| blob | 331617937755f26c4cf39af62ebc52234e0d4eb2 |
3 <html>
4 <head>
8 </head>
9 <body>
11 <h1>
12 Writing an LLVM Pass
13 </h1>
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>
61 </ul></li>
63 <ul>
67 method</a></li>
70 </ul></li>
72 class</a>
73 <ul>
76 </ul></li>
77 </ul>
79 <ul>
81 </ul></li>
83 <ul>
85 method</a></li>
86 <li><a href="#AU::addRequired">The <tt>AnalysisUsage::addRequired<></tt> and <tt>AnalysisUsage::addRequiredTransitive<></tt> methods</a></li>
91 </ul></li>
93 <ul>
96 </ul></li>
99 <ul>
101 </ul></li>
103 <ul>
106 </ul></li>
108 <ul>
111 </ul></li>
113 <ul>
115 </ul></li>
116 </ol>
121 </div>
123 <!-- *********************************************************************** -->
124 <h2>
126 </h2>
127 <!-- *********************************************************************** -->
131 <p>The LLVM Pass Framework is an important part of the LLVM system, because LLVM
132 passes are where most of the interesting parts of the compiler exist. Passes
133 perform the transformations and optimizations that make up the compiler, they
134 build the analysis results that are used by these transformations, and they are,
135 above all, a structuring technique for compiler code.</p>
139 class, which implement functionality by overriding virtual methods inherited
147 more information about what your pass does, and how it can be combined with
148 other passes. One of the main features of the LLVM Pass Framework is that it
149 schedules passes to run in an efficient way based on the constraints that your
150 pass meets (which are indicated by which class they derive from).</p>
152 <p>We start by showing you how to construct a pass, everything from setting up
153 the code, to compiling, loading, and executing it. After the basics are down,
154 more advanced features are discussed.</p>
156 </div>
158 <!-- *********************************************************************** -->
159 <h2>
161 </h2>
162 <!-- *********************************************************************** -->
167 is designed to simply print out the name of non-external functions that exist in
168 the program being compiled. It does not modify the program at all, it just
169 inspects it. The source code and files for this pass are available in the LLVM
172 </div>
174 <!-- ======================================================================= -->
175 <h3>
177 </h3>
181 <p>First, configure and build LLVM. This needs to be done directly inside the
182 LLVM source tree rather than in a separate objects directory.
183 Next, you need to create a new directory somewhere in the LLVM source
184 base. For this example, we'll assume that you made
186 (Makefile) that will compile the source code for the new pass. To do this,
188 <hr>
191 # Makefile for hello pass
193 # Path to top level of LLVM hierarchy
194 LEVEL = ../../..
196 # Name of the library to build
197 LIBRARYNAME = Hello
199 # Make the shared library become a loadable module so the tools can
200 # dlopen/dlsym on the resulting library.
201 LOADABLE_MODULE = 1
203 # Include the makefile implementation stuff
204 include $(LEVEL)/Makefile.common
205 </pre></div>
208 directory are to be compiled and linked together into a shared object
211 If your operating system uses a suffix other than .so (such as windows or
212 Mac OS/X), the appropriate extension will be used.</p>
214 <p>If you are used CMake to build LLVM, see
217 <p>Now that we have the build scripts set up, we just need to write the code for
218 the pass itself.</p>
220 </div>
222 <!-- ======================================================================= -->
223 <h3>
225 </h3>
229 <p>Now that we have a way to compile our new pass, we just have to write it.
230 Start out with:</p>
235 <b>#include</b> "<a href="http://llvm.org/doxygen/raw__ostream_8h.html">llvm/Support/raw_ostream.h</a>"
236 </pre></div>
240 we are operating on <tt><a
242 and we will be doing some printing.</p>
247 </pre></div>
248 <p>... which is required because the functions from the include files
249 live in the llvm namespace.
250 </p>
256 </pre></div>
258 <p>... which starts out an anonymous namespace. Anonymous namespaces are to C++
259 what the "<tt>static</tt>" keyword is to C (at global scope). It makes the
260 things declared inside of the anonymous namespace only visible to the current
261 file. If you're not familiar with them, consult a decent C++ book for more
262 information.</p>
268 </pre></div><p>
272 The different builtin pass subclasses are described in detail <a
275 time.</p>
278 static char ID;
279 Hello() : FunctionPass(ID) {}
280 </pre></div><p>
282 <p> This declares pass identifier used by LLVM to identify pass. This allows LLVM to
283 avoid using expensive C++ runtime information.</p>
289 }
291 </pre></div>
294 which overloads an abstract virtual method inherited from <a
296 to do our thing, so we just print out our message with the name of each
297 function.</p>
300 char Hello::ID = 0;
301 </pre></div>
303 <p> We initialize pass ID here. LLVM uses ID's address to identify pass so
304 initialization value is not important.</p>
308 false /* Only looks at CFG */,
309 false /* Analysis Pass */);
