1 ========================
2 LLVM Programmer's Manual
3 ========================
9 This is always a work in progress.
16 This document is meant to highlight some of the important classes and interfaces
17 available in the LLVM source-base. This manual is not intended to explain what
18 LLVM is, how it works, and what LLVM code looks like. It assumes that you know
19 the basics of LLVM and are interested in writing transformations or otherwise
20 analyzing or manipulating the code.
22 This document should get you oriented so that you can find your way in the
23 continuously growing source code that makes up the LLVM infrastructure. Note
24 that this manual is not intended to serve as a replacement for reading the
25 source code, so if you think there should be a method in one of these classes to
26 do something, but it's not listed, check the source. Links to the `doxygen
27 <http://llvm.org/doxygen/>`__ sources are provided to make this as easy as
30 The first section of this document describes general information that is useful
31 to know when working in the LLVM infrastructure, and the second describes the
32 Core LLVM classes. In the future this manual will be extended with information
33 describing how to use extension libraries, such as dominator information, CFG
34 traversal routines, and useful utilities like the ``InstVisitor`` (`doxygen
35 <http://llvm.org/doxygen/InstVisitor_8h-source.html>`__) template.
42 This section contains general information that is useful if you are working in
43 the LLVM source-base, but that isn't specific to any particular API.
47 The C++ Standard Template Library
48 ---------------------------------
50 LLVM makes heavy use of the C++ Standard Template Library (STL), perhaps much
51 more than you are used to, or have seen before. Because of this, you might want
52 to do a little background reading in the techniques used and capabilities of the
53 library. There are many good pages that discuss the STL, and several books on
54 the subject that you can get, so it will not be discussed in this document.
56 Here are some useful links:
59 <http://en.cppreference.com/w/>`_ - an excellent
60 reference for the STL and other parts of the standard C++ library.
62 #. `C++ In a Nutshell <http://www.tempest-sw.com/cpp/>`_ - This is an O'Reilly
63 book in the making. It has a decent Standard Library Reference that rivals
64 Dinkumware's, and is unfortunately no longer free since the book has been
67 #. `C++ Frequently Asked Questions <http://www.parashift.com/c++-faq-lite/>`_.
69 #. `SGI's STL Programmer's Guide <http://www.sgi.com/tech/stl/>`_ - Contains a
70 useful `Introduction to the STL
71 <http://www.sgi.com/tech/stl/stl_introduction.html>`_.
73 #. `Bjarne Stroustrup's C++ Page
74 <http://www.research.att.com/%7Ebs/C++.html>`_.
76 #. `Bruce Eckel's Thinking in C++, 2nd ed. Volume 2 Revision 4.0
77 (even better, get the book)
78 <http://www.mindview.net/Books/TICPP/ThinkingInCPP2e.html>`_.
80 You are also encouraged to take a look at the :doc:`LLVM Coding Standards
81 <CodingStandards>` guide which focuses on how to write maintainable code more
82 than where to put your curly braces.
86 Other useful references
87 -----------------------
89 #. `Using static and shared libraries across platforms
90 <http://www.fortran-2000.com/ArnaudRecipes/sharedlib.html>`_
94 Important and useful LLVM APIs
95 ==============================
97 Here we highlight some LLVM APIs that are generally useful and good to know
98 about when writing transformations.
102 The ``isa<>``, ``cast<>`` and ``dyn_cast<>`` templates
103 ------------------------------------------------------
105 The LLVM source-base makes extensive use of a custom form of RTTI. These
106 templates have many similarities to the C++ ``dynamic_cast<>`` operator, but
107 they don't have some drawbacks (primarily stemming from the fact that
108 ``dynamic_cast<>`` only works on classes that have a v-table). Because they are
109 used so often, you must know what they do and how they work. All of these
110 templates are defined in the ``llvm/Support/Casting.h`` (`doxygen
111 <http://llvm.org/doxygen/Casting_8h-source.html>`__) file (note that you very
112 rarely have to include this file directly).
115 The ``isa<>`` operator works exactly like the Java "``instanceof``" operator.
116 It returns true or false depending on whether a reference or pointer points to
117 an instance of the specified class. This can be very useful for constraint
118 checking of various sorts (example below).
121 The ``cast<>`` operator is a "checked cast" operation. It converts a pointer
122 or reference from a base class to a derived class, causing an assertion
123 failure if it is not really an instance of the right type. This should be
124 used in cases where you have some information that makes you believe that
125 something is of the right type. An example of the ``isa<>`` and ``cast<>``
130 static bool isLoopInvariant(const Value *V, const Loop *L) {
131 if (isa<Constant>(V) || isa<Argument>(V) || isa<GlobalValue>(V))
134 // Otherwise, it must be an instruction...
135 return !L->contains(cast<Instruction>(V)->getParent());
138 Note that you should **not** use an ``isa<>`` test followed by a ``cast<>``,
139 for that use the ``dyn_cast<>`` operator.
142 The ``dyn_cast<>`` operator is a "checking cast" operation. It checks to see
143 if the operand is of the specified type, and if so, returns a pointer to it
144 (this operator does not work with references). If the operand is not of the
145 correct type, a null pointer is returned. Thus, this works very much like
146 the ``dynamic_cast<>`` operator in C++, and should be used in the same
147 circumstances. Typically, the ``dyn_cast<>`` operator is used in an ``if``
148 statement or some other flow control statement like this:
152 if (auto *AI = dyn_cast<AllocationInst>(Val)) {
156 This form of the ``if`` statement effectively combines together a call to
157 ``isa<>`` and a call to ``cast<>`` into one statement, which is very
160 Note that the ``dyn_cast<>`` operator, like C++'s ``dynamic_cast<>`` or Java's
161 ``instanceof`` operator, can be abused. In particular, you should not use big
162 chained ``if/then/else`` blocks to check for lots of different variants of
163 classes. If you find yourself wanting to do this, it is much cleaner and more
164 efficient to use the ``InstVisitor`` class to dispatch over the instruction
168 The ``cast_or_null<>`` operator works just like the ``cast<>`` operator,
169 except that it allows for a null pointer as an argument (which it then
170 propagates). This can sometimes be useful, allowing you to combine several
171 null checks into one.
173 ``dyn_cast_or_null<>``:
174 The ``dyn_cast_or_null<>`` operator works just like the ``dyn_cast<>``
175 operator, except that it allows for a null pointer as an argument (which it
176 then propagates). This can sometimes be useful, allowing you to combine
177 several null checks into one.
179 These five templates can be used with any classes, whether they have a v-table
180 or not. If you want to add support for these templates, see the document
181 :doc:`How to set up LLVM-style RTTI for your class hierarchy
182 <HowToSetUpLLVMStyleRTTI>`
186 Passing strings (the ``StringRef`` and ``Twine`` classes)
187 ---------------------------------------------------------
189 Although LLVM generally does not do much string manipulation, we do have several
190 important APIs which take strings. Two important examples are the Value class
191 -- which has names for instructions, functions, etc. -- and the ``StringMap``
192 class which is used extensively in LLVM and Clang.
194 These are generic classes, and they need to be able to accept strings which may
195 have embedded null characters. Therefore, they cannot simply take a ``const
196 char *``, and taking a ``const std::string&`` requires clients to perform a heap
197 allocation which is usually unnecessary. Instead, many LLVM APIs use a
198 ``StringRef`` or a ``const Twine&`` for passing strings efficiently.
202 The ``StringRef`` class
203 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
205 The ``StringRef`` data type represents a reference to a constant string (a
206 character array and a length) and supports the common operations available on
207 ``std::string``, but does not require heap allocation.
209 It can be implicitly constructed using a C style null-terminated string, an
210 ``std::string``, or explicitly with a character pointer and length. For
211 example, the ``StringRef`` find function is declared as:
215 iterator find(StringRef Key);
217 and clients can call it using any one of:
221 Map.find("foo"); // Lookup "foo"
222 Map.find(std::string("bar")); // Lookup "bar"
223 Map.find(StringRef("\0baz", 4)); // Lookup "\0baz"
225 Similarly, APIs which need to return a string may return a ``StringRef``
226 instance, which can be used directly or converted to an ``std::string`` using
227 the ``str`` member function. See ``llvm/ADT/StringRef.h`` (`doxygen
228 <http://llvm.org/doxygen/classllvm_1_1StringRef_8h-source.html>`__) for more
231 You should rarely use the ``StringRef`` class directly, because it contains
232 pointers to external memory it is not generally safe to store an instance of the
233 class (unless you know that the external storage will not be freed).
