
69 </ul>
71 <ul>
73 </ul></li>
75 <ul>
77 method</a></li>
78 <li><a href="#AU::addRequired">The <tt>AnalysisUsage::addRequired<></tt> and <tt>AnalysisUsage::addRequiredTransitive<></tt> methods</a></li>
81 <li><a href="#getAnalysis">The <tt>getAnalysis<></tt> and <tt>getAnalysisToUpdate<></tt> methods</a></li>
82 </ul></li>
84 <ul>
87 </ul></li>
90 <ul>
92 </ul></li>
94 <ul>
97 </ul></li>
99 <ul>
102 </ul></li>
104 <ul>
106 </ul></li>
107 </ol>
112 </div>
114 <!-- *********************************************************************** -->
117 </div>
118 <!-- *********************************************************************** -->
122 <p>The LLVM Pass Framework is an important part of the LLVM system, because LLVM
123 passes are where most of the interesting parts of the compiler exist. Passes
124 perform the transformations and optimizations that make up the compiler, they
125 build the analysis results that are used by these transformations, and they are,
126 above all, a structuring technique for compiler code.</p>
130 class, which implement functionality by overriding virtual methods inherited
137 more information about what your pass does, and how it can be combined with
138 other passes. One of the main features of the LLVM Pass Framework is that it
139 schedules passes to run in an efficient way based on the constraints that your
140 pass meets (which are indicated by which class they derive from).</p>
142 <p>We start by showing you how to construct a pass, everything from setting up
143 the code, to compiling, loading, and executing it. After the basics are down,
144 more advanced features are discussed.</p>
146 </div>
148 <!-- *********************************************************************** -->
151 </div>
152 <!-- *********************************************************************** -->
157 is designed to simply print out the name of non-external functions that exist in
158 the program being compiled. It does not modify the program at all, it just
159 inspects it. The source code and files for this pass are available in the LLVM
162 </div>
164 <!-- ======================================================================= -->
167 </div>
171 <p>First, you need to create a new directory somewhere in the LLVM source
172 base. For this example, we'll assume that you made
174 (Makefile) that will compile the source code for the new pass. To do this,
176 <hr/>
179 # Makefile for hello pass
181 # Path to top level of LLVM heirarchy
182 LEVEL = ../../..
184 # Name of the library to build
185 LIBRARYNAME = Hello
187 # Make the shared library become a loadable module so the tools can
188 # dlopen/dlsym on the resulting library.
189 LOADABLE_MODULE = 1
191 # Tell the build system which LLVM libraries your pass needs. You'll probably
192 # need at least LLVMSystem.a, LLVMSupport.a, LLVMCore.a but possibly several
193 # others too.
194 LLVMLIBS = LLVMCore.a LLVMSupport.a LLVMSystem.a
196 # Include the makefile implementation stuff
197 include $(LEVEL)/Makefile.common
198 </pre></div>
201 directory are to be compiled and linked together into a
204 If your operating system uses a suffix other than .so (such as windows or
205 Mac OS/X), the appropriate extension will be used.</p>
207 <p>Now that we have the build scripts set up, we just need to write the code for
208 the pass itself.</p>
210 </div>
212 <!-- ======================================================================= -->
215 </div>
219 <p>Now that we have a way to compile our new pass, we just have to write it.
220 Start out with:</p>
225 </pre></div>
229 we are operating on <tt><a
235 </pre></div>
236 <p>... which is required because the functions from the include files
237 live in the llvm namespace.
238 </p>
244 </pre></div>
246 <p>... which starts out an anonymous namespace. Anonymous namespaces are to C++
247 what the "<tt>static</tt>" keyword is to C (at global scope). It makes the
248 things declared inside of the anonymous namespace only visible to the current
249 file. If you're not familiar with them, consult a decent C++ book for more
250 information.</p>
256 </pre></div><p>
260 The different builtin pass subclasses are described in detail <a
263 time.</p>
266 static char ID;
267 Hello() : FunctionPass((intptr_t)&ID) {}
268 </pre></div><p>
270 <p> This declares pass identifier used by LLVM to identify pass. This allows LLVM to
271 avoid using expensive C++ runtime information.</p>
277 }
279 </pre></div>
282 which overloads an abstract virtual method inherited from <a
284 to do our thing, so we just print out our message with the name of each
285 function.</p>
288 char Hello::ID = 0;
289 </pre></div>
291 <p> We initialize pass ID here. LLVM uses ID's address to identify pass so
292 initialization value is not important.</p>
297 </pre></div>
300 giving it a command line
314 static char ID;
315 Hello() : FunctionPass((intptr_t)&ID) {}
320 }
321 };
323 char Hello::ID = 0;
325 }
326 </pre></div>
329 command in the local directory and you should get a new
330 "<tt>Debug/lib/Hello.so</tt> file. Note that everything in this file is
331 contained in an anonymous namespace: this reflects the fact that passes are self
332 contained units that do not need external interfaces (although they can have
333 them) to be useful.</p>
335 </div>
337 <!-- ======================================================================= -->
340 </div>
344 <p>Now that you have a brand new shiny shared object file, we can use the
345 <tt>opt</tt> command to run an LLVM program through your pass. Because you
346 registered your pass with the <tt>RegisterPass</tt> template, you will be able to
347 use the <tt>opt</tt> tool to access it, once loaded.</p>
349 <p>To test it, follow the example at the end of the <a
351 LLVM. We can now run the bitcode file (<tt>hello.bc</tt>) for the program
352 through our transformation like this (or course, any bitcode file will
353 work):</p>
356 $ opt -load ../../../Debug/lib/Hello.so -hello < hello.bc > /dev/null
357 Hello: __main
358 Hello: puts
359 Hello: main
360 </pre></div>
362 <p>The '<tt>-load</tt>' option specifies that '<tt>opt</tt>' should load your
363 pass as a shared object, which makes '<tt>-hello</tt>' a valid command line
365 pass</a>). Because the hello pass does not modify the program in any
366 interesting way, we just throw away the result of <tt>opt</tt> (sending it to
367 <tt>/dev/null</tt>).</p>
369 <p>To see what happened to the other string you registered, try running
370 <tt>opt</tt> with the <tt>--help</tt> option:</p>
373 $ opt -load ../../../Debug/lib/Hello.so --help
374 OVERVIEW: llvm .bc -> .bc modular optimizer
376 USAGE: opt [options] <input bitcode>
378 OPTIONS:
379 Optimizations available:
380 ...
381 -funcresolve - Resolve Functions
382 -gcse - Global Common Subexpression Elimination
383 -globaldce - Dead Global Elimination
384 <b>-hello - Hello World Pass</b>
385 -indvars - Canonicalize Induction Variables
386 -inline - Function Integration/Inlining
387 -instcombine - Combine redundant instructions
388 ...