311 </pre></div>
314 giving it a command line
316 Last two arguments describe its behavior.
317 If a pass walks CFG without modifying it then third argument is set to true.
318 If a pass is an analysis pass, for example dominator tree pass, then true
319 is supplied as fourth argument. </p>
326 <b>#include</b> "<a href="http://llvm.org/doxygen/raw__ostream_8h.html">llvm/Support/raw_ostream.h</a>"
333 static char ID;
334 Hello() : FunctionPass(ID) {}
339 }
340 };
342 char Hello::ID = 0;
344 }
346 </pre></div>
349 command in the local directory and you should get a new file
350 "<tt>Debug+Asserts/lib/Hello.so</tt>" under the top level directory of the LLVM
351 source tree (not in the local directory). Note that everything in this file is
352 contained in an anonymous namespace: this reflects the fact that passes are self
353 contained units that do not need external interfaces (although they can have
354 them) to be useful.</p>
356 </div>
358 <!-- ======================================================================= -->
359 <h3>
361 </h3>
365 <p>Now that you have a brand new shiny shared object file, we can use the
373 through our transformation like this (or course, any bitcode file will
374 work):</p>
378 Hello: __main
379 Hello: puts
380 Hello: main
381 </pre></div>
386 pass</a>). Because the hello pass does not modify the program in any
390 <p>To see what happened to the other string you registered, try running
394 $ opt -load ../../../Debug+Asserts/lib/Hello.so -help
395 OVERVIEW: llvm .bc -> .bc modular optimizer
399 OPTIONS:
400 Optimizations available:
401 ...
402 -funcresolve - Resolve Functions
403 -gcse - Global Common Subexpression Elimination
404 -globaldce - Dead Global Elimination
406 -indvars - Canonicalize Induction Variables
407 -inline - Function Integration/Inlining
408 -instcombine - Combine redundant instructions
409 ...
410 </pre></div>
412 <p>The pass name get added as the information string for your pass, giving some
414 would go ahead and make it do the cool transformations you want. Once you get
415 it all working and tested, it may become useful to find out how fast your pass
418 the execution time of your pass along with the other passes you queue up. For
419 example:</p>
422 $ opt -load ../../../Debug+Asserts/lib/Hello.so -hello -time-passes < hello.bc > /dev/null
423 Hello: __main
424 Hello: puts
425 Hello: main
426 ===============================================================================
427 ... Pass execution timing report ...
428 ===============================================================================
431 ---User Time--- --System Time-- --User+System-- ---Wall Time--- --- Pass Name ---
437 </pre></div>
439 <p>As you can see, our implementation above is pretty fast :). The additional
441 that the LLVM emitted by your pass is still valid and well formed LLVM, which
442 hasn't been broken somehow.</p>
444 <p>Now that you have seen the basics of the mechanics behind passes, we can talk
445 about some more details of how they work and how to use them.</p>
447 </div>
449 <!-- *********************************************************************** -->
450 <h2>
452 </h2>
453 <!-- *********************************************************************** -->
457 <p>One of the first things that you should do when designing a new pass is to
458 decide what class you should subclass for your pass. The <a
461 did not discuss why or when this should occur. Here we talk about the classes
462 available, from the most general to the most specific.</p>
465 specific</b> class possible, while still being able to meet the requirements
466 listed. This gives the LLVM Pass Infrastructure information necessary to
467 optimize how passes are run, so that the resultant compiler isn't unnecessarily
468 slow.</p>
470 </div>
472 <!-- ======================================================================= -->
473 <h3>
475 </h3>
481 class. This pass type is used for passes that do not have to be run, do not
482 change state, and never need to be updated. This is not a normal type of
483 transformation or analysis, but can provide information about the current
484 compiler configuration.</p>
486 <p>Although this pass class is very infrequently used, it is important for
487 providing information about the current target machine being compiled for, and
488 other static information that can affect the various transformations.</p>
493 </div>
495 <!-- ======================================================================= -->
496 <h3>
498 </h3>
504 class is the most general of all superclasses that you can use. Deriving from
506 referring to function bodies in no predictable order, or adding and removing
508 subclasses, no optimization can be done for their execution.</p>
510 <p>A module pass can use function level passes (e.g. dominators) using
511 the getAnalysis interface
513 function to retrieve analysis result for, if the function pass does not require
514 any module or immutable passes. Note that this can only be done for functions for which the
515 analysis ran, e.g. in the case of dominators you should only ask for the
516 DominatorTree for function definitions, not declarations.</p>
520 following signature:</p>
522 </div>
524 <!-- _______________________________________________________________________ -->
525 <h4>
527 </h4>
533 </pre></div>
536 It should return true if the module was modified by the transformation and
537 false otherwise.</p>
539 </div>
541 <!-- ======================================================================= -->
542 <h3>