234 ``StringRef`` is small and pervasive enough in LLVM that it should always be
240 The ``Twine`` (`doxygen <http://llvm.org/doxygen/classllvm_1_1Twine.html>`__)
241 class is an efficient way for APIs to accept concatenated strings. For example,
242 a common LLVM paradigm is to name one instruction based on the name of another
243 instruction with a suffix, for example:
247 New = CmpInst::Create(..., SO->getName() + ".cmp");
249 The ``Twine`` class is effectively a lightweight `rope
250 <http://en.wikipedia.org/wiki/Rope_(computer_science)>`_ which points to
251 temporary (stack allocated) objects. Twines can be implicitly constructed as
252 the result of the plus operator applied to strings (i.e., a C strings, an
253 ``std::string``, or a ``StringRef``). The twine delays the actual concatenation
254 of strings until it is actually required, at which point it can be efficiently
255 rendered directly into a character array. This avoids unnecessary heap
256 allocation involved in constructing the temporary results of string
257 concatenation. See ``llvm/ADT/Twine.h`` (`doxygen
258 <http://llvm.org/doxygen/Twine_8h_source.html>`__) and :ref:`here <dss_twine>`
259 for more information.
261 As with a ``StringRef``, ``Twine`` objects point to external memory and should
262 almost never be stored or mentioned directly. They are intended solely for use
263 when defining a function which should be able to efficiently accept concatenated
271 Proper error handling helps us identify bugs in our code, and helps end-users
272 understand errors in their tool usage. Errors fall into two broad categories:
273 *programmatic* and *recoverable*, with different strategies for handling and
279 Programmatic errors are violations of program invariants or API contracts, and
280 represent bugs within the program itself. Our aim is to document invariants, and
281 to abort quickly at the point of failure (providing some basic diagnostic) when
282 invariants are broken at runtime.
284 The fundamental tools for handling programmatic errors are assertions and the
285 llvm_unreachable function. Assertions are used to express invariant conditions,
286 and should include a message describing the invariant:
290 assert(isPhysReg(R) && "All virt regs should have been allocated already.");
292 The llvm_unreachable function can be used to document areas of control flow
293 that should never be entered if the program invariants hold:
297 enum { Foo, Bar, Baz } X = foo();
300 case Foo: /* Handle Foo */; break;
301 case Bar: /* Handle Bar */; break;
303 llvm_unreachable("X should be Foo or Bar here");
309 Recoverable errors represent an error in the program's environment, for example
310 a resource failure (a missing file, a dropped network connection, etc.), or
311 malformed input. These errors should be detected and communicated to a level of
312 the program where they can be handled appropriately. Handling the error may be
313 as simple as reporting the issue to the user, or it may involve attempts at
316 Recoverable errors are modeled using LLVM's ``Error`` scheme. This scheme
317 represents errors using function return values, similar to classic C integer
318 error codes, or C++'s ``std::error_code``. However, the ``Error`` class is
319 actually a lightweight wrapper for user-defined error types, allowing arbitrary
320 information to be attached to describe the error. This is similar to the way C++
321 exceptions allow throwing of user-defined types.
323 Success values are created by calling ``Error::success()``, E.g.:
330 return Error::success();
333 Success values are very cheap to construct and return - they have minimal
334 impact on program performance.
336 Failure values are constructed using ``make_error<T>``, where ``T`` is any class
337 that inherits from the ErrorInfo utility, E.g.:
341 class BadFileFormat : public ErrorInfo<BadFileFormat> {
346 BadFileFormat(StringRef Path) : Path(Path.str()) {}
348 void log(raw_ostream &OS) const override {
349 OS << Path << " is malformed";
352 std::error_code convertToErrorCode() const override {
353 return make_error_code(object_error::parse_failed);
357 char FileExists::ID; // This should be declared in the C++ file.
359 Error printFormattedFile(StringRef Path) {
360 if (<check for valid format>)
361 return make_error<InvalidObjectFile>(Path);
362 // print file contents.
363 return Error::success();
366 Error values can be implicitly converted to bool: true for error, false for
367 success, enabling the following idiom:
374 if (auto Err = mayFail())
376 // Success! We can proceed.
379 For functions that can fail but need to return a value the ``Expected<T>``
380 utility can be used. Values of this type can be constructed with either a
381 ``T``, or an ``Error``. Expected<T> values are also implicitly convertible to
382 boolean, but with the opposite convention to ``Error``: true for success, false
383 for error. If success, the ``T`` value can be accessed via the dereference
384 operator. If failure, the ``Error`` value can be extracted using the
385 ``takeError()`` method. Idiomatic usage looks like:
389 Expected<FormattedFile> openFormattedFile(StringRef Path) {
390 // If badly formatted, return an error.
391 if (auto Err = checkFormat(Path))
392 return std::move(Err);
393 // Otherwise return a FormattedFile instance.
394 return FormattedFile(Path);
397 Error processFormattedFile(StringRef Path) {
398 // Try to open a formatted file
399 if (auto FileOrErr = openFormattedFile(Path)) {
400 // On success, grab a reference to the file and continue.
401 auto &File = *FileOrErr;
404 // On error, extract the Error value and return it.
405 return FileOrErr.takeError();
408 If an ``Expected<T>`` value is in success mode then the ``takeError()`` method
409 will return a success value. Using this fact, the above function can be
414 Error processFormattedFile(StringRef Path) {
415 // Try to open a formatted file
416 auto FileOrErr = openFormattedFile(Path);
417 if (auto Err = FileOrErr.takeError())
418 // On error, extract the Error value and return it.
420 // On success, grab a reference to the file and continue.
421 auto &File = *FileOrErr;
425 This second form is often more readable for functions that involve multiple
426 ``Expected<T>`` values as it limits the indentation required.
428 All ``Error`` instances, whether success or failure, must be either checked or
429 moved from (via ``std::move`` or a return) before they are destructed.
430 Accidentally discarding an unchecked error will cause a program abort at the
431 point where the unchecked value's destructor is run, making it easy to identify
432 and fix violations of this rule.
434 Success values are considered checked once they have been tested (by invoking
435 the boolean conversion operator):
439 if (auto Err = canFail(...))
440 return Err; // Failure value - move error to caller.
442 // Safe to continue: Err was checked.
444 In contrast, the following code will always cause an abort, even if ``canFail``
445 returns a success value:
450 // Program will always abort here, even if canFail() returns Success, since
451 // the value is not checked.
453 Failure values are considered checked once a handler for the error type has
459 processFormattedFile(...),
460 [](const BadFileFormat &BFF) {
461 report("Unable to process " + BFF.Path + ": bad format");
463 [](const FileNotFound &FNF) {
464 report("File not found " + FNF.Path);
467 The ``handleErrors`` function takes an error as its first argument, followed by
468 a variadic list of "handlers", each of which must be a callable type (a
469 function, lambda, or class with a call operator) with one argument. The
470 ``handleErrors`` function will visit each handler in the sequence and check its
471 argument type against the dynamic type of the error, running the first handler
472 that matches. This is the same decision process that is used decide which catch
473 clause to run for a C++ exception.
475 Since the list of handlers passed to ``handleErrors`` may not cover every error
476 type that can occur, the ``handleErrors`` function also returns an Error value
477 that must be checked or propagated. If the error value that is passed to
478 ``handleErrors`` does not match any of the handlers it will be returned from
479 handleErrors. Idiomatic use of ``handleErrors`` thus looks like:
485 processFormattedFile(...),
486 [](const BadFileFormat &BFF) {
487 report("Unable to process " + BFF.Path + ": bad format");
489 [](const FileNotFound &FNF) {
490 report("File not found " + FNF.Path);
494 In cases where you truly know that the handler list is exhaustive the
495 ``handleAllErrors`` function can be used instead. This is identical to
496 ``handleErrors`` except that it will terminate the program if an unhandled
497 error is passed in, and can therefore return void. The ``handleAllErrors``
498 function should generally be avoided: the introduction of a new error type
499 elsewhere in the program can easily turn a formerly exhaustive list of errors
500 into a non-exhaustive list, risking unexpected program termination. Where
501 possible, use handleErrors and propagate unknown errors up the stack instead.
503 For tool code, where errors can be handled by printing an error message then
504 exiting with an error code, the :ref:`ExitOnError <err_exitonerr>` utility
505 may be a better choice than handleErrors, as it simplifies control flow when
506 calling fallible functions.
511 Many kinds of errors have no recovery strategy, the only action that can be
512 taken is to report them to the user so that the user can attempt to fix the
513 environment. In this case representing the error as a string makes perfect
514 sense. LLVM provides the ``StringError`` class for this purpose. It takes two
515 arguments: A string error message, and an equivalent ``std::error_code`` for
520 make_error<StringError>("Bad executable",
521 make_error_code(errc::executable_format_error"));
523 If you're certain that the error you're building will never need to be converted
524 to a ``std::error_code`` you can use the ``inconvertibleErrorCode()`` function:
528 make_error<StringError>("Bad executable", inconvertibleErrorCode());
530 This should be done only after careful consideration. If any attempt is made to
531 convert this error to a ``std::error_code`` it will trigger immediate program
532 termination. Unless you are certain that your errors will not need
533 interoperability you should look for an existing ``std::error_code`` that you
534 can convert to, and even (as painful as it is) consider introducing a new one as
537 Interoperability with std::error_code and ErrorOr
538 """""""""""""""""""""""""""""""""""""""""""""""""
540 Many existing LLVM APIs use ``std::error_code`` and its partner ``ErrorOr<T>``
541 (which plays the same role as ``Expected<T>``, but wraps a ``std::error_code``
542 rather than an ``Error``). The infectious nature of error types means that an
543 attempt to change one of these functions to return ``Error`` or ``Expected<T>``
544 instead often results in an avalanche of changes to callers, callers of callers,
545 and so on. (The first such attempt, returning an ``Error`` from
546 MachOObjectFile's constructor, was abandoned after the diff reached 3000 lines,
547 impacted half a dozen libraries, and was still growing).