389 </pre></div>
391 <p>The pass name get added as the information string for your pass, giving some
392 documentation to users of <tt>opt</tt>. Now that you have a working pass, you
393 would go ahead and make it do the cool transformations you want. Once you get
394 it all working and tested, it may become useful to find out how fast your pass
396 line option (<tt>--time-passes</tt>) that allows you to get information about
397 the execution time of your pass along with the other passes you queue up. For
398 example:</p>
401 $ opt -load ../../../Debug/lib/Hello.so -hello -time-passes < hello.bc > /dev/null
402 Hello: __main
403 Hello: puts
404 Hello: main
405 ===============================================================================
406 ... Pass execution timing report ...
407 ===============================================================================
408 Total Execution Time: 0.02 seconds (0.0479059 wall clock)
410 ---User Time--- --System Time-- --User+System-- ---Wall Time--- --- Pass Name ---
411 0.0100 (100.0%) 0.0000 ( 0.0%) 0.0100 ( 50.0%) 0.0402 ( 84.0%) Bitcode Writer
412 0.0000 ( 0.0%) 0.0100 (100.0%) 0.0100 ( 50.0%) 0.0031 ( 6.4%) Dominator Set Construction
413 0.0000 ( 0.0%) 0.0000 ( 0.0%) 0.0000 ( 0.0%) 0.0013 ( 2.7%) Module Verifier
414 <b> 0.0000 ( 0.0%) 0.0000 ( 0.0%) 0.0000 ( 0.0%) 0.0033 ( 6.9%) Hello World Pass</b>
415 0.0100 (100.0%) 0.0100 (100.0%) 0.0200 (100.0%) 0.0479 (100.0%) TOTAL
416 </pre></div>
418 <p>As you can see, our implementation above is pretty fast :). The additional
419 passes listed are automatically inserted by the '<tt>opt</tt>' tool to verify
420 that the LLVM emitted by your pass is still valid and well formed LLVM, which
421 hasn't been broken somehow.</p>
423 <p>Now that you have seen the basics of the mechanics behind passes, we can talk
424 about some more details of how they work and how to use them.</p>
426 </div>
428 <!-- *********************************************************************** -->
431 </div>
432 <!-- *********************************************************************** -->
436 <p>One of the first things that you should do when designing a new pass is to
437 decide what class you should subclass for your pass. The <a
440 did not discuss why or when this should occur. Here we talk about the classes
441 available, from the most general to the most specific.</p>
443 <p>When choosing a superclass for your Pass, you should choose the <b>most
444 specific</b> class possible, while still being able to meet the requirements
445 listed. This gives the LLVM Pass Infrastructure information necessary to
446 optimize how passes are run, so that the resultant compiler isn't unneccesarily
447 slow.</p>
449 </div>
451 <!-- ======================================================================= -->
454 </div>
460 class. This pass type is used for passes that do not have to be run, do not
461 change state, and never need to be updated. This is not a normal type of
462 transformation or analysis, but can provide information about the current
463 compiler configuration.</p>
465 <p>Although this pass class is very infrequently used, it is important for
466 providing information about the current target machine being compiled for, and
467 other static information that can affect the various transformations.</p>
469 <p><tt>ImmutablePass</tt>es never invalidate other transformations, are never
472 </div>
474 <!-- ======================================================================= -->
477 </div>
483 class is the most general of all superclasses that you can use. Deriving from
484 <tt>ModulePass</tt> indicates that your pass uses the entire program as a unit,
485 refering to function bodies in no predictable order, or adding and removing
486 functions. Because nothing is known about the behavior of <tt>ModulePass</tt>
487 subclasses, no optimization can be done for their execution. A module pass
488 can use function level passes (e.g. dominators) using getAnalysis interface
489 <tt> getAnalysis<DominatorTree>(Function)</tt>. </p>
491 <p>To write a correct <tt>ModulePass</tt> subclass, derive from
492 <tt>ModulePass</tt> and overload the <tt>runOnModule</tt> method with the
493 following signature:</p>
495 </div>
497 <!-- _______________________________________________________________________ -->
500 </div>
505 <b>virtual bool</b> runOnModule(Module &M) = 0;
506 </pre></div>
508 <p>The <tt>runOnModule</tt> method performs the interesting work of the pass.
509 It should return true if the module was modified by the transformation and
510 false otherwise.</p>
512 </div>
514 <!-- ======================================================================= -->
517 </div>
523 is used by passes that need to traverse the program bottom-up on the call graph
524 (callees before callers). Deriving from CallGraphSCCPass provides some
525 mechanics for building and traversing the CallGraph, but also allows the system
526 to optimize execution of CallGraphSCCPass's. If your pass meets the
527 requirements outlined below, and doesn't meet the requirements of a <tt><a
530 <tt>CallGraphSCCPass</tt>.</p>
532 <p><b>TODO</b>: explain briefly what SCC, Tarjan's algo, and B-U mean.</p>
534 <p>To be explicit, <tt>CallGraphSCCPass</tt> subclasses are:</p>
536 <ol>
538 <li>... <em>not allowed</em> to modify any <tt>Function</tt>s that are not in
539 the current SCC.</li>
541 <li>... <em>not allowed</em> to inspect any Function's other than those in the
542 current SCC and the direct callees of the SCC.</li>
544 <li>... <em>required</em> to preserve the current CallGraph object, updating it
545 to reflect any changes made to the program.</li>
547 <li>... <em>not allowed</em> to add or remove SCC's from the current Module,
548 though they may change the contents of an SCC.</li>
550 <li>... <em>allowed</em> to add or remove global variables from the current
551 Module.</li>
553 <li>... <em>allowed</em> to maintain state across invocations of
555 </ol>
557 <p>Implementing a <tt>CallGraphSCCPass</tt> is slightly tricky in some cases
558 because it has to handle SCCs with more than one node in it. All of the virtual
559 methods described below should return true if they modified the program, or
560 false if they didn't.</p>
562 </div>
564 <!-- _______________________________________________________________________ -->
567 method</a>
568 </div>
573 <b>virtual bool</b> doInitialization(CallGraph &CG);
574 </pre></div>
576 <p>The <tt>doIninitialize</tt> method is allowed to do most of the things that
577 <tt>CallGraphSCCPass</tt>'s are not allowed to do. They can add and remove
578 functions, get pointers to functions, etc. The <tt>doInitialization</tt> method
579 is designed to do simple initialization type of stuff that does not depend on
580 the SCCs being processed. The <tt>doInitialization</tt> method call is not
581 scheduled to overlap with any other pass executions (thus it should be very
582 fast).</p>
584 </div>
586 <!-- _______________________________________________________________________ -->
589 </div>
594 <b>virtual bool</b> runOnSCC(const std::vector<CallGraphNode *> &SCCM) = 0;
595 </pre></div>
597 <p>The <tt>runOnSCC</tt> method performs the interesting work of the pass, and
598 should return true if the module was modified by the transformation, false
599 otherwise.</p>
601 </div>
603 <!-- _______________________________________________________________________ -->
606 &)</tt> method</a>
607 </div>
612 <b>virtual bool</b> doFinalization(CallGraph &CG);
613 </pre></div>
615 <p>The <tt>doFinalization</tt> method is an infrequently used method that is
616 called when the pass framework has finished calling <a
618 program being compiled.</p>
620 </div>
622 <!-- ======================================================================= -->
625 </div>
629 <p>In contrast to <tt>ModulePass</tt> subclasses, <tt><a
631 subclasses do have a predictable, local behavior that can be expected by the
632 system. All <tt>FunctionPass</tt> execute on each function in the program
633 independent of all of the other functions in the program.