544 </h3>
550 is used by passes that need to traverse the program bottom-up on the call graph
551 (callees before callers). Deriving from CallGraphSCCPass provides some
552 mechanics for building and traversing the CallGraph, but also allows the system
553 to optimize execution of CallGraphSCCPass's. If your pass meets the
554 requirements outlined below, and doesn't meet the requirements of a <tt><a
563 <ol>
566 than those in the current SCC and the direct callers and direct callees of the
567 SCC.</li>
570 to reflect any changes made to the program.</li>
573 though they may change the contents of an SCC.</li>
576 Module.</li>
580 </ol>
583 because it has to handle SCCs with more than one node in it. All of the virtual
584 methods described below should return true if they modified the program, or
585 false if they didn't.</p>
587 </div>
589 <!-- _______________________________________________________________________ -->
590 <h4>
593 </a>
594 </h4>
600 </pre></div>
605 is designed to do simple initialization type of stuff that does not depend on
607 scheduled to overlap with any other pass executions (thus it should be very
608 fast).</p>
610 </div>
612 <!-- _______________________________________________________________________ -->
613 <h4>
615 </h4>
621 </pre></div>
624 should return true if the module was modified by the transformation, false
625 otherwise.</p>
627 </div>
629 <!-- _______________________________________________________________________ -->
630 <h4>
633 </a>
634 </h4>
640 </pre></div>
643 called when the pass framework has finished calling <a
645 program being compiled.</p>
647 </div>
649 <!-- ======================================================================= -->
650 <h3>
652 </h3>
658 subclasses do have a predictable, local behavior that can be expected by the
660 independent of all of the other functions in the program.
666 <ol>
670 <li>Maintain state across invocations of
672 </ol>
676 may overload three virtual methods to do their work. All of these methods
677 should return true if they modified the program, or false if they didn't.</p>
679 </div>
681 <!-- _______________________________________________________________________ -->
682 <h4>
685 </a>
686 </h4>
692 </pre></div>
697 is designed to do simple initialization type of stuff that does not depend on
699 scheduled to overlap with any other pass executions (thus it should be very
700 fast).</p>
707 free functions that it needs, adding prototypes to the module if necessary.</p>
709 </div>
711 <!-- _______________________________________________________________________ -->
712 <h4>
714 </h4>
720 </pre></div><p>
723 the transformation or analysis work of your pass. As usual, a true value should
724 be returned if the function is modified.</p>
726 </div>
728 <!-- _______________________________________________________________________ -->
729 <h4>
732 </a>
733 </h4>
739 </pre></div>
742 called when the pass framework has finished calling <a
744 program being compiled.</p>
746 </div>
748 <!-- ======================================================================= -->
749 <h3>
751 </h3>
757 loop nest order such that outer most loop is processed last. </p>
762 do their work. All these methods should return true if they modified the
763 program, or false if they didn't. </p>
764 </div>
766 <!-- _______________________________________________________________________ -->
767 <h4>
770 </a>
771 </h4>
777 </pre></div>
780 type of stuff that does not depend on the functions being processed. The
782 other pass executions (thus it should be very fast). LPPassManager
783 interface should be used to access Function or Module level analysis
784 information.</p>
786 </div>
789 <!-- _______________________________________________________________________ -->
790 <h4>
792 </h4>
798 </pre></div><p>
801 the transformation or analysis work of your pass. As usual, a true value should
803 should be used to update loop nest.</p>
805 </div>
807 <!-- _______________________________________________________________________ -->
808 <h4>
810 </h4>
816 </pre></div>
819 called when the pass framework has finished calling <a
821 program being compiled. </p>
823 </div>
825 <!-- ======================================================================= -->
826 <h3>
828 </h3>
833 but executes on each single entry single exit region in the function.
835 region is processed last. </p>
840 methods should return true if they modified the program, or false if they didn not.
841 </p>
842 </div>
844 <!-- _______________________________________________________________________ -->
845 <h4>
848 </a>
849 </h4>
855 </pre></div>
858 type of stuff that does not depend on the functions being processed. The
860 other pass executions (thus it should be very fast). RPPassManager
861 interface should be used to access Function or Module level analysis
862 information.</p>
864 </div>
867 <!-- _______________________________________________________________________ -->
868 <h4>
870 </h4>
876 </pre></div><p>
879 the transformation or analysis work of your pass. As usual, a true value should
881 should be used to update region tree.</p>
883 </div>
885 <!-- _______________________________________________________________________ -->
886 <h4>
888 </h4>
894 </pre></div>
897 called when the pass framework has finished calling <a
899 program being compiled. </p>
901 </div>
905 <!-- ======================================================================= -->
906 <h3>
908 </h3>
914 their scope of inspection and modification to a single basic block at a time.