549 To solve this problem, the ``Error``/``std::error_code`` interoperability requirement was
550 introduced. Two pairs of functions allow any ``Error`` value to be converted to a
551 ``std::error_code``, any ``Expected<T>`` to be converted to an ``ErrorOr<T>``, and vice
556 std::error_code errorToErrorCode(Error Err);
557 Error errorCodeToError(std::error_code EC);
559 template <typename T> ErrorOr<T> expectedToErrorOr(Expected<T> TOrErr);
560 template <typename T> Expected<T> errorOrToExpected(ErrorOr<T> TOrEC);
563 Using these APIs it is easy to make surgical patches that update individual
564 functions from ``std::error_code`` to ``Error``, and from ``ErrorOr<T>`` to
567 Returning Errors from error handlers
568 """"""""""""""""""""""""""""""""""""
570 Error recovery attempts may themselves fail. For that reason, ``handleErrors``
571 actually recognises three different forms of handler signature:
575 // Error must be handled, no new errors produced:
576 void(UserDefinedError &E);
578 // Error must be handled, new errors can be produced:
579 Error(UserDefinedError &E);
581 // Original error can be inspected, then re-wrapped and returned (or a new
582 // error can be produced):
583 Error(std::unique_ptr<UserDefinedError> E);
585 Any error returned from a handler will be returned from the ``handleErrors``
586 function so that it can be handled itself, or propagated up the stack.
590 Using ExitOnError to simplify tool code
591 """""""""""""""""""""""""""""""""""""""
593 Library code should never call ``exit`` for a recoverable error, however in tool
594 code (especially command line tools) this can be a reasonable approach. Calling
595 ``exit`` upon encountering an error dramatically simplifies control flow as the
596 error no longer needs to be propagated up the stack. This allows code to be
597 written in straight-line style, as long as each fallible call is wrapped in a
598 check and call to exit. The ``ExitOnError`` class supports this pattern by
599 providing call operators that inspect ``Error`` values, stripping the error away
600 in the success case and logging to ``stderr`` then exiting in the failure case.
602 To use this class, declare a global ``ExitOnError`` variable in your program:
606 ExitOnError ExitOnErr;
608 Calls to fallible functions can then be wrapped with a call to ``ExitOnErr``,
609 turning them into non-failing calls:
614 Expected<int> mayFail2();
617 ExitOnErr(mayFail());
618 int X = ExitOnErr(mayFail2());
621 On failure, the error's log message will be written to ``stderr``, optionally
622 preceded by a string "banner" that can be set by calling the setBanner method. A
623 mapping can also be supplied from ``Error`` values to exit codes using the
624 ``setExitCodeMapper`` method:
628 int main(int argc, char *argv[]) {
629 ExitOnErr.setBanner(std::string(argv[0]) + " error:");
630 ExitOnErr.setExitCodeMapper(
631 [](const Error &Err) {
632 if (Err.isA<BadFileFormat>())
637 Use ``ExitOnError`` in your tool code where possible as it can greatly improve
640 Fallible constructors
641 """""""""""""""""""""
643 Some classes require resource acquisition or other complex initialization that
644 can fail during construction. Unfortunately constructors can't return errors,
645 and having clients test objects after they're constructed to ensure that they're
646 valid is error prone as it's all too easy to forget the test. To work around
647 this, use the named constructor idiom and return an ``Expected<T>``:
654 static Expected<Foo> Create(Resource R1, Resource R2) {
658 return std::move(Err);
664 Foo(Resource R1, Resource R2, Error &Err) {
665 ErrorAsOutParameter EAO(&Err);
666 if (auto Err2 = R1.acquire()) {
667 Err = std::move(Err2);
675 Here, the named constructor passes an ``Error`` by reference into the actual
676 constructor, which the constructor can then use to return errors. The
677 ``ErrorAsOutParameter`` utility sets the ``Error`` value's checked flag on entry
678 to the constructor so that the error can be assigned to, then resets it on exit
679 to force the client (the named constructor) to check the error.
681 By using this idiom, clients attempting to construct a Foo receive either a
682 well-formed Foo or an Error, never an object in an invalid state.
684 Propagating and consuming errors based on types
685 """""""""""""""""""""""""""""""""""""""""""""""
687 In some contexts, certain types of error are known to be benign. For example,
688 when walking an archive, some clients may be happy to skip over badly formatted
689 object files rather than terminating the walk immediately. Skipping badly
690 formatted objects could be achieved using an elaborate handler method, but the
691 Error.h header provides two utilities that make this idiom much cleaner: the
692 type inspection method, ``isA``, and the ``consumeError`` function:
696 Error walkArchive(Archive A) {
697 for (unsigned I = 0; I != A.numMembers(); ++I) {
698 auto ChildOrErr = A.getMember(I);
699 if (auto Err = ChildOrErr.takeError()) {
700 if (Err.isA<BadFileFormat>())
701 consumeError(std::move(Err))
705 auto &Child = *ChildOrErr;
709 return Error::success();
712 Concatenating Errors with joinErrors
713 """"""""""""""""""""""""""""""""""""
715 In the archive walking example above ``BadFileFormat`` errors are simply
716 consumed and ignored. If the client had wanted report these errors after
717 completing the walk over the archive they could use the ``joinErrors`` utility:
721 Error walkArchive(Archive A) {
722 Error DeferredErrs = Error::success();
723 for (unsigned I = 0; I != A.numMembers(); ++I) {
724 auto ChildOrErr = A.getMember(I);
725 if (auto Err = ChildOrErr.takeError())
726 if (Err.isA<BadFileFormat>())
727 DeferredErrs = joinErrors(std::move(DeferredErrs), std::move(Err));
730 auto &Child = *ChildOrErr;
737 The ``joinErrors`` routine builds a special error type called ``ErrorList``,
738 which holds a list of user defined errors. The ``handleErrors`` routine
739 recognizes this type and will attempt to handle each of the contained erorrs in
740 order. If all contained errors can be handled, ``handleErrors`` will return
741 ``Error::success()``, otherwise ``handleErrors`` will concatenate the remaining
742 errors and return the resulting ``ErrorList``.
744 Building fallible iterators and iterator ranges
745 """""""""""""""""""""""""""""""""""""""""""""""
747 The archive walking examples above retrieve archive members by index, however
748 this requires considerable boiler-plate for iteration and error checking. We can
749 clean this up by using ``Error`` with the "fallible iterator" pattern. The usual
750 C++ iterator patterns do not allow for failure on increment, but we can
751 incorporate support for it by having iterators hold an Error reference through
752 which they can report failure. In this pattern, if an increment operation fails
753 the failure is recorded via the Error reference and the iterator value is set to
754 the end of the range in order to terminate the loop. This ensures that the
755 dereference operation is safe anywhere that an ordinary iterator dereference
756 would be safe (i.e. when the iterator is not equal to end). Where this pattern
757 is followed (as in the ``llvm::object::Archive`` class) the result is much
758 cleaner iteration idiom:
763 for (auto &Child : Ar->children(Err)) {
764 // Use Child - we only enter the loop when it's valid
767 // Check Err after the loop to ensure it didn't break due to an error.
773 More information on Error and its related utilities can be found in the
776 Passing functions and other callable objects
777 --------------------------------------------
779 Sometimes you may want a function to be passed a callback object. In order to
780 support lambda expressions and other function objects, you should not use the
781 traditional C approach of taking a function pointer and an opaque cookie:
785 void takeCallback(bool (*Callback)(Function *, void *), void *Cookie);
787 Instead, use one of the following approaches:
792 If you don't mind putting the definition of your function into a header file,
793 make it a function template that is templated on the callable type.
797 template<typename Callable>
798 void takeCallback(Callable Callback) {
802 The ``function_ref`` class template
803 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
806 (`doxygen <http://llvm.org/docs/doxygen/html/classllvm_1_1function__ref_3_01Ret_07Params_8_8_8_08_4.html>`__) class
807 template represents a reference to a callable object, templated over the type
808 of the callable. This is a good choice for passing a callback to a function,
809 if you don't need to hold onto the callback after the function returns. In this
810 way, ``function_ref`` is to ``std::function`` as ``StringRef`` is to
813 ``function_ref<Ret(Param1, Param2, ...)>`` can be implicitly constructed from
814 any callable object that can be called with arguments of type ``Param1``,
815 ``Param2``, ..., and returns a value that can be converted to type ``Ret``.