634 <tt>FunctionPass</tt>'s do not require that they are executed in a particular
635 order, and <tt>FunctionPass</tt>'s do not modify external functions.</p>
637 <p>To be explicit, <tt>FunctionPass</tt> subclasses are not allowed to:</p>
639 <ol>
640 <li>Modify a Function other than the one currently being processed.</li>
641 <li>Add or remove Function's from the current Module.</li>
642 <li>Add or remove global variables from the current Module.</li>
643 <li>Maintain state across invocations of
645 </ol>
647 <p>Implementing a <tt>FunctionPass</tt> is usually straightforward (See the <a
649 may overload three virtual methods to do their work. All of these methods
650 should return true if they modified the program, or false if they didn't.</p>
652 </div>
654 <!-- _______________________________________________________________________ -->
657 method</a>
658 </div>
663 <b>virtual bool</b> doInitialization(Module &M);
664 </pre></div>
666 <p>The <tt>doIninitialize</tt> method is allowed to do most of the things that
667 <tt>FunctionPass</tt>'s are not allowed to do. They can add and remove
668 functions, get pointers to functions, etc. The <tt>doInitialization</tt> method
669 is designed to do simple initialization type of stuff that does not depend on
670 the functions being processed. The <tt>doInitialization</tt> method call is not
671 scheduled to overlap with any other pass executions (thus it should be very
672 fast).</p>
674 <p>A good example of how this method should be used is the <a
676 pass. This pass converts <tt>malloc</tt> and <tt>free</tt> instructions into
677 platform dependent <tt>malloc()</tt> and <tt>free()</tt> function calls. It
678 uses the <tt>doInitialization</tt> method to get a reference to the malloc and
679 free functions that it needs, adding prototypes to the module if necessary.</p>
681 </div>
683 <!-- _______________________________________________________________________ -->
686 </div>
691 <b>virtual bool</b> runOnFunction(Function &F) = 0;
692 </pre></div><p>
694 <p>The <tt>runOnFunction</tt> method must be implemented by your subclass to do
695 the transformation or analysis work of your pass. As usual, a true value should
696 be returned if the function is modified.</p>
698 </div>
700 <!-- _______________________________________________________________________ -->
703 &)</tt> method</a>
704 </div>
709 <b>virtual bool</b> doFinalization(Module &M);
710 </pre></div>
712 <p>The <tt>doFinalization</tt> method is an infrequently used method that is
713 called when the pass framework has finished calling <a
715 program being compiled.</p>
717 </div>
719 <!-- ======================================================================= -->
722 </div>
726 <p> All <tt>LoopPass</tt> execute on each loop in the function independent of
727 all of the other loops in the function. <tt>LoopPass</tt> processes loops in
728 loop nest order such that outer most loop is processed last. </p>
730 <p> <tt>LoopPass</tt> subclasses are allowed to update loop nest using
731 <tt>LPPassManager</tt> interface. Implementing a loop pass is usually
732 straightforward. <tt>Looppass</tt>'s may overload three virtual methods to
733 do their work. All these methods should return true if they modified the
734 program, or false if they didn't. </p>
735 </div>
737 <!-- _______________________________________________________________________ -->
740 LPPassManager &)</tt>
741 method</a>
742 </div>
747 <b>virtual bool</b> doInitialization(Loop *, LPPassManager &LPM);
748 </pre></div>
750 <p>The <tt>doInitialization</tt> method is designed to do simple initialization
751 type of stuff that does not depend on the functions being processed. The
752 <tt>doInitialization</tt> method call is not scheduled to overlap with any
753 other pass executions (thus it should be very fast). LPPassManager
754 interface should be used to access Function or Module level analysis
755 information.</p>
757 </div>
760 <!-- _______________________________________________________________________ -->
763 </div>
768 <b>virtual bool</b> runOnLoop(Loop *, LPPassManager &LPM) = 0;
769 </pre></div><p>
771 <p>The <tt>runOnLoop</tt> method must be implemented by your subclass to do
772 the transformation or analysis work of your pass. As usual, a true value should
773 be returned if the function is modified. <tt>LPPassManager</tt> interface
774 should be used to update loop nest.</p>
776 </div>
778 <!-- _______________________________________________________________________ -->
781 </div>
786 <b>virtual bool</b> doFinalization();
787 </pre></div>
789 <p>The <tt>doFinalization</tt> method is an infrequently used method that is
790 called when the pass framework has finished calling <a
792 program being compiled. </p>
794 </div>
798 <!-- ======================================================================= -->
801 </div>
805 <p><tt>BasicBlockPass</tt>'s are just like <a
807 their scope of inspection and modification to a single basic block at a time.