917 <ol>
919 <li>Maintain state across invocations of
922 <li>Any of the things forbidden for
924 </ol>
927 optimizations. They may override the same <a
930 href="#FunctionPass"><tt>FunctionPass</tt></a>'s have, but also have the following virtual methods that may also be implemented:</p>
932 </div>
934 <!-- _______________________________________________________________________ -->
935 <h4>
938 </a>
939 </h4>
945 </pre></div>
950 to do simple initialization that does not depend on the
952 scheduled to overlap with any other pass executions (thus it should be very
953 fast).</p>
955 </div>
957 <!-- _______________________________________________________________________ -->
958 <h4>
960 </h4>
966 </pre></div>
969 function is not allowed to inspect or modify basic blocks other than the
970 parameter, and are not allowed to modify the CFG. A true value must be returned
971 if the basic block is modified.</p>
973 </div>
975 <!-- _______________________________________________________________________ -->
976 <h4>
979 </a>
980 </h4>
986 </pre></div>
989 called when the pass framework has finished calling <a
991 program being compiled. This can be used to perform per-function
992 finalization.</p>
994 </div>
996 <!-- ======================================================================= -->
997 <h3>
999 </h3>
1004 executes on the machine-dependent representation of each LLVM function in the
1005 program.</p>
1007 <p>Code generator passes are registered and initialized specially by
1010 commands.</p>
1017 <ol>
1018 <li>Modify or create any LLVM IR Instructions, BasicBlocks, Arguments,
1019 Functions, GlobalVariables, GlobalAliases, or Modules.</li>
1023 data)</li>
1024 </ol>
1026 </div>
1028 <!-- _______________________________________________________________________ -->
1029 <h4>
1032 </a>
1033 </h4>
1039 </pre></div>
1054 </div>
1056 <!-- *********************************************************************** -->
1057 <h2>
1059 </h2>
1060 <!-- *********************************************************************** -->
1065 pass registration works, and discussed some of the reasons that it is used and
1066 what it does. Here we discuss how and why passes are registered.</p>
1069 template. The template parameter is the name of the pass that is to be used on
1070 the command line to specify that the pass should be added to a program (for
1073 programs, as
1076 <p>If you want your pass to be easily dumpable, you should
1079 </div>
1081 <!-- _______________________________________________________________________ -->
1082 <h4>
1084 </h4>
1090 </pre></div>
1093 a human readable version of the analysis results. This is useful for debugging
1094 an analysis itself, as well as for other people to figure out how an analysis
1099 program that has been analyzed. Note however that this pointer may be null in
1101 debugger), so it should only be used to enhance debug output, it should not be
1102 depended on.</p>
1104 </div>
1106 <!-- *********************************************************************** -->
1107 <h2>
1109 </h2>
1110 <!-- *********************************************************************** -->
1117 know how the passes interact with each other and what dependencies exist between
1118 the various passes. To track this, each pass can declare the set of passes that
1119 are required to be executed before the current pass, and the passes which are
1120 invalidated by the current pass.</p>
1122 <p>Typically this functionality is used to require that analysis results are
1123 computed before your pass is run. Running arbitrary transformation passes can
1124 invalidate the computed analysis results, which is what the invalidation set
1125 specifies. If a pass does not implement the <tt><a
1129 </div>
1131 <!-- _______________________________________________________________________ -->
1132 <h4>
1134 </h4>
1140 </pre></div>
1143 invalidated sets may be specified for your transformation. The implementation
1144 should fill in the <tt><a
1146 object with information about which passes are required and not invalidated. To
1147 do this, a pass may call any of the following methods on the AnalysisUsage
1148 object:</p>
1149 </div>
1151 <!-- _______________________________________________________________________ -->
1152 <h4>
1156 </a>
1157 </h4>
1160 <p>
1161 If your pass requires a previous pass to be executed (an analysis for example),
1162 it can use one of these methods to arrange for it to be run before your pass.
1163 LLVM has many different types of analyses and passes that can be required,
1166 be no critical edges in the CFG when your pass has been run.
1167 </p>
1169 <p>
1170 Some analyses chain to other analyses to do their job. For example, an <a
1175 that the transitively required pass should be alive as long as the requiring
1176 pass is.
1177 </p>
1178 </div>
1180 <!-- _______________________________________________________________________ -->
1181 <h4>
1184 </a>
1185 </h4>
1188 <p>
1189 One of the jobs of the PassManager is to optimize how and when analyses are run.
1190 In particular, it attempts to avoid recomputing data unless it needs to. For
1191 this reason, passes are allowed to declare that they preserve (i.e., they don't
1192 invalidate) an existing analysis if it's available. For example, a simple
1193 constant folding pass would not modify the CFG, so it can't possibly affect the
1194 results of dominator analysis. By default, all passes are assumed to invalidate
1195 all others.