820 void visitBasicBlocks(Function *F, function_ref<bool (BasicBlock*)> Callback) {
821 for (BasicBlock &BB : *F)
830 visitBasicBlocks(F, [&](BasicBlock *BB) {
836 Note that a ``function_ref`` object contains pointers to external memory, so it
837 is not generally safe to store an instance of the class (unless you know that
838 the external storage will not be freed). If you need this ability, consider
839 using ``std::function``. ``function_ref`` is small enough that it should always
844 The ``DEBUG()`` macro and ``-debug`` option
845 -------------------------------------------
847 Often when working on your pass you will put a bunch of debugging printouts and
848 other code into your pass. After you get it working, you want to remove it, but
849 you may need it again in the future (to work out new bugs that you run across).
851 Naturally, because of this, you don't want to delete the debug printouts, but
852 you don't want them to always be noisy. A standard compromise is to comment
853 them out, allowing you to enable them if you need them in the future.
855 The ``llvm/Support/Debug.h`` (`doxygen
856 <http://llvm.org/doxygen/Debug_8h-source.html>`__) file provides a macro named
857 ``DEBUG()`` that is a much nicer solution to this problem. Basically, you can
858 put arbitrary code into the argument of the ``DEBUG`` macro, and it is only
859 executed if '``opt``' (or any other tool) is run with the '``-debug``' command
864 DEBUG(errs() << "I am here!\n");
866 Then you can run your pass like this:
870 $ opt < a.bc > /dev/null -mypass
872 $ opt < a.bc > /dev/null -mypass -debug
875 Using the ``DEBUG()`` macro instead of a home-brewed solution allows you to not
876 have to create "yet another" command line option for the debug output for your
877 pass. Note that ``DEBUG()`` macros are disabled for non-asserts builds, so they
878 do not cause a performance impact at all (for the same reason, they should also
879 not contain side-effects!).
881 One additional nice thing about the ``DEBUG()`` macro is that you can enable or
882 disable it directly in gdb. Just use "``set DebugFlag=0``" or "``set
883 DebugFlag=1``" from the gdb if the program is running. If the program hasn't
884 been started yet, you can always just run it with ``-debug``.
888 Fine grained debug info with ``DEBUG_TYPE`` and the ``-debug-only`` option
889 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
891 Sometimes you may find yourself in a situation where enabling ``-debug`` just
892 turns on **too much** information (such as when working on the code generator).
893 If you want to enable debug information with more fine-grained control, you
894 should define the ``DEBUG_TYPE`` macro and use the ``-debug-only`` option as
899 #define DEBUG_TYPE "foo"
900 DEBUG(errs() << "'foo' debug type\n");
902 #define DEBUG_TYPE "bar"
903 DEBUG(errs() << "'bar' debug type\n"));
906 Then you can run your pass like this:
910 $ opt < a.bc > /dev/null -mypass
912 $ opt < a.bc > /dev/null -mypass -debug
915 $ opt < a.bc > /dev/null -mypass -debug-only=foo
917 $ opt < a.bc > /dev/null -mypass -debug-only=bar
919 $ opt < a.bc > /dev/null -mypass -debug-only=foo,bar
923 Of course, in practice, you should only set ``DEBUG_TYPE`` at the top of a file,
924 to specify the debug type for the entire module. Be careful that you only do
925 this after including Debug.h and not around any #include of headers. Also, you
926 should use names more meaningful than "foo" and "bar", because there is no
927 system in place to ensure that names do not conflict. If two different modules
928 use the same string, they will all be turned on when the name is specified.
929 This allows, for example, all debug information for instruction scheduling to be
930 enabled with ``-debug-only=InstrSched``, even if the source lives in multiple
931 files. The name must not include a comma (,) as that is used to separate the
932 arguments of the ``-debug-only`` option.
934 For performance reasons, -debug-only is not available in optimized build
935 (``--enable-optimized``) of LLVM.
937 The ``DEBUG_WITH_TYPE`` macro is also available for situations where you would
938 like to set ``DEBUG_TYPE``, but only for one specific ``DEBUG`` statement. It
939 takes an additional first parameter, which is the type to use. For example, the
940 preceding example could be written as:
944 DEBUG_WITH_TYPE("foo", errs() << "'foo' debug type\n");
945 DEBUG_WITH_TYPE("bar", errs() << "'bar' debug type\n"));
949 The ``Statistic`` class & ``-stats`` option
950 -------------------------------------------
952 The ``llvm/ADT/Statistic.h`` (`doxygen
953 <http://llvm.org/doxygen/Statistic_8h-source.html>`__) file provides a class
954 named ``Statistic`` that is used as a unified way to keep track of what the LLVM
955 compiler is doing and how effective various optimizations are. It is useful to
956 see what optimizations are contributing to making a particular program run
959 Often you may run your pass on some big program, and you're interested to see
960 how many times it makes a certain transformation. Although you can do this with
961 hand inspection, or some ad-hoc method, this is a real pain and not very useful
962 for big programs. Using the ``Statistic`` class makes it very easy to keep
963 track of this information, and the calculated information is presented in a
964 uniform manner with the rest of the passes being executed.
966 There are many examples of ``Statistic`` uses, but the basics of using it are as
969 #. Define your statistic like this:
973 #define DEBUG_TYPE "mypassname" // This goes before any #includes.
974 STATISTIC(NumXForms, "The # of times I did stuff");
976 The ``STATISTIC`` macro defines a static variable, whose name is specified by
977 the first argument. The pass name is taken from the ``DEBUG_TYPE`` macro, and
978 the description is taken from the second argument. The variable defined
979 ("NumXForms" in this case) acts like an unsigned integer.
981 #. Whenever you make a transformation, bump the counter:
985 ++NumXForms; // I did stuff!
987 That's all you have to do. To get '``opt``' to print out the statistics
988 gathered, use the '``-stats``' option:
992 $ opt -stats -mypassname < program.bc > /dev/null
993 ... statistics output ...
995 Note that in order to use the '``-stats``' option, LLVM must be
996 compiled with assertions enabled.
998 When running ``opt`` on a C file from the SPEC benchmark suite, it gives a
999 report that looks like this:
1001 .. code-block:: none
1003 7646 bitcodewriter - Number of normal instructions
1004 725 bitcodewriter - Number of oversized instructions
1005 129996 bitcodewriter - Number of bitcode bytes written
1006 2817 raise - Number of insts DCEd or constprop'd
1007 3213 raise - Number of cast-of-self removed
1008 5046 raise - Number of expression trees converted
1009 75 raise - Number of other getelementptr's formed
1010 138 raise - Number of load/store peepholes
1011 42 deadtypeelim - Number of unused typenames removed from symtab
1012 392 funcresolve - Number of varargs functions resolved
1013 27 globaldce - Number of global variables removed
1014 2 adce - Number of basic blocks removed
1015 134 cee - Number of branches revectored
1016 49 cee - Number of setcc instruction eliminated
1017 532 gcse - Number of loads removed
1018 2919 gcse - Number of instructions removed
1019 86 indvars - Number of canonical indvars added
1020 87 indvars - Number of aux indvars removed
1021 25 instcombine - Number of dead inst eliminate
1022 434 instcombine - Number of insts combined
1023 248 licm - Number of load insts hoisted
1024 1298 licm - Number of insts hoisted to a loop pre-header
1025 3 licm - Number of insts hoisted to multiple loop preds (bad, no loop pre-header)
1026 75 mem2reg - Number of alloca's promoted
1027 1444 cfgsimplify - Number of blocks simplified
1029 Obviously, with so many optimizations, having a unified framework for this stuff
1030 is very nice. Making your pass fit well into the framework makes it more
1031 maintainable and useful.
1035 Viewing graphs while debugging code
1036 -----------------------------------
1038 Several of the important data structures in LLVM are graphs: for example CFGs
1039 made out of LLVM :ref:`BasicBlocks <BasicBlock>`, CFGs made out of LLVM
1040 :ref:`MachineBasicBlocks <MachineBasicBlock>`, and :ref:`Instruction Selection
1041 DAGs <SelectionDAG>`. In many cases, while debugging various parts of the
1042 compiler, it is nice to instantly visualize these graphs.
1044 LLVM provides several callbacks that are available in a debug build to do
1045 exactly that. If you call the ``Function::viewCFG()`` method, for example, the
1046 current LLVM tool will pop up a window containing the CFG for the function where
1047 each basic block is a node in the graph, and each node contains the instructions
1048 in the block. Similarly, there also exists ``Function::viewCFGOnly()`` (does
1049 not include the instructions), the ``MachineFunction::viewCFG()`` and
1050 ``MachineFunction::viewCFGOnly()``, and the ``SelectionDAG::viewGraph()``
1051 methods. Within GDB, for example, you can usually use something like ``call
1052 DAG.viewGraph()`` to pop up a window. Alternatively, you can sprinkle calls to
1053 these functions in your code in places you want to debug.