808 As such, they are <b>not</b> allowed to do any of the following:</p>
810 <ol>
811 <li>Modify or inspect any basic blocks outside of the current one</li>
812 <li>Maintain state across invocations of
814 <li>Modify the control flow graph (by altering terminator instructions)</li>
815 <li>Any of the things forbidden for
817 </ol>
820 optimizations. They may override the same <a
823 href="#FunctionPass"><tt>FunctionPass</tt></a>'s have, but also have the following virtual methods that may also be implemented:</p>
825 </div>
827 <!-- _______________________________________________________________________ -->
830 &)</tt> method</a>
831 </div>
836 <b>virtual bool</b> doInitialization(Function &F);
837 </pre></div>
839 <p>The <tt>doIninitialize</tt> method is allowed to do most of the things that
840 <tt>BasicBlockPass</tt>'s are not allowed to do, but that
841 <tt>FunctionPass</tt>'s can. The <tt>doInitialization</tt> method is designed
842 to do simple initialization that does not depend on the
843 BasicBlocks being processed. The <tt>doInitialization</tt> method call is not
844 scheduled to overlap with any other pass executions (thus it should be very
845 fast).</p>
847 </div>
849 <!-- _______________________________________________________________________ -->
852 </div>
857 <b>virtual bool</b> runOnBasicBlock(BasicBlock &BB) = 0;
858 </pre></div>
860 <p>Override this function to do the work of the <tt>BasicBlockPass</tt>. This
861 function is not allowed to inspect or modify basic blocks other than the
862 parameter, and are not allowed to modify the CFG. A true value must be returned
863 if the basic block is modified.</p>
865 </div>
867 <!-- _______________________________________________________________________ -->
870 method</a>
871 </div>
876 <b>virtual bool</b> doFinalization(Function &F);
877 </pre></div>
879 <p>The <tt>doFinalization</tt> method is an infrequently used method that is
880 called when the pass framework has finished calling <a
882 program being compiled. This can be used to perform per-function
883 finalization.</p>
885 </div>
887 <!-- ======================================================================= -->
890 </div>
894 <p>A <tt>MachineFunctionPass</tt> is a part of the LLVM code generator that
895 executes on the machine-dependent representation of each LLVM function in the
896 program. A <tt>MachineFunctionPass</tt> is also a <tt>FunctionPass</tt>, so all
897 the restrictions that apply to a <tt>FunctionPass</tt> also apply to it.
898 <tt>MachineFunctionPass</tt>es also have additional restrictions. In particular,
899 <tt>MachineFunctionPass</tt>es are not allowed to do any of the following:</p>
901 <ol>
902 <li>Modify any LLVM Instructions, BasicBlocks or Functions.</li>
903 <li>Modify a MachineFunction other than the one currently being processed.</li>
904 <li>Add or remove MachineFunctions from the current Module.</li>
905 <li>Add or remove global variables from the current Module.</li>
906 <li>Maintain state across invocations of <a
908 data)</li>
909 </ol>
911 </div>
913 <!-- _______________________________________________________________________ -->
916 &MF)</tt> method</a>
917 </div>
922 <b>virtual bool</b> runOnMachineFunction(MachineFunction &MF) = 0;
923 </pre></div>
925 <p><tt>runOnMachineFunction</tt> can be considered the main entry point of a
926 <tt>MachineFunctionPass</tt>; that is, you should override this method to do the
927 work of your <tt>MachineFunctionPass</tt>.</p>
929 <p>The <tt>runOnMachineFunction</tt> method is called on every
930 <tt>MachineFunction</tt> in a <tt>Module</tt>, so that the
931 <tt>MachineFunctionPass</tt> may perform optimizations on the machine-dependent
932 representation of the function. If you want to get at the LLVM <tt>Function</tt>
933 for the <tt>MachineFunction</tt> you're working on, use
934 <tt>MachineFunction</tt>'s <tt>getFunction()</tt> accessor method -- but
935 remember, you may not modify the LLVM <tt>Function</tt> or its contents from a
936 <tt>MachineFunctionPass</tt>.</p>
938 </div>
940 <!-- *********************************************************************** -->
943 </div>
944 <!-- *********************************************************************** -->
949 pass registration works, and discussed some of the reasons that it is used and
950 what it does. Here we discuss how and why passes are registered.</p>
952 <p>As we saw above, passes are registered with the <b><tt>RegisterPass</tt></b>
953 template, which requires you to pass at least two
954 parameters. The first parameter is the name of the pass that is to be used on
955 the command line to specify that the pass should be added to a program (for
956 example, with <tt>opt</tt> or <tt>bugpoint</tt>). The second argument is the
957 name of the pass, which is to be used for the <tt>--help</tt> output of
958 programs, as
959 well as for debug output generated by the <tt>--debug-pass</tt> option.</p>
961 <p>If you want your pass to be easily dumpable, you should
962 implement the virtual <tt>print</tt> method:</p>
964 </div>
966 <!-- _______________________________________________________________________ -->
969 </div>
974 <b>virtual void</b> print(llvm::OStream &O, <b>const</b> Module *M) <b>const</b>;
975 </pre></div>
978 a human readable version of the analysis results. This is useful for debugging
979 an analysis itself, as well as for other people to figure out how an analysis
980 works. Use the <tt>opt -analyze</tt> argument to invoke this method.</p>
982 <p>The <tt>llvm::OStream</tt> parameter specifies the stream to write the results on,
983 and the <tt>Module</tt> parameter gives a pointer to the top level module of the
984 program that has been analyzed. Note however that this pointer may be null in
985 certain circumstances (such as calling the <tt>Pass::dump()</tt> from a
986 debugger), so it should only be used to enhance debug output, it should not be
987 depended on.</p>
989 </div>
991 <!-- *********************************************************************** -->
994 </div>
995 <!-- *********************************************************************** -->
999 <p>One of the main responsibilities of the <tt>PassManager</tt> is to make sure
1000 that passes interact with each other correctly. Because <tt>PassManager</tt>
1002 know how the passes interact with each other and what dependencies exist between
1003 the various passes. To track this, each pass can declare the set of passes that
1004 are required to be executed before the current pass, and the passes which are
1005 invalidated by the current pass.</p>
1007 <p>Typically this functionality is used to require that analysis results are
1008 computed before your pass is run. Running arbitrary transformation passes can
1009 invalidate the computed analysis results, which is what the invalidation set
1010 specifies. If a pass does not implement the <tt><a
1012 having any prerequisite passes, and invalidating <b>all</b> other passes.</p>
1014 </div>
1016 <!-- _______________________________________________________________________ -->
1019 </div>
1024 <b>virtual void</b> getAnalysisUsage(AnalysisUsage &Info) <b>const</b>;
1025 </pre></div>
1027 <p>By implementing the <tt>getAnalysisUsage</tt> method, the required and
1028 invalidated sets may be specified for your transformation. The implementation
1029 should fill in the <tt><a
1031 object with information about which passes are required and not invalidated. To
1032 do this, a pass may call any of the following methods on the AnalysisUsage
1033 object:</p>
1034 </div>
1036 <!-- _______________________________________________________________________ -->
1038 <a name="AU::addRequired">The <tt>AnalysisUsage::addRequired<></tt> and <tt>AnalysisUsage::addRequiredTransitive<></tt> methods</a>
1039 </div>
1042 <p>
1043 If your pass requires a previous pass to be executed (an analysis for example),
1044 it can use one of these methods to arrange for it to be run before your pass.
1045 LLVM has many different types of analyses and passes that can be required,
1046 spanning the range from <tt>DominatorSet</tt> to <tt>BreakCriticalEdges</tt>.
1047 Requiring <tt>BreakCriticalEdges</tt>, for example, guarantees that there will
1048 be no critical edges in the CFG when your pass has been run.