1196 </p>
1198 <p>
1202 not modify the LLVM program at all (which is true for analyses), and the
1204 instructions in the program but do not modify the CFG or terminator instructions
1205 (note that this property is implicitly set for <a
1207 </p>
1209 <p>
1212 and dominator related analyses if they exist, so it can preserve them, despite
1213 the fact that it hacks on the CFG.
1214 </p>
1215 </div>
1217 <!-- _______________________________________________________________________ -->
1218 <h4>
1221 </a>
1222 </h4>
1228 <b>void</b> <a href="http://llvm.org/doxygen/structLICM.html">LICM</a>::getAnalysisUsage(AnalysisUsage &AU) <b>const</b> {
1229 AU.setPreservesCFG();
1230 AU.addRequired<<a href="http://llvm.org/doxygen/classllvm_1_1LoopInfo.html">LoopInfo</a>>();
1231 }
1232 </pre></div>
1234 </div>
1236 <!-- _______________________________________________________________________ -->
1237 <h4>
1241 </a>
1242 </h4>
1247 your class, providing you with access to the passes that you declared that you
1249 method. It takes a single template argument that specifies which pass class you
1250 want, and returns a reference to that pass. For example:</p>
1253 bool LICM::runOnFunction(Function &F) {
1255 ...
1256 }
1257 </pre></div>
1259 <p>This method call returns a reference to the pass desired. You may get a
1260 runtime assertion failure if you attempt to get an analysis that you did not
1261 declare as required in your <a
1266 A module level pass can use function level analysis info using this interface.
1267 For example:</p>
1270 bool ModuleLevelPass::runOnModule(Module &M) {
1271 ...
1273 ...
1274 }
1275 </pre></div>
1277 <p>In above example, runOnFunction for DominatorTree is called by pass manager
1278 before returning a reference to the desired pass.</p>
1280 <p>
1281 If your pass is capable of updating analyses if they exist (e.g.,
1284 if it is active. For example:</p>
1287 ...
1290 }
1291 ...
1292 </pre></div>
1294 </div>
1296 <!-- *********************************************************************** -->
1297 <h2>
1299 </h2>
1300 <!-- *********************************************************************** -->
1304 <p>Now that we understand the basics of how passes are defined, how they are
1305 used, and how they are required from other passes, it's time to get a little bit
1306 fancier. All of the pass relationships that we have seen so far are very
1307 simple: one pass depends on one other specific pass to be run before it can run.
1308 For many applications, this is great, for others, more flexibility is
1309 required.</p>
1311 <p>In particular, some analyses are defined such that there is a single simple
1312 interface to the analysis results, but multiple ways of calculating them.
1313 Consider alias analysis for example. The most trivial alias analysis returns
1314 "may alias" for any alias query. The most sophisticated analysis a
1315 flow-sensitive, context-sensitive interprocedural analysis that can take a
1316 significant amount of time to execute (and obviously, there is a lot of room
1317 between these two extremes for other implementations). To cleanly support
1318 situations like this, the LLVM Pass Infrastructure supports the notion of
1319 Analysis Groups.</p>
1321 </div>
1323 <!-- _______________________________________________________________________ -->
1324 <h4>
1326 </h4>
1330 <p>An Analysis Group is a single simple interface that may be implemented by
1331 multiple different passes. Analysis Groups can be given human readable names
1333 class. An analysis group may have one or more implementations, one of which is
1336 <p>Analysis groups are used by client passes just like other passes are: the
1339 scans the available passes to see if any implementations of the analysis group
1340 are available. If none is available, the default implementation is created for
1345 passes, all analysis group implementations must be registered, and must use the
1347 implementation pool. Also, a default implementation of the interface
1353 analysis group. The default implementation of the alias analysis interface (the
1354 <tt><a
1356 pass) just does a few simple checks that don't require significant analysis to
1357 compute (such as: two different globals can never alias each other, etc).
1358 Passes that use the <tt><a
1360 interface (for example the <tt><a
1362 not care which implementation of alias analysis is actually provided, they just
1363 use the designated interface.</p>
1365 <p>From the user's perspective, commands work just like normal. Issuing the
1367 instantiated and added to the pass sequence. Issuing the command '<tt>opt
1370 hypothetical example) instead.</p>
1372 </div>
1374 <!-- _______________________________________________________________________ -->
1375 <h4>
1377 </h4>
1383 implementations to the analysis group. First,
1384 an analysis group should be registered, with a human readable name
1385 provided for it.