1055 Getting this to work requires a small amount of setup. On Unix systems
1056 with X11, install the `graphviz <http://www.graphviz.org>`_ toolkit, and make
1057 sure 'dot' and 'gv' are in your path. If you are running on Mac OS X, download
1058 and install the Mac OS X `Graphviz program
1059 <http://www.pixelglow.com/graphviz/>`_ and add
1060 ``/Applications/Graphviz.app/Contents/MacOS/`` (or wherever you install it) to
1061 your path. The programs need not be present when configuring, building or
1062 running LLVM and can simply be installed when needed during an active debug
1065 ``SelectionDAG`` has been extended to make it easier to locate *interesting*
1066 nodes in large complex graphs. From gdb, if you ``call DAG.setGraphColor(node,
1067 "color")``, then the next ``call DAG.viewGraph()`` would highlight the node in
1068 the specified color (choices of colors can be found at `colors
1069 <http://www.graphviz.org/doc/info/colors.html>`_.) More complex node attributes
1070 can be provided with ``call DAG.setGraphAttrs(node, "attributes")`` (choices can
1071 be found at `Graph attributes <http://www.graphviz.org/doc/info/attrs.html>`_.)
1072 If you want to restart and clear all the current graph attributes, then you can
1073 ``call DAG.clearGraphAttrs()``.
1075 Note that graph visualization features are compiled out of Release builds to
1076 reduce file size. This means that you need a Debug+Asserts or Release+Asserts
1077 build to use these features.
1081 Picking the Right Data Structure for a Task
1082 ===========================================
1084 LLVM has a plethora of data structures in the ``llvm/ADT/`` directory, and we
1085 commonly use STL data structures. This section describes the trade-offs you
1086 should consider when you pick one.
1088 The first step is a choose your own adventure: do you want a sequential
1089 container, a set-like container, or a map-like container? The most important
1090 thing when choosing a container is the algorithmic properties of how you plan to
1091 access the container. Based on that, you should use:
1094 * a :ref:`map-like <ds_map>` container if you need efficient look-up of a
1095 value based on another value. Map-like containers also support efficient
1096 queries for containment (whether a key is in the map). Map-like containers
1097 generally do not support efficient reverse mapping (values to keys). If you
1098 need that, use two maps. Some map-like containers also support efficient
1099 iteration through the keys in sorted order. Map-like containers are the most
1100 expensive sort, only use them if you need one of these capabilities.
1102 * a :ref:`set-like <ds_set>` container if you need to put a bunch of stuff into
1103 a container that automatically eliminates duplicates. Some set-like
1104 containers support efficient iteration through the elements in sorted order.
1105 Set-like containers are more expensive than sequential containers.
1107 * a :ref:`sequential <ds_sequential>` container provides the most efficient way
1108 to add elements and keeps track of the order they are added to the collection.
1109 They permit duplicates and support efficient iteration, but do not support
1110 efficient look-up based on a key.
1112 * a :ref:`string <ds_string>` container is a specialized sequential container or
1113 reference structure that is used for character or byte arrays.
1115 * a :ref:`bit <ds_bit>` container provides an efficient way to store and
1116 perform set operations on sets of numeric id's, while automatically
1117 eliminating duplicates. Bit containers require a maximum of 1 bit for each
1118 identifier you want to store.
1120 Once the proper category of container is determined, you can fine tune the
1121 memory use, constant factors, and cache behaviors of access by intelligently
1122 picking a member of the category. Note that constant factors and cache behavior
1123 can be a big deal. If you have a vector that usually only contains a few
1124 elements (but could contain many), for example, it's much better to use
1125 :ref:`SmallVector <dss_smallvector>` than :ref:`vector <dss_vector>`. Doing so
1126 avoids (relatively) expensive malloc/free calls, which dwarf the cost of adding
1127 the elements to the container.
1131 Sequential Containers (std::vector, std::list, etc)
1132 ---------------------------------------------------
1134 There are a variety of sequential containers available for you, based on your
1135 needs. Pick the first in this section that will do what you want.
1142 The ``llvm::ArrayRef`` class is the preferred class to use in an interface that
1143 accepts a sequential list of elements in memory and just reads from them. By
1144 taking an ``ArrayRef``, the API can be passed a fixed size array, an
1145 ``std::vector``, an ``llvm::SmallVector`` and anything else that is contiguous
1148 .. _dss_fixedarrays:
1153 Fixed size arrays are very simple and very fast. They are good if you know
1154 exactly how many elements you have, or you have a (low) upper bound on how many
1159 Heap Allocated Arrays
1160 ^^^^^^^^^^^^^^^^^^^^^
1162 Heap allocated arrays (``new[]`` + ``delete[]``) are also simple. They are good
1163 if the number of elements is variable, if you know how many elements you will
1164 need before the array is allocated, and if the array is usually large (if not,
1165 consider a :ref:`SmallVector <dss_smallvector>`). The cost of a heap allocated
1166 array is the cost of the new/delete (aka malloc/free). Also note that if you
1167 are allocating an array of a type with a constructor, the constructor and
1168 destructors will be run for every element in the array (re-sizable vectors only
1169 construct those elements actually used).
1171 .. _dss_tinyptrvector:
1173 llvm/ADT/TinyPtrVector.h
1174 ^^^^^^^^^^^^^^^^^^^^^^^^
1176 ``TinyPtrVector<Type>`` is a highly specialized collection class that is
1177 optimized to avoid allocation in the case when a vector has zero or one
1178 elements. It has two major restrictions: 1) it can only hold values of pointer
1179 type, and 2) it cannot hold a null pointer.
1181 Since this container is highly specialized, it is rarely used.
1183 .. _dss_smallvector:
1185 llvm/ADT/SmallVector.h
1186 ^^^^^^^^^^^^^^^^^^^^^^
1188 ``SmallVector<Type, N>`` is a simple class that looks and smells just like
1189 ``vector<Type>``: it supports efficient iteration, lays out elements in memory
1190 order (so you can do pointer arithmetic between elements), supports efficient
1191 push_back/pop_back operations, supports efficient random access to its elements,
1194 The advantage of SmallVector is that it allocates space for some number of
1195 elements (N) **in the object itself**. Because of this, if the SmallVector is
1196 dynamically smaller than N, no malloc is performed. This can be a big win in
1197 cases where the malloc/free call is far more expensive than the code that
1198 fiddles around with the elements.
1200 This is good for vectors that are "usually small" (e.g. the number of
1201 predecessors/successors of a block is usually less than 8). On the other hand,
1202 this makes the size of the SmallVector itself large, so you don't want to
1203 allocate lots of them (doing so will waste a lot of space). As such,
1204 SmallVectors are most useful when on the stack.
1206 SmallVector also provides a nice portable and efficient replacement for
1211 Prefer to use ``SmallVectorImpl<T>`` as a parameter type.
1213 In APIs that don't care about the "small size" (most?), prefer to use
1214 the ``SmallVectorImpl<T>`` class, which is basically just the "vector
1215 header" (and methods) without the elements allocated after it. Note that
1216 ``SmallVector<T, N>`` inherits from ``SmallVectorImpl<T>`` so the
1217 conversion is implicit and costs nothing. E.g.
1221 // BAD: Clients cannot pass e.g. SmallVector<Foo, 4>.
1222 hardcodedSmallSize(SmallVector<Foo, 2> &Out);
1223 // GOOD: Clients can pass any SmallVector<Foo, N>.
1224 allowsAnySmallSize(SmallVectorImpl<Foo> &Out);
1227 SmallVector<Foo, 8> Vec;
1228 hardcodedSmallSize(Vec); // Error.
1229 allowsAnySmallSize(Vec); // Works.
1232 Even though it has "``Impl``" in the name, this is so widely used that
1233 it really isn't "private to the implementation" anymore. A name like
1234 ``SmallVectorHeader`` would be more appropriate.
1241 ``std::vector`` is well loved and respected. It is useful when SmallVector
1242 isn't: when the size of the vector is often large (thus the small optimization
1243 will rarely be a benefit) or if you will be allocating many instances of the
1244 vector itself (which would waste space for elements that aren't in the
1245 container). vector is also useful when interfacing with code that expects
1248 One worthwhile note about std::vector: avoid code like this:
1257 Instead, write this as:
1267 Doing so will save (at least) one heap allocation and free per iteration of the
1275 ``std::deque`` is, in some senses, a generalized version of ``std::vector``.
1276 Like ``std::vector``, it provides constant time random access and other similar
1277 properties, but it also provides efficient access to the front of the list. It
1278 does not guarantee continuity of elements within memory.