1049 </p>
1051 <p>
1052 Some analyses chain to other analyses to do their job. For example, an <a
1055 cases where analyses chain, the <tt>addRequiredTransitive</tt> method should be
1056 used instead of the <tt>addRequired</tt> method. This informs the PassManager
1057 that the transitively required pass should be alive as long as the requiring
1058 pass is.
1059 </p>
1060 </div>
1062 <!-- _______________________________________________________________________ -->
1065 </div>
1068 <p>
1069 One of the jobs of the PassManager is to optimize how and when analyses are run.
1070 In particular, it attempts to avoid recomputing data unless it needs to. For
1071 this reason, passes are allowed to declare that they preserve (i.e., they don't
1072 invalidate) an existing analysis if it's available. For example, a simple
1073 constant folding pass would not modify the CFG, so it can't possibly affect the
1074 results of dominator analysis. By default, all passes are assumed to invalidate
1075 all others.
1076 </p>
1078 <p>
1079 The <tt>AnalysisUsage</tt> class provides several methods which are useful in
1080 certain circumstances that are related to <tt>addPreserved</tt>. In particular,
1081 the <tt>setPreservesAll</tt> method can be called to indicate that the pass does
1082 not modify the LLVM program at all (which is true for analyses), and the
1083 <tt>setPreservesCFG</tt> method can be used by transformations that change
1084 instructions in the program but do not modify the CFG or terminator instructions
1085 (note that this property is implicitly set for <a
1087 </p>
1089 <p>
1090 <tt>addPreserved</tt> is particularly useful for transformations like
1091 <tt>BreakCriticalEdges</tt>. This pass knows how to update a small set of loop
1092 and dominator related analyses if they exist, so it can preserve them, despite
1093 the fact that it hacks on the CFG.
1094 </p>
1095 </div>
1097 <!-- _______________________________________________________________________ -->
1100 </div>
1105 <i>// This is an example implementation from an analysis, which does not modify
1106 // the program at all, yet has a prerequisite.</i>
1107 <b>void</b> <a href="http://llvm.org/doxygen/classllvm_1_1PostDominanceFrontier.html">PostDominanceFrontier</a>::getAnalysisUsage(AnalysisUsage &AU) <b>const</b> {
1108 AU.setPreservesAll();
1109 AU.addRequired<<a href="http://llvm.org/doxygen/classllvm_1_1PostDominatorTree.html">PostDominatorTree</a>>();
1110 }
1111 </pre></div>
1113 <p>and:</p>
1116 <i>// This example modifies the program, but does not modify the CFG</i>
1117 <b>void</b> <a href="http://llvm.org/doxygen/structLICM.html">LICM</a>::getAnalysisUsage(AnalysisUsage &AU) <b>const</b> {
1118 AU.setPreservesCFG();
1119 AU.addRequired<<a href="http://llvm.org/doxygen/classllvm_1_1LoopInfo.html">LoopInfo</a>>();
1120 }
1121 </pre></div>
1123 </div>
1125 <!-- _______________________________________________________________________ -->
1127 <a name="getAnalysis">The <tt>getAnalysis<></tt> and <tt>getAnalysisToUpdate<></tt> methods</a>
1128 </div>
1132 <p>The <tt>Pass::getAnalysis<></tt> method is automatically inherited by
1133 your class, providing you with access to the passes that you declared that you
1135 method. It takes a single template argument that specifies which pass class you
1136 want, and returns a reference to that pass. For example:</p>
1139 bool LICM::runOnFunction(Function &F) {
1140 LoopInfo &LI = getAnalysis<LoopInfo>();
1141 ...
1142 }
1143 </pre></div>
1145 <p>This method call returns a reference to the pass desired. You may get a
1146 runtime assertion failure if you attempt to get an analysis that you did not
1147 declare as required in your <a
1149 method can be called by your <tt>run*</tt> method implementation, or by any
1150 other local method invoked by your <tt>run*</tt> method.
1152 A module level pass can use function level analysis info using this interface.
1153 For example:</p>
1156 bool ModuleLevelPass::runOnModule(Module &M) {
1157 ...
1158 DominatorTree &DT = getAnalysis<DominatorTree>(Func);
1159 ...
1160 }
1161 </pre></div>
1163 <p>In above example, runOnFunction for DominatorTree is called by pass manager
1164 before returning a reference to the desired pass.</p>
1166 <p>
1167 If your pass is capable of updating analyses if they exist (e.g.,
1168 <tt>BreakCriticalEdges</tt>, as described above), you can use the
1169 <tt>getAnalysisToUpdate</tt> method, which returns a pointer to the analysis if
1170 it is active. For example:</p>
1173 ...
1174 if (DominatorSet *DS = getAnalysisToUpdate<DominatorSet>()) {
1175 <i>// A DominatorSet is active. This code will update it.</i>
1176 }
1177 ...
1178 </pre></div>
1180 </div>
1182 <!-- *********************************************************************** -->
1185 </div>
1186 <!-- *********************************************************************** -->
1190 <p>Now that we understand the basics of how passes are defined, how they are
1191 used, and how they are required from other passes, it's time to get a little bit
1192 fancier. All of the pass relationships that we have seen so far are very
1193 simple: one pass depends on one other specific pass to be run before it can run.
1194 For many applications, this is great, for others, more flexibility is
1195 required.</p>
1197 <p>In particular, some analyses are defined such that there is a single simple
1198 interface to the analysis results, but multiple ways of calculating them.
1199 Consider alias analysis for example. The most trivial alias analysis returns
1201 flow-sensitive, context-sensitive interprocedural analysis that can take a
1202 significant amount of time to execute (and obviously, there is a lot of room
1203 between these two extremes for other implementations). To cleanly support
1204 situations like this, the LLVM Pass Infrastructure supports the notion of
1205 Analysis Groups.</p>
1207 </div>
1209 <!-- _______________________________________________________________________ -->
1212 </div>
1216 <p>An Analysis Group is a single simple interface that may be implemented by
1217 multiple different passes. Analysis Groups can be given human readable names
1218 just like passes, but unlike passes, they need not derive from the <tt>Pass</tt>
1219 class. An analysis group may have one or more implementations, one of which is
1222 <p>Analysis groups are used by client passes just like other passes are: the
1223 <tt>AnalysisUsage::addRequired()</tt> and <tt>Pass::getAnalysis()</tt> methods.
1225 scans the available passes to see if any implementations of the analysis group
1226 are available. If none is available, the default implementation is created for
1228 between passes</a> still apply.</p>
1231 passes, all analysis group implementations must be registered, and must use the
1233 implementation pool. Also, a default implementation of the interface
1234 <b>must</b> be registered with <A
1237 <p>As a concrete example of an Analysis Group in action, consider the <a
1239 analysis group. The default implementation of the alias analysis interface (the
1240 <tt><a
1242 pass) just does a few simple checks that don't require significant analysis to
1243 compute (such as: two different globals can never alias each other, etc).