1386 Unlike registration of passes, there is no command line argument to be specified
1390 <b>static</b> RegisterAnalysisGroup<<a href="http://llvm.org/doxygen/classllvm_1_1AliasAnalysis.html">AliasAnalysis</a>> A("<i>Alias Analysis</i>");
1391 </pre></div>
1393 <p>Once the analysis is registered, passes can declare that they are valid
1394 implementations of the interface by using the following code:</p>
1399 INITIALIZE_AG_PASS(FancyAA, <a href="http://llvm.org/doxygen/classllvm_1_1AliasAnalysis.html">AliasAnalysis</a>, "<i>somefancyaa</i>",
1400 "<i>A more complex alias analysis implementation</i>",
1404 );
1405 }
1406 </pre></div>
1410 to "join" the <tt><a href="http://llvm.org/doxygen/classllvm_1_1AliasAnalysis.html">AliasAnalysis</a></tt>
1411 analysis group. Every implementation of an analysis group should join using
1412 this macro.</p>
1417 INITIALIZE_AG_PASS(BasicAA, <a href="http://llvm.org/doxygen/classllvm_1_1AliasAnalysis.html">AliasAnalysis</a>, "<i>basicaa</i>",
1418 "<i>Basic Alias Analysis (default AA impl)</i>",
1422 );
1423 }
1424 </pre></div>
1426 <p>Here we show how the default implementation is specified (using the final
1428 one default implementation available at all times for an Analysis Group to be
1430 Here we declare that the
1431 <tt><a href="http://llvm.org/doxygen/structBasicAliasAnalysis.html">BasicAliasAnalysis</a></tt>
1432 pass is the default implementation for the interface.</p>
1434 </div>
1436 <!-- *********************************************************************** -->
1437 <h2>
1439 </h2>
1440 <!-- *********************************************************************** -->
1445 class is designed to be an easy way to expose various success
1446 metrics from passes. These statistics are printed at the end of a
1447 run, when the -stats command line option is enabled on the command
1448 line. See the <a href="http://llvm.org/docs/ProgrammersManual.html#Statistic">Statistics section</a> in the Programmer's Manual for details.
1450 </div>
1453 <!-- *********************************************************************** -->
1454 <h2>
1456 </h2>
1457 <!-- *********************************************************************** -->
1463 <a
1466 are set up correctly, and then schedules passes to run efficiently. All of the
1468 passes.</p>
1471 time of a series of passes:</p>
1473 <ol>
1475 recomputing analysis results as much as possible. This means keeping track of
1476 which analyses are available already, which analyses get invalidated, and which
1477 analyses are needed to be run for a pass. An important part of work is that the
1480 results as soon as they are no longer needed.</li>
1484 of a series of passes by pipelining the passes together. This means that, given
1487 the first function, then all of the <a
1489 etc... until the entire program has been run through the passes.
1491 <p>This improves the cache behavior of the compiler, because it is only touching
1492 the LLVM program representation for a single function at a time, instead of
1493 traversing the entire program. It reduces the memory consumption of compiler,
1494 because, for example, only one <a
1496 needs to be calculated at a time. This also makes it possible to implement
1497 some <a
1500 </ol>
1503 much information it has about the behaviors of the passes it is scheduling. For
1504 example, the "preserved" set is intentionally conservative in the face of an
1506 Not implementing when it should be implemented will have the effect of not
1507 allowing any analysis results to live across the execution of your pass.</p>
1510 options that is useful for debugging pass execution, seeing how things work, and
1511 diagnosing when you should be preserving more analyses than you currently are
1520 $ opt -load ../../../Debug+Asserts/lib/Hello.so -gcse -licm --debug-pass=Structure < hello.bc > /dev/null
1521 Module Pass Manager
1522 Function Pass Manager
1523 Dominator Set Construction
1524 Immediate Dominators Construction
1525 Global Common Subexpression Elimination
1526 -- Immediate Dominators Construction
1527 -- Global Common Subexpression Elimination
1528 Natural Loop Construction
1529 Loop Invariant Code Motion
1530 -- Natural Loop Construction
1531 -- Loop Invariant Code Motion
1532 Module Verifier
1533 -- Dominator Set Construction
1534 -- Module Verifier
1535 Bitcode Writer
1536 --Bitcode Writer
1537 </pre></div>
1539 <p>This output shows us when passes are constructed and when the analysis
1541 GCSE uses dominator and immediate dominator information to do its job. The LICM
1542 pass uses natural loop information, which uses dominator sets, but not immediate
1543 dominators. Because immediate dominators are no longer useful after the GCSE
1544 pass, it is immediately destroyed. The dominator sets are then reused to
1545 compute natural loop information, which is then used by the LICM pass.</p>
1547 <p>After the LICM pass, the module verifier runs (which is automatically added