1280 In exchange for this extra flexibility, ``std::deque`` has significantly higher
1281 constant factor costs than ``std::vector``. If possible, use ``std::vector`` or
1289 ``std::list`` is an extremely inefficient class that is rarely useful. It
1290 performs a heap allocation for every element inserted into it, thus having an
1291 extremely high constant factor, particularly for small data types.
1292 ``std::list`` also only supports bidirectional iteration, not random access
1295 In exchange for this high cost, std::list supports efficient access to both ends
1296 of the list (like ``std::deque``, but unlike ``std::vector`` or
1297 ``SmallVector``). In addition, the iterator invalidation characteristics of
1298 std::list are stronger than that of a vector class: inserting or removing an
1299 element into the list does not invalidate iterator or pointers to other elements
1307 ``ilist<T>`` implements an 'intrusive' doubly-linked list. It is intrusive,
1308 because it requires the element to store and provide access to the prev/next
1309 pointers for the list.
1311 ``ilist`` has the same drawbacks as ``std::list``, and additionally requires an
1312 ``ilist_traits`` implementation for the element type, but it provides some novel
1313 characteristics. In particular, it can efficiently store polymorphic objects,
1314 the traits class is informed when an element is inserted or removed from the
1315 list, and ``ilist``\ s are guaranteed to support a constant-time splice
1318 These properties are exactly what we want for things like ``Instruction``\ s and
1319 basic blocks, which is why these are implemented with ``ilist``\ s.
1321 Related classes of interest are explained in the following subsections:
1323 * :ref:`ilist_traits <dss_ilist_traits>`
1325 * :ref:`iplist <dss_iplist>`
1327 * :ref:`llvm/ADT/ilist_node.h <dss_ilist_node>`
1329 * :ref:`Sentinels <dss_ilist_sentinel>`
1331 .. _dss_packedvector:
1333 llvm/ADT/PackedVector.h
1334 ^^^^^^^^^^^^^^^^^^^^^^^
1336 Useful for storing a vector of values using only a few number of bits for each
1337 value. Apart from the standard operations of a vector-like container, it can
1338 also perform an 'or' set operation.
1346 FirstCondition = 0x1,
1347 SecondCondition = 0x2,
1352 PackedVector<State, 2> Vec1;
1353 Vec1.push_back(FirstCondition);
1355 PackedVector<State, 2> Vec2;
1356 Vec2.push_back(SecondCondition);
1359 return Vec1[0]; // returns 'Both'.
1362 .. _dss_ilist_traits:
1367 ``ilist_traits<T>`` is ``ilist<T>``'s customization mechanism. ``iplist<T>``
1368 (and consequently ``ilist<T>``) publicly derive from this traits class.
1375 ``iplist<T>`` is ``ilist<T>``'s base and as such supports a slightly narrower
1376 interface. Notably, inserters from ``T&`` are absent.
1378 ``ilist_traits<T>`` is a public base of this class and can be used for a wide
1379 variety of customizations.
1383 llvm/ADT/ilist_node.h
1384 ^^^^^^^^^^^^^^^^^^^^^
1386 ``ilist_node<T>`` implements the forward and backward links that are expected
1387 by the ``ilist<T>`` (and analogous containers) in the default manner.
1389 ``ilist_node<T>``\ s are meant to be embedded in the node type ``T``, usually
1390 ``T`` publicly derives from ``ilist_node<T>``.
1392 .. _dss_ilist_sentinel:
1397 ``ilist``\ s have another specialty that must be considered. To be a good
1398 citizen in the C++ ecosystem, it needs to support the standard container
1399 operations, such as ``begin`` and ``end`` iterators, etc. Also, the
1400 ``operator--`` must work correctly on the ``end`` iterator in the case of
1401 non-empty ``ilist``\ s.
1403 The only sensible solution to this problem is to allocate a so-called *sentinel*
1404 along with the intrusive list, which serves as the ``end`` iterator, providing
1405 the back-link to the last element. However conforming to the C++ convention it
1406 is illegal to ``operator++`` beyond the sentinel and it also must not be
1409 These constraints allow for some implementation freedom to the ``ilist`` how to
1410 allocate and store the sentinel. The corresponding policy is dictated by
1411 ``ilist_traits<T>``. By default a ``T`` gets heap-allocated whenever the need
1412 for a sentinel arises.
1414 While the default policy is sufficient in most cases, it may break down when
1415 ``T`` does not provide a default constructor. Also, in the case of many
1416 instances of ``ilist``\ s, the memory overhead of the associated sentinels is
1417 wasted. To alleviate the situation with numerous and voluminous
1418 ``T``-sentinels, sometimes a trick is employed, leading to *ghostly sentinels*.
1420 Ghostly sentinels are obtained by specially-crafted ``ilist_traits<T>`` which
1421 superpose the sentinel with the ``ilist`` instance in memory. Pointer
1422 arithmetic is used to obtain the sentinel, which is relative to the ``ilist``'s
1423 ``this`` pointer. The ``ilist`` is augmented by an extra pointer, which serves
1424 as the back-link of the sentinel. This is the only field in the ghostly
1425 sentinel which can be legally accessed.
1429 Other Sequential Container options
1430 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
1432 Other STL containers are available, such as ``std::string``.
1434 There are also various STL adapter classes such as ``std::queue``,
1435 ``std::priority_queue``, ``std::stack``, etc. These provide simplified access
1436 to an underlying container but don't affect the cost of the container itself.
1440 String-like containers
1441 ----------------------
1443 There are a variety of ways to pass around and use strings in C and C++, and
1444 LLVM adds a few new options to choose from. Pick the first option on this list
1445 that will do what you need, they are ordered according to their relative cost.
1447 Note that it is generally preferred to *not* pass strings around as ``const
1448 char*``'s. These have a number of problems, including the fact that they
1449 cannot represent embedded nul ("\0") characters, and do not have a length
1450 available efficiently. The general replacement for '``const char*``' is
1453 For more information on choosing string containers for APIs, please see
1454 :ref:`Passing Strings <string_apis>`.
1458 llvm/ADT/StringRef.h
1459 ^^^^^^^^^^^^^^^^^^^^
1461 The StringRef class is a simple value class that contains a pointer to a
1462 character and a length, and is quite related to the :ref:`ArrayRef
1463 <dss_arrayref>` class (but specialized for arrays of characters). Because
1464 StringRef carries a length with it, it safely handles strings with embedded nul
1465 characters in it, getting the length does not require a strlen call, and it even
1466 has very convenient APIs for slicing and dicing the character range that it
1469 StringRef is ideal for passing simple strings around that are known to be live,
1470 either because they are C string literals, std::string, a C array, or a
1471 SmallVector. Each of these cases has an efficient implicit conversion to
1472 StringRef, which doesn't result in a dynamic strlen being executed.
1474 StringRef has a few major limitations which make more powerful string containers
1477 #. You cannot directly convert a StringRef to a 'const char*' because there is
1478 no way to add a trailing nul (unlike the .c_str() method on various stronger
1481 #. StringRef doesn't own or keep alive the underlying string bytes.
1482 As such it can easily lead to dangling pointers, and is not suitable for
1483 embedding in datastructures in most cases (instead, use an std::string or
1484 something like that).
1486 #. For the same reason, StringRef cannot be used as the return value of a
1487 method if the method "computes" the result string. Instead, use std::string.
1489 #. StringRef's do not allow you to mutate the pointed-to string bytes and it
1490 doesn't allow you to insert or remove bytes from the range. For editing
1491 operations like this, it interoperates with the :ref:`Twine <dss_twine>`
1494 Because of its strengths and limitations, it is very common for a function to
1495 take a StringRef and for a method on an object to return a StringRef that points
1496 into some string that it owns.
1503 The Twine class is used as an intermediary datatype for APIs that want to take a
1504 string that can be constructed inline with a series of concatenations. Twine
1505 works by forming recursive instances of the Twine datatype (a simple value
1506 object) on the stack as temporary objects, linking them together into a tree
1507 which is then linearized when the Twine is consumed. Twine is only safe to use
1508 as the argument to a function, and should always be a const reference, e.g.:
1512 void foo(const Twine &T);
1516 foo(X + "." + Twine(i));
1518 This example forms a string like "blarg.42" by concatenating the values
1519 together, and does not form intermediate strings containing "blarg" or "blarg.".
1521 Because Twine is constructed with temporary objects on the stack, and because
1522 these instances are destroyed at the end of the current statement, it is an
1523 inherently dangerous API. For example, this simple variant contains undefined
1524 behavior and will probably crash:
1528 void foo(const Twine &T);
1532 const Twine &Tmp = X + "." + Twine(i);
1535 ... because the temporaries are destroyed before the call. That said, Twine's
1536 are much more efficient than intermediate std::string temporaries, and they work
1537 really well with StringRef. Just be aware of their limitations.