1244 Passes that use the <tt><a
1246 interface (for example the <tt><a
1248 not care which implementation of alias analysis is actually provided, they just
1249 use the designated interface.</p>
1251 <p>From the user's perspective, commands work just like normal. Issuing the
1252 command '<tt>opt -gcse ...</tt>' will cause the <tt>basicaa</tt> class to be
1253 instantiated and added to the pass sequence. Issuing the command '<tt>opt
1254 -somefancyaa -gcse ...</tt>' will cause the <tt>gcse</tt> pass to use the
1255 <tt>somefancyaa</tt> alias analysis (which doesn't actually exist, it's just a
1256 hypothetical example) instead.</p>
1258 </div>
1260 <!-- _______________________________________________________________________ -->
1263 </div>
1267 <p>The <tt>RegisterAnalysisGroup</tt> template is used to register the analysis
1268 group itself as well as add pass implementations to the analysis group. First,
1269 an analysis should be registered, with a human readable name provided for it.
1270 Unlike registration of passes, there is no command line argument to be specified
1274 <b>static</b> RegisterAnalysisGroup<<a href="http://llvm.org/doxygen/classllvm_1_1AliasAnalysis.html">AliasAnalysis</a>> A("<i>Alias Analysis</i>");
1275 </pre></div>
1277 <p>Once the analysis is registered, passes can declare that they are valid
1278 implementations of the interface by using the following code:</p>
1281 <b>namespace</b> {
1282 //<i> Analysis Group implementations <b>must</b> be registered normally...</i>
1283 RegisterPass<FancyAA>
1286 //<i> Declare that we implement the AliasAnalysis interface</i>
1287 RegisterAnalysisGroup<<a href="http://llvm.org/doxygen/classllvm_1_1AliasAnalysis.html">AliasAnalysis</a>> C(B);
1288 }
1289 </pre></div>
1291 <p>This just shows a class <tt>FancyAA</tt> that is registered normally, then
1294 analysis group. Every implementation of an analysis group should join using
1295 this template. A single pass may join multiple different analysis groups with
1296 no problem.</p>
1299 <b>namespace</b> {
1300 //<i> Analysis Group implementations <b>must</b> be registered normally...</i>
1301 RegisterPass<<a href="http://llvm.org/doxygen/structBasicAliasAnalysis.html">BasicAliasAnalysis</a>>
1304 //<i> Declare that we implement the AliasAnalysis interface</i>
1305 RegisterAnalysisGroup<<a href="http://llvm.org/doxygen/classllvm_1_1AliasAnalysis.html">AliasAnalysis</a>, <b>true</b>> E(D);
1306 }
1307 </pre></div>
1309 <p>Here we show how the default implementation is specified (using the extra
1310 argument to the <tt>RegisterAnalysisGroup</tt> template). There must be exactly
1311 one default implementation available at all times for an Analysis Group to be
1312 used. Only default implementation can derive from <tt>ImmutablePass</tt>.
1313 Here we declare that the
1314 <tt><a href="http://llvm.org/doxygen/structBasicAliasAnalysis.html">BasicAliasAnalysis</a></tt>
1315 pass is the default implementation for the interface.</p>
1317 </div>
1319 <!-- *********************************************************************** -->
1322 </div>
1323 <!-- *********************************************************************** -->
1326 <p>The <a
1328 class is designed to be an easy way to expose various success
1329 metrics from passes. These statistics are printed at the end of a
1330 run, when the -stats command line option is enabled on the command
1331 line. See the <a href="http://llvm.org/docs/ProgrammersManual.html#Statistic">Statistics section</a> in the Programmer's Manual for details.
1333 </div>
1336 <!-- *********************************************************************** -->
1339 </div>
1340 <!-- *********************************************************************** -->
1344 <p>The <a
1346 <a
1349 are set up correctly, and then schedules passes to run efficiently. All of the
1350 LLVM tools that run passes use the <tt>PassManager</tt> for execution of these
1351 passes.</p>
1353 <p>The <tt>PassManager</tt> does two main things to try to reduce the execution
1354 time of a series of passes:</p>
1356 <ol>
1357 <li><b>Share analysis results</b> - The PassManager attempts to avoid
1358 recomputing analysis results as much as possible. This means keeping track of
1359 which analyses are available already, which analyses get invalidated, and which
1360 analyses are needed to be run for a pass. An important part of work is that the
1361 <tt>PassManager</tt> tracks the exact lifetime of all analysis results, allowing
1363 results as soon as they are no longer needed.</li>
1365 <li><b>Pipeline the execution of passes on the program</b> - The
1366 <tt>PassManager</tt> attempts to get better cache and memory usage behavior out
1367 of a series of passes by pipelining the passes together. This means that, given
1370 the first function, then all of the <a
1372 etc... until the entire program has been run through the passes.