1549 resultant LLVM code is well formed. After it finishes, the dominator set
1550 information is destroyed, after being computed once, and shared by three
1551 passes.</p>
1557 $ opt -load ../../../Debug+Asserts/lib/Hello.so -gcse -hello -licm --debug-pass=Structure < hello.bc > /dev/null
1558 Module Pass Manager
1559 Function Pass Manager
1560 Dominator Set Construction
1561 Immediate Dominators Construction
1562 Global Common Subexpression Elimination
1564 -- Immediate Dominators Construction
1565 -- Global Common Subexpression Elimination
1566 <b> Hello World Pass
1567 -- Hello World Pass
1568 Dominator Set Construction</b>
1569 Natural Loop Construction
1570 Loop Invariant Code Motion
1571 -- Natural Loop Construction
1572 -- Loop Invariant Code Motion
1573 Module Verifier
1574 -- Dominator Set Construction
1575 -- Module Verifier
1576 Bitcode Writer
1577 --Bitcode Writer
1578 Hello: __main
1579 Hello: puts
1580 Hello: main
1581 </pre></div>
1584 Dominator Set pass, even though it doesn't modify the code at all! To fix this,
1585 we need to add the following <a
1591 AU.setPreservesAll();
1592 }
1593 </pre></div>
1598 $ opt -load ../../../Debug+Asserts/lib/Hello.so -gcse -hello -licm --debug-pass=Structure < hello.bc > /dev/null
1599 Pass Arguments: -gcse -hello -licm
1600 Module Pass Manager
1601 Function Pass Manager
1602 Dominator Set Construction
1603 Immediate Dominators Construction
1604 Global Common Subexpression Elimination
1605 -- Immediate Dominators Construction
1606 -- Global Common Subexpression Elimination
1607 Hello World Pass
1608 -- Hello World Pass
1609 Natural Loop Construction
1610 Loop Invariant Code Motion
1611 -- Loop Invariant Code Motion
1612 -- Natural Loop Construction
1613 Module Verifier
1614 -- Dominator Set Construction
1615 -- Module Verifier
1616 Bitcode Writer
1617 --Bitcode Writer
1618 Hello: __main
1619 Hello: puts
1620 Hello: main
1621 </pre></div>
1623 <p>Which shows that we don't accidentally invalidate dominator information
1624 anymore, and therefore do not have to compute it twice.</p>
1626 </div>
1628 <!-- _______________________________________________________________________ -->
1629 <h4>
1631 </h4>
1637 </pre></div>
1640 results, and how long to keep them around for. Because the lifetime of the pass
1641 object itself is effectively the entire duration of the compilation process, we
1642 need some way to free analysis results when they are no longer useful. The
1645 <p>If you are writing an analysis or any other pass that retains a significant
1646 amount of state (for use by another pass which "requires" your pass and uses the
1652 </div>
1654 <!-- *********************************************************************** -->
1655 <h2>
1657 </h2>
1658 <!-- *********************************************************************** -->
1663 both for the purposes of distribution, and for regulating the resident code size
1664 when running on the target system. Therefore, it becomes desirable to
1665 selectively use some passes, while omitting others and maintain the flexibility
1666 to change configurations later on. You want to be able to do all this, and,
1667 provide feedback to the user. This is where pass registration comes into
1668 play.</p>
1670 <p>The fundamental mechanisms for pass registration are the
1676 communicates additions and deletions to the command line interface.</p>
1679 information provided about a particular pass. This information includes the
1680 command line name, the command help string and the address of the function used
1681 to create an instance of the pass. A global static constructor of one of these
1684 in the tool will be registered at start up. A dynamically loaded pass will
1685 register on load and unregister at unload.</p>
1687 </div>
1689 <!-- _______________________________________________________________________ -->
1690 <h3>
1692 </h3>
1696 <p>There are predefined registries to track instruction scheduling
1699 allocator machine pass.</p>
1701 <p>Implement your register allocator machine pass. In your register allocator
1702 .cpp file add the following include;</p>
1705 #include "llvm/CodeGen/RegAllocRegistry.h"
1706 </pre></div>
1708 <p>Also in your register allocator .cpp file, define a creator function in the
1709 form; </p>
1712 FunctionPass *createMyRegisterAllocator() {
1713 return new MyRegisterAllocator();
1714 }
1715 </pre></div>
1717 <p>Note that the signature of this function should match the type of
1722 static RegisterRegAlloc myRegAlloc("myregalloc",
1723 " my register allocator help string",
1724 createMyRegisterAllocator);
1725 </pre></div>
1727 <p>Note the two spaces prior to the help string produces a tidy result on the
1728 -help query.</p>
1731 $ llc -help
1732 ...
1733 -regalloc - Register allocator to use (default=linearscan)
1734 =linearscan - linear scan register allocator
1735 =local - local register allocator
1736 =simple - simple register allocator
1737 =myregalloc - my register allocator help string
1738 ...