1539 .. _dss_smallstring:
1541 llvm/ADT/SmallString.h
1542 ^^^^^^^^^^^^^^^^^^^^^^
1544 SmallString is a subclass of :ref:`SmallVector <dss_smallvector>` that adds some
1545 convenience APIs like += that takes StringRef's. SmallString avoids allocating
1546 memory in the case when the preallocated space is enough to hold its data, and
1547 it calls back to general heap allocation when required. Since it owns its data,
1548 it is very safe to use and supports full mutation of the string.
1550 Like SmallVector's, the big downside to SmallString is their sizeof. While they
1551 are optimized for small strings, they themselves are not particularly small.
1552 This means that they work great for temporary scratch buffers on the stack, but
1553 should not generally be put into the heap: it is very rare to see a SmallString
1554 as the member of a frequently-allocated heap data structure or returned
1562 The standard C++ std::string class is a very general class that (like
1563 SmallString) owns its underlying data. sizeof(std::string) is very reasonable
1564 so it can be embedded into heap data structures and returned by-value. On the
1565 other hand, std::string is highly inefficient for inline editing (e.g.
1566 concatenating a bunch of stuff together) and because it is provided by the
1567 standard library, its performance characteristics depend a lot of the host
1568 standard library (e.g. libc++ and MSVC provide a highly optimized string class,
1569 GCC contains a really slow implementation).
1571 The major disadvantage of std::string is that almost every operation that makes
1572 them larger can allocate memory, which is slow. As such, it is better to use
1573 SmallVector or Twine as a scratch buffer, but then use std::string to persist
1578 Set-Like Containers (std::set, SmallSet, SetVector, etc)
1579 --------------------------------------------------------
1581 Set-like containers are useful when you need to canonicalize multiple values
1582 into a single representation. There are several different choices for how to do
1583 this, providing various trade-offs.
1585 .. _dss_sortedvectorset:
1590 If you intend to insert a lot of elements, then do a lot of queries, a great
1591 approach is to use a vector (or other sequential container) with
1592 std::sort+std::unique to remove duplicates. This approach works really well if
1593 your usage pattern has these two distinct phases (insert then query), and can be
1594 coupled with a good choice of :ref:`sequential container <ds_sequential>`.
1596 This combination provides the several nice properties: the result data is
1597 contiguous in memory (good for cache locality), has few allocations, is easy to
1598 address (iterators in the final vector are just indices or pointers), and can be
1599 efficiently queried with a standard binary search (e.g.
1600 ``std::lower_bound``; if you want the whole range of elements comparing
1601 equal, use ``std::equal_range``).
1608 If you have a set-like data structure that is usually small and whose elements
1609 are reasonably small, a ``SmallSet<Type, N>`` is a good choice. This set has
1610 space for N elements in place (thus, if the set is dynamically smaller than N,
1611 no malloc traffic is required) and accesses them with a simple linear search.
1612 When the set grows beyond N elements, it allocates a more expensive
1613 representation that guarantees efficient access (for most types, it falls back
1614 to :ref:`std::set <dss_set>`, but for pointers it uses something far better,
1615 :ref:`SmallPtrSet <dss_smallptrset>`.
1617 The magic of this class is that it handles small sets extremely efficiently, but
1618 gracefully handles extremely large sets without loss of efficiency. The
1619 drawback is that the interface is quite small: it supports insertion, queries
1620 and erasing, but does not support iteration.
1622 .. _dss_smallptrset:
1624 llvm/ADT/SmallPtrSet.h
1625 ^^^^^^^^^^^^^^^^^^^^^^
1627 ``SmallPtrSet`` has all the advantages of ``SmallSet`` (and a ``SmallSet`` of
1628 pointers is transparently implemented with a ``SmallPtrSet``), but also supports
1629 iterators. If more than N insertions are performed, a single quadratically
1630 probed hash table is allocated and grows as needed, providing extremely
1631 efficient access (constant time insertion/deleting/queries with low constant
1632 factors) and is very stingy with malloc traffic.
1634 Note that, unlike :ref:`std::set <dss_set>`, the iterators of ``SmallPtrSet``
1635 are invalidated whenever an insertion occurs. Also, the values visited by the
1636 iterators are not visited in sorted order.
1640 llvm/ADT/StringSet.h
1641 ^^^^^^^^^^^^^^^^^^^^
1643 ``StringSet`` is a thin wrapper around :ref:`StringMap\<char\> <dss_stringmap>`,
1644 and it allows efficient storage and retrieval of unique strings.
1646 Functionally analogous to ``SmallSet<StringRef>``, ``StringSet`` also supports
1647 iteration. (The iterator dereferences to a ``StringMapEntry<char>``, so you
1648 need to call ``i->getKey()`` to access the item of the StringSet.) On the
1649 other hand, ``StringSet`` doesn't support range-insertion and
1650 copy-construction, which :ref:`SmallSet <dss_smallset>` and :ref:`SmallPtrSet
1651 <dss_smallptrset>` do support.
1658 DenseSet is a simple quadratically probed hash table. It excels at supporting
1659 small values: it uses a single allocation to hold all of the pairs that are
1660 currently inserted in the set. DenseSet is a great way to unique small values
1661 that are not simple pointers (use :ref:`SmallPtrSet <dss_smallptrset>` for
1662 pointers). Note that DenseSet has the same requirements for the value type that
1663 :ref:`DenseMap <dss_densemap>` has.
1667 llvm/ADT/SparseSet.h
1668 ^^^^^^^^^^^^^^^^^^^^
1670 SparseSet holds a small number of objects identified by unsigned keys of
1671 moderate size. It uses a lot of memory, but provides operations that are almost
1672 as fast as a vector. Typical keys are physical registers, virtual registers, or
1673 numbered basic blocks.
1675 SparseSet is useful for algorithms that need very fast clear/find/insert/erase
1676 and fast iteration over small sets. It is not intended for building composite
1679 .. _dss_sparsemultiset:
1681 llvm/ADT/SparseMultiSet.h
1682 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
1684 SparseMultiSet adds multiset behavior to SparseSet, while retaining SparseSet's
1685 desirable attributes. Like SparseSet, it typically uses a lot of memory, but
1686 provides operations that are almost as fast as a vector. Typical keys are
1687 physical registers, virtual registers, or numbered basic blocks.
1689 SparseMultiSet is useful for algorithms that need very fast
1690 clear/find/insert/erase of the entire collection, and iteration over sets of
1691 elements sharing a key. It is often a more efficient choice than using composite
1692 data structures (e.g. vector-of-vectors, map-of-vectors). It is not intended for
1693 building composite data structures.
1697 llvm/ADT/FoldingSet.h
1698 ^^^^^^^^^^^^^^^^^^^^^
1700 FoldingSet is an aggregate class that is really good at uniquing
1701 expensive-to-create or polymorphic objects. It is a combination of a chained
1702 hash table with intrusive links (uniqued objects are required to inherit from
1703 FoldingSetNode) that uses :ref:`SmallVector <dss_smallvector>` as part of its ID
1706 Consider a case where you want to implement a "getOrCreateFoo" method for a
1707 complex object (for example, a node in the code generator). The client has a
1708 description of **what** it wants to generate (it knows the opcode and all the
1709 operands), but we don't want to 'new' a node, then try inserting it into a set
1710 only to find out it already exists, at which point we would have to delete it
1711 and return the node that already exists.
1713 To support this style of client, FoldingSet perform a query with a
1714 FoldingSetNodeID (which wraps SmallVector) that can be used to describe the
1715 element that we want to query for. The query either returns the element
1716 matching the ID or it returns an opaque ID that indicates where insertion should
1717 take place. Construction of the ID usually does not require heap traffic.
1719 Because FoldingSet uses intrusive links, it can support polymorphic objects in
1720 the set (for example, you can have SDNode instances mixed with LoadSDNodes).
1721 Because the elements are individually allocated, pointers to the elements are
1722 stable: inserting or removing elements does not invalidate any pointers to other
1730 ``std::set`` is a reasonable all-around set class, which is decent at many
1731 things but great at nothing. std::set allocates memory for each element
1732 inserted (thus it is very malloc intensive) and typically stores three pointers
1733 per element in the set (thus adding a large amount of per-element space
1734 overhead). It offers guaranteed log(n) performance, which is not particularly
1735 fast from a complexity standpoint (particularly if the elements of the set are
1736 expensive to compare, like strings), and has extremely high constant factors for
1737 lookup, insertion and removal.
1739 The advantages of std::set are that its iterators are stable (deleting or
1740 inserting an element from the set does not affect iterators or pointers to other
1741 elements) and that iteration over the set is guaranteed to be in sorted order.
1742 If the elements in the set are large, then the relative overhead of the pointers
1743 and malloc traffic is not a big deal, but if the elements of the set are small,
1744 std::set is almost never a good choice.