1374 <p>This improves the cache behavior of the compiler, because it is only touching
1375 the LLVM program representation for a single function at a time, instead of
1376 traversing the entire program. It reduces the memory consumption of compiler,
1377 because, for example, only one <a
1379 needs to be calculated at a time. This also makes it possible to implement
1380 some <a
1383 </ol>
1385 <p>The effectiveness of the <tt>PassManager</tt> is influenced directly by how
1386 much information it has about the behaviors of the passes it is scheduling. For
1389 Not implementing when it should be implemented will have the effect of not
1390 allowing any analysis results to live across the execution of your pass.</p>
1392 <p>The <tt>PassManager</tt> class exposes a <tt>--debug-pass</tt> command line
1393 options that is useful for debugging pass execution, seeing how things work, and
1394 diagnosing when you should be preserving more analyses than you currently are
1395 (To get information about all of the variants of the <tt>--debug-pass</tt>
1396 option, just type '<tt>opt --help-hidden</tt>').</p>
1398 <p>By using the <tt>--debug-pass=Structure</tt> option, for example, we can see
1400 Lets try it out with the <tt>gcse</tt> and <tt>licm</tt> passes:</p>
1403 $ opt -load ../../../Debug/lib/Hello.so -gcse -licm --debug-pass=Structure < hello.bc > /dev/null
1404 Module Pass Manager
1405 Function Pass Manager
1406 Dominator Set Construction
1407 Immediate Dominators Construction
1408 Global Common Subexpression Elimination
1409 -- Immediate Dominators Construction
1410 -- Global Common Subexpression Elimination
1411 Natural Loop Construction
1412 Loop Invariant Code Motion
1413 -- Natural Loop Construction
1414 -- Loop Invariant Code Motion
1415 Module Verifier
1416 -- Dominator Set Construction
1417 -- Module Verifier
1418 Bitcode Writer
1419 --Bitcode Writer
1420 </pre></div>
1422 <p>This output shows us when passes are constructed and when the analysis
1423 results are known to be dead (prefixed with '<tt>--</tt>'). Here we see that
1424 GCSE uses dominator and immediate dominator information to do its job. The LICM
1425 pass uses natural loop information, which uses dominator sets, but not immediate
1426 dominators. Because immediate dominators are no longer useful after the GCSE
1427 pass, it is immediately destroyed. The dominator sets are then reused to
1428 compute natural loop information, which is then used by the LICM pass.</p>
1430 <p>After the LICM pass, the module verifier runs (which is automatically added
1431 by the '<tt>opt</tt>' tool), which uses the dominator set to check that the
1432 resultant LLVM code is well formed. After it finishes, the dominator set
1433 information is destroyed, after being computed once, and shared by three
1434 passes.</p>
1437 World</a> pass in between the two passes:</p>
1440 $ opt -load ../../../Debug/lib/Hello.so -gcse -hello -licm --debug-pass=Structure < hello.bc > /dev/null
1441 Module Pass Manager
1442 Function Pass Manager
1443 Dominator Set Construction
1444 Immediate Dominators Construction
1445 Global Common Subexpression Elimination
1446 <b>-- Dominator Set Construction</b>
1447 -- Immediate Dominators Construction
1448 -- Global Common Subexpression Elimination
1449 <b> Hello World Pass
1450 -- Hello World Pass
1451 Dominator Set Construction</b>
1452 Natural Loop Construction
1453 Loop Invariant Code Motion
1454 -- Natural Loop Construction
1455 -- Loop Invariant Code Motion
1456 Module Verifier
1457 -- Dominator Set Construction
1458 -- Module Verifier
1459 Bitcode Writer
1460 --Bitcode Writer
1461 Hello: __main
1462 Hello: puts
1463 Hello: main
1464 </pre></div>
1467 Dominator Set pass, even though it doesn't modify the code at all! To fix this,
1468 we need to add the following <a
1472 <i>// We don't modify the program, so we preserve all analyses</i>
1473 <b>virtual void</b> getAnalysisUsage(AnalysisUsage &AU) <b>const</b> {
1474 AU.setPreservesAll();
1475 }
1476 </pre></div>
1478 <p>Now when we run our pass, we get this output:</p>
1481 $ opt -load ../../../Debug/lib/Hello.so -gcse -hello -licm --debug-pass=Structure < hello.bc > /dev/null
1482 Pass Arguments: -gcse -hello -licm
1483 Module Pass Manager
1484 Function Pass Manager
1485 Dominator Set Construction
1486 Immediate Dominators Construction
1487 Global Common Subexpression Elimination
1488 -- Immediate Dominators Construction
1489 -- Global Common Subexpression Elimination
1490 Hello World Pass
1491 -- Hello World Pass
1492 Natural Loop Construction
1493 Loop Invariant Code Motion
1494 -- Loop Invariant Code Motion
1495 -- Natural Loop Construction
1496 Module Verifier
1497 -- Dominator Set Construction
1498 -- Module Verifier
1499 Bitcode Writer
1500 --Bitcode Writer
1501 Hello: __main
1502 Hello: puts
1503 Hello: main
1504 </pre></div>
1506 <p>Which shows that we don't accidentally invalidate dominator information
1507 anymore, and therefore do not have to compute it twice.</p>
1509 </div>
1511 <!-- _______________________________________________________________________ -->
1514 </div>
1519 <b>virtual void</b> releaseMemory();
1520 </pre></div>
1522 <p>The <tt>PassManager</tt> automatically determines when to compute analysis
1523 results, and how long to keep them around for. Because the lifetime of the pass
1524 object itself is effectively the entire duration of the compilation process, we
1525 need some way to free analysis results when they are no longer useful. The
1526 <tt>releaseMemory</tt> virtual method is the way to do this.</p>
1528 <p>If you are writing an analysis or any other pass that retains a significant
1531 <tt>releaseMEmory</tt> to, well, release the memory allocated to maintain this
1532 internal state. This method is called after the <tt>run*</tt> method for the
1533 class, before the next call of <tt>run*</tt> in your pass.</p>
1535 </div>
1537 <!-- *********************************************************************** -->
1540 </div>
1541 <!-- *********************************************************************** -->
1545 <p><i>Size matters</i> when constructing production quality tools using llvm,
1546 both for the purposes of distribution, and for regulating the resident code size
1547 when running on the target system. Therefore, it becomes desirable to
1548 selectively use some passes, while omitting others and maintain the flexibility
1549 to change configurations later on. You want to be able to do all this, and,
1550 provide feedback to the user. This is where pass registration comes into
1551 play.</p>
1553 <p>The fundamental mechanisms for pass registration are the
1554 <tt>MachinePassRegistry</tt> class and subclasses of
1555 <tt>MachinePassRegistryNode</tt>.</p>
1557 <p>An instance of <tt>MachinePassRegistry</tt> is used to maintain a list of
1558 <tt>MachinePassRegistryNode</tt> objects. This instance maintains the list and
1559 communicates additions and deletions to the command line interface.</p>
1561 <p>An instance of <tt>MachinePassRegistryNode</tt> subclass is used to maintain
1562 information provided about a particular pass. This information includes the
1563 command line name, the command help string and the address of the function used
1564 to create an instance of the pass. A global static constructor of one of these
1565 instances <i>registers</i> with a corresponding <tt>MachinePassRegistry</tt>,
1566 the static destructor <i>unregisters</i>. Thus a pass that is statically linked
1567 in the tool will be registered at start up. A dynamically loaded pass will
1568 register on load and unregister at unload.</p>
1570 </div>
1572 <!-- _______________________________________________________________________ -->
1575 </div>
1579 <p>There are predefined registries to track instruction scheduling
1580 (<tt>RegisterScheduler</tt>) and register allocation (<tt>RegisterRegAlloc</tt>)
1581 machine passes. Here we will describe how to <i>register</i> a register
1582 allocator machine pass.</p>
1584 <p>Implement your register allocator machine pass. In your register allocator
1585 .cpp file add the following include;</p>
1589 </pre></div>
1591 <p>Also in your register allocator .cpp file, define a creator function in the
1592 form; </p>
1595 FunctionPass *createMyRegisterAllocator() {
1596 return new MyRegisterAllocator();
1597 }
1598 </pre></div>
1600 <p>Note that the signature of this function should match the type of
1601 <tt>RegisterRegAlloc::FunctionPassCtor</tt>. In the same file add the
1607 createMyRegisterAllocator);
1608 </pre></div>
1610 <p>Note the two spaces prior to the help string produces a tidy result on the
1611 --help query.</p>
1614 $ llc --help
1615 ...