1739 </pre></div>
1742 an option. Registering instruction schedulers is similar except use the
1747 <p>To force the load/linking of your register allocator into the llc/lli tools,
1751 </div>
1754 <!-- _______________________________________________________________________ -->
1755 <h3>
1757 </h3>
1761 <p>The easiest way to get started is to clone one of the existing registries; we
1765 <p>Then you need to declare the registry. Example: if your pass registry is
1769 MachinePassRegistry RegisterMyPasses::Registry;
1770 </pre></div>
1775 cl::opt<RegisterMyPasses::FunctionPassCtor, false,
1777 MyPassOpt("mypass",
1778 cl::init(&createDefaultMyPass),
1779 cl::desc("my pass option help"));
1780 </pre></div>
1783 creator.</p>
1785 </div>
1787 <!-- *********************************************************************** -->
1788 <h2>
1790 </h2>
1791 <!-- *********************************************************************** -->
1795 <p>Unfortunately, using GDB with dynamically loaded passes is not as easy as it
1796 should be. First of all, you can't set a breakpoint in a shared object that has
1797 not been loaded yet, and second of all there are problems with inlined functions
1798 in shared objects. Here are some suggestions to debugging your pass with
1799 GDB.</p>
1801 <p>For sake of discussion, I'm going to assume that you are debugging a
1803 on that.</p>
1805 </div>
1807 <!-- _______________________________________________________________________ -->
1808 <h4>
1810 </h4>
1818 GNU gdb 5.0
1819 Copyright 2000 Free Software Foundation, Inc.
1820 GDB is free software, covered by the GNU General Public License, and you are
1821 welcome to change it and/or distribute copies of it under certain conditions.
1822 Type "show copying" to see the conditions.
1823 There is absolutely no warranty for GDB. Type "show warranty" for details.
1824 This GDB was configured as "sparc-sun-solaris2.6"...
1825 (gdb)
1826 </pre></div>
1829 time to load. Be patient. Since we cannot set a breakpoint in our pass yet
1830 (the shared object isn't loaded until runtime), we must execute the process, and
1831 have it stop before it invokes our pass, but after it has loaded the shared
1832 object. The most foolproof way of doing this is to set a breakpoint in
1834 want:</p>
1840 Starting program: opt test.bc -load $(LLVMTOP)/llvm/Debug+Asserts/lib/[libname].so -[passoption]
1843 (gdb)
1844 </pre></div>
1847 now free to set breakpoints in your pass so that you can trace through execution
1848 or do other standard debugging stuff.</p>
1850 </div>
1852 <!-- _______________________________________________________________________ -->
1853 <h4>
1855 </h4>
1859 <p>Once you have the basics down, there are a couple of problems that GDB has,
1860 some with solutions, some without.</p>
1862 <ul>
1863 <li>Inline functions have bogus stack information. In general, GDB does a
1864 pretty good job getting stack traces and stepping through inline functions.
1865 When a pass is dynamically loaded however, it somehow completely loses this
1866 capability. The only solution I know of is to de-inline a function (move it
1867 from the body of a class to a .cpp file).</li>
1869 <li>Restarting the program breaks breakpoints. After following the information
1870 above, you have succeeded in getting some breakpoints planted in your pass. Nex
1872 and you start getting errors about breakpoints being unsettable. The only way I
1874 already set in your pass, run the program, and re-set the breakpoints once
1877 </ul>
1879 <p>Hopefully these tips will help with common case debugging situations. If
1880 you'd like to contribute some tips of your own, just contact <a
1883 </div>
1885 <!-- *********************************************************************** -->
1886 <h2>
1888 </h2>
1889 <!-- *********************************************************************** -->
1893 <p>Although the LLVM Pass Infrastructure is very capable as it stands, and does
1894 some nifty stuff, there are things we'd like to add in the future. Here is
1895 where we are going:</p>
1897 </div>
1899 <!-- _______________________________________________________________________ -->
1900 <h4>
1902 </h4>
1906 <p>Multiple CPU machines are becoming more common and compilation can never be
1907 fast enough: obviously we should allow for a multithreaded compiler. Because of
1908 the semantics defined for passes above (specifically they cannot maintain state
1911 to create multiple instances of each pass object, and allow the separate
1912 instances to be hacking on different parts of the program at the same time.</p>
1914 <p>This implementation would prevent each of the passes from having to implement
1915 multithreaded constructs, requiring only the LLVM core to have locking in a few
1916 places (for global resources). Although this is a simple extension, we simply
1917 haven't had time (or multiprocessor machines, thus a reason) to implement this.
1918 Despite that, we have kept the LLVM passes SMP ready, and you should too.</p>
1920 </div>
1922 <!-- *********************************************************************** -->
1923 <hr>
1924 <address>
1932 Last modified: $Date$
1933 </address>
1935 </body>
1936 </html>