1616 -regalloc - Register allocator to use: (default = linearscan)
1617 =linearscan - linear scan register allocator
1618 =local - local register allocator
1619 =simple - simple register allocator
1620 =myregalloc - my register allocator help string
1621 ...
1622 </pre></div>
1624 <p>And that's it. The user is now free to use <tt>-regalloc=myregalloc</tt> as
1625 an option. Registering instruction schedulers is similar except use the
1626 <tt>RegisterScheduler</tt> class. Note that the
1627 <tt>RegisterScheduler::FunctionPassCtor</tt> is significantly different from
1628 <tt>RegisterRegAlloc::FunctionPassCtor</tt>.</p>
1630 <p>To force the load/linking of your register allocator into the llc/lli tools,
1632 call line to <tt>llvm/Codegen/LinkAllCodegenComponents.h</tt>.</p>
1634 </div>
1637 <!-- _______________________________________________________________________ -->
1640 </div>
1644 <p>The easiest way to get started is to clone one of the existing registries; we
1645 recommend <tt>llvm/CodeGen/RegAllocRegistry.h</tt>. The key things to modify
1646 are the class name and the <tt>FunctionPassCtor</tt> type.</p>
1648 <p>Then you need to declare the registry. Example: if your pass registry is
1649 <tt>RegisterMyPasses</tt> then define;</p>
1652 MachinePassRegistry RegisterMyPasses::Registry;
1653 </pre></div>
1655 <p>And finally, declare the command line option for your passes. Example:</p>
1658 cl::opt<RegisterMyPasses::FunctionPassCtor, false,
1659 RegisterPassParser<RegisterMyPasses> >
1661 cl::init(&createDefaultMyPass),
1663 </pre></div>
1666 creator.</p>
1668 </div>
1670 <!-- *********************************************************************** -->
1673 </div>
1674 <!-- *********************************************************************** -->
1678 <p>Unfortunately, using GDB with dynamically loaded passes is not as easy as it
1679 should be. First of all, you can't set a breakpoint in a shared object that has
1680 not been loaded yet, and second of all there are problems with inlined functions
1681 in shared objects. Here are some suggestions to debugging your pass with
1682 GDB.</p>
1684 <p>For sake of discussion, I'm going to assume that you are debugging a
1685 transformation invoked by <tt>opt</tt>, although nothing described here depends
1686 on that.</p>
1688 </div>
1690 <!-- _______________________________________________________________________ -->
1693 </div>
1697 <p>First thing you do is start <tt>gdb</tt> on the <tt>opt</tt> process:</p>
1700 $ <b>gdb opt</b>
1701 GNU gdb 5.0
1702 Copyright 2000 Free Software Foundation, Inc.
1703 GDB is free software, covered by the GNU General Public License, and you are
1704 welcome to change it and/or distribute copies of it under certain conditions.
1708 (gdb)
1709 </pre></div>
1711 <p>Note that <tt>opt</tt> has a lot of debugging information in it, so it takes
1712 time to load. Be patient. Since we cannot set a breakpoint in our pass yet
1713 (the shared object isn't loaded until runtime), we must execute the process, and
1714 have it stop before it invokes our pass, but after it has loaded the shared
1715 object. The most foolproof way of doing this is to set a breakpoint in
1716 <tt>PassManager::run</tt> and then run the process with the arguments you
1717 want:</p>
1720 (gdb) <b>break llvm::PassManager::run</b>
1721 Breakpoint 1 at 0x2413bc: file Pass.cpp, line 70.
1722 (gdb) <b>run test.bc -load $(LLVMTOP)/llvm/Debug/lib/[libname].so -[passoption]</b>
1723 Starting program: opt test.bc -load $(LLVMTOP)/llvm/Debug/lib/[libname].so -[passoption]
1724 Breakpoint 1, PassManager::run (this=0xffbef174, M=@0x70b298) at Pass.cpp:70
1725 70 bool PassManager::run(Module &M) { return PM->run(M); }
1726 (gdb)
1727 </pre></div>
1729 <p>Once the <tt>opt</tt> stops in the <tt>PassManager::run</tt> method you are
1730 now free to set breakpoints in your pass so that you can trace through execution
1731 or do other standard debugging stuff.</p>
1733 </div>
1735 <!-- _______________________________________________________________________ -->
1738 </div>
1742 <p>Once you have the basics down, there are a couple of problems that GDB has,
1743 some with solutions, some without.</p>
1745 <ul>
1746 <li>Inline functions have bogus stack information. In general, GDB does a
1747 pretty good job getting stack traces and stepping through inline functions.
1748 When a pass is dynamically loaded however, it somehow completely loses this
1749 capability. The only solution I know of is to de-inline a function (move it
1750 from the body of a class to a .cpp file).</li>
1752 <li>Restarting the program breaks breakpoints. After following the information
1753 above, you have succeeded in getting some breakpoints planted in your pass. Nex
1754 thing you know, you restart the program (i.e., you type '<tt>run</tt>' again),
1755 and you start getting errors about breakpoints being unsettable. The only way I
1757 already set in your pass, run the program, and re-set the breakpoints once
1758 execution stops in <tt>PassManager::run</tt>.</li>
1760 </ul>
1762 <p>Hopefully these tips will help with common case debugging situations. If
1763 you'd like to contribute some tips of your own, just contact <a
1766 </div>
1768 <!-- *********************************************************************** -->
1771 </div>
1772 <!-- *********************************************************************** -->
1776 <p>Although the LLVM Pass Infrastructure is very capable as it stands, and does
1777 some nifty stuff, there are things we'd like to add in the future. Here is
1778 where we are going:</p>
1780 </div>
1782 <!-- _______________________________________________________________________ -->
1785 </div>
1789 <p>Multiple CPU machines are becoming more common and compilation can never be
1790 fast enough: obviously we should allow for a multithreaded compiler. Because of
1791 the semantics defined for passes above (specifically they cannot maintain state
1792 across invocations of their <tt>run*</tt> methods), a nice clean way to
1793 implement a multithreaded compiler would be for the <tt>PassManager</tt> class
1794 to create multiple instances of each pass object, and allow the separate
1795 instances to be hacking on different parts of the program at the same time.</p>
1797 <p>This implementation would prevent each of the passes from having to implement
1798 multithreaded constructs, requiring only the LLVM core to have locking in a few
1799 places (for global resources). Although this is a simple extension, we simply
1800 haven't had time (or multiprocessor machines, thus a reason) to implement this.
1801 Despite that, we have kept the LLVM passes SMP ready, and you should too.</p>
1803 </div>
1805 <!-- *********************************************************************** -->
1806 <hr>
1807 <address>
1815 Last modified: $Date$
1816 </address>
1818 </body>
1819 </html>