1748 llvm/ADT/SetVector.h
1749 ^^^^^^^^^^^^^^^^^^^^
1751 LLVM's ``SetVector<Type>`` is an adapter class that combines your choice of a
1752 set-like container along with a :ref:`Sequential Container <ds_sequential>` The
1753 important property that this provides is efficient insertion with uniquing
1754 (duplicate elements are ignored) with iteration support. It implements this by
1755 inserting elements into both a set-like container and the sequential container,
1756 using the set-like container for uniquing and the sequential container for
1759 The difference between SetVector and other sets is that the order of iteration
1760 is guaranteed to match the order of insertion into the SetVector. This property
1761 is really important for things like sets of pointers. Because pointer values
1762 are non-deterministic (e.g. vary across runs of the program on different
1763 machines), iterating over the pointers in the set will not be in a well-defined
1766 The drawback of SetVector is that it requires twice as much space as a normal
1767 set and has the sum of constant factors from the set-like container and the
1768 sequential container that it uses. Use it **only** if you need to iterate over
1769 the elements in a deterministic order. SetVector is also expensive to delete
1770 elements out of (linear time), unless you use its "pop_back" method, which is
1773 ``SetVector`` is an adapter class that defaults to using ``std::vector`` and a
1774 size 16 ``SmallSet`` for the underlying containers, so it is quite expensive.
1775 However, ``"llvm/ADT/SetVector.h"`` also provides a ``SmallSetVector`` class,
1776 which defaults to using a ``SmallVector`` and ``SmallSet`` of a specified size.
1777 If you use this, and if your sets are dynamically smaller than ``N``, you will
1778 save a lot of heap traffic.
1780 .. _dss_uniquevector:
1782 llvm/ADT/UniqueVector.h
1783 ^^^^^^^^^^^^^^^^^^^^^^^
1785 UniqueVector is similar to :ref:`SetVector <dss_setvector>` but it retains a
1786 unique ID for each element inserted into the set. It internally contains a map
1787 and a vector, and it assigns a unique ID for each value inserted into the set.
1789 UniqueVector is very expensive: its cost is the sum of the cost of maintaining
1790 both the map and vector, it has high complexity, high constant factors, and
1791 produces a lot of malloc traffic. It should be avoided.
1793 .. _dss_immutableset:
1795 llvm/ADT/ImmutableSet.h
1796 ^^^^^^^^^^^^^^^^^^^^^^^
1798 ImmutableSet is an immutable (functional) set implementation based on an AVL
1799 tree. Adding or removing elements is done through a Factory object and results
1800 in the creation of a new ImmutableSet object. If an ImmutableSet already exists
1801 with the given contents, then the existing one is returned; equality is compared
1802 with a FoldingSetNodeID. The time and space complexity of add or remove
1803 operations is logarithmic in the size of the original set.
1805 There is no method for returning an element of the set, you can only check for
1810 Other Set-Like Container Options
1811 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
1813 The STL provides several other options, such as std::multiset and the various
1814 "hash_set" like containers (whether from C++ TR1 or from the SGI library). We
1815 never use hash_set and unordered_set because they are generally very expensive
1816 (each insertion requires a malloc) and very non-portable.
1818 std::multiset is useful if you're not interested in elimination of duplicates,
1819 but has all the drawbacks of :ref:`std::set <dss_set>`. A sorted vector
1820 (where you don't delete duplicate entries) or some other approach is almost
1825 Map-Like Containers (std::map, DenseMap, etc)
1826 ---------------------------------------------
1828 Map-like containers are useful when you want to associate data to a key. As
1829 usual, there are a lot of different ways to do this. :)
1831 .. _dss_sortedvectormap:
1836 If your usage pattern follows a strict insert-then-query approach, you can
1837 trivially use the same approach as :ref:`sorted vectors for set-like containers
1838 <dss_sortedvectorset>`. The only difference is that your query function (which
1839 uses std::lower_bound to get efficient log(n) lookup) should only compare the
1840 key, not both the key and value. This yields the same advantages as sorted
1845 llvm/ADT/StringMap.h
1846 ^^^^^^^^^^^^^^^^^^^^
1848 Strings are commonly used as keys in maps, and they are difficult to support
1849 efficiently: they are variable length, inefficient to hash and compare when
1850 long, expensive to copy, etc. StringMap is a specialized container designed to
1851 cope with these issues. It supports mapping an arbitrary range of bytes to an
1852 arbitrary other object.
1854 The StringMap implementation uses a quadratically-probed hash table, where the
1855 buckets store a pointer to the heap allocated entries (and some other stuff).
1856 The entries in the map must be heap allocated because the strings are variable
1857 length. The string data (key) and the element object (value) are stored in the
1858 same allocation with the string data immediately after the element object.
1859 This container guarantees the "``(char*)(&Value+1)``" points to the key string
1862 The StringMap is very fast for several reasons: quadratic probing is very cache
1863 efficient for lookups, the hash value of strings in buckets is not recomputed
1864 when looking up an element, StringMap rarely has to touch the memory for
1865 unrelated objects when looking up a value (even when hash collisions happen),
1866 hash table growth does not recompute the hash values for strings already in the
1867 table, and each pair in the map is store in a single allocation (the string data
1868 is stored in the same allocation as the Value of a pair).
1870 StringMap also provides query methods that take byte ranges, so it only ever
1871 copies a string if a value is inserted into the table.
1873 StringMap iteratation order, however, is not guaranteed to be deterministic, so
1874 any uses which require that should instead use a std::map.
1878 llvm/ADT/IndexedMap.h
1879 ^^^^^^^^^^^^^^^^^^^^^
1881 IndexedMap is a specialized container for mapping small dense integers (or
1882 values that can be mapped to small dense integers) to some other type. It is
1883 internally implemented as a vector with a mapping function that maps the keys
1884 to the dense integer range.
1886 This is useful for cases like virtual registers in the LLVM code generator: they
1887 have a dense mapping that is offset by a compile-time constant (the first
1888 virtual register ID).
1895 DenseMap is a simple quadratically probed hash table. It excels at supporting
1896 small keys and values: it uses a single allocation to hold all of the pairs
1897 that are currently inserted in the map. DenseMap is a great way to map
1898 pointers to pointers, or map other small types to each other.
1900 There are several aspects of DenseMap that you should be aware of, however.
1901 The iterators in a DenseMap are invalidated whenever an insertion occurs,
1902 unlike map. Also, because DenseMap allocates space for a large number of
1903 key/value pairs (it starts with 64 by default), it will waste a lot of space if
1904 your keys or values are large. Finally, you must implement a partial
1905 specialization of DenseMapInfo for the key that you want, if it isn't already
1906 supported. This is required to tell DenseMap about two special marker values
1907 (which can never be inserted into the map) that it needs internally.
1909 DenseMap's find_as() method supports lookup operations using an alternate key
1910 type. This is useful in cases where the normal key type is expensive to
1911 construct, but cheap to compare against. The DenseMapInfo is responsible for
1912 defining the appropriate comparison and hashing methods for each alternate key
1920 ValueMap is a wrapper around a :ref:`DenseMap <dss_densemap>` mapping
1921 ``Value*``\ s (or subclasses) to another type. When a Value is deleted or
1922 RAUW'ed, ValueMap will update itself so the new version of the key is mapped to
1923 the same value, just as if the key were a WeakVH. You can configure exactly how
1924 this happens, and what else happens on these two events, by passing a ``Config``
1925 parameter to the ValueMap template.
1927 .. _dss_intervalmap:
1929 llvm/ADT/IntervalMap.h
1930 ^^^^^^^^^^^^^^^^^^^^^^
1932 IntervalMap is a compact map for small keys and values. It maps key intervals
1933 instead of single keys, and it will automatically coalesce adjacent intervals.
1934 When the map only contains a few intervals, they are stored in the map object
1935 itself to avoid allocations.
1937 The IntervalMap iterators are quite big, so they should not be passed around as
1938 STL iterators. The heavyweight iterators allow a smaller data structure.
1945 std::map has similar characteristics to :ref:`std::set <dss_set>`: it uses a
1946 single allocation per pair inserted into the map, it offers log(n) lookup with
1947 an extremely large constant factor, imposes a space penalty of 3 pointers per
1948 pair in the map, etc.
1950 std::map is most useful when your keys or values are very large, if you need to
1951 iterate over the collection in sorted order, or if you need stable iterators
1952 into the map (i.e. they don't get invalidated if an insertion or deletion of
1953 another element takes place).
1957 llvm/ADT/MapVector.h
1958 ^^^^^^^^^^^^^^^^^^^^
1960 ``MapVector<KeyT,ValueT>`` provides a subset of the DenseMap interface. The
1961 main difference is that the iteration order is guaranteed to be the insertion
1962 order, making it an easy (but somewhat expensive) solution for non-deterministic
1963 iteration over maps of pointers.
1965 It is implemented by mapping from key to an index in a vector of key,value
1966 pairs. This provides fast lookup and iteration, but has two main drawbacks:
1967 the key is stored twice and removing elements takes linear time. If it is
1968 necessary to remove elements, it's best to remove them in bulk using
1971 .. _dss_inteqclasses:
1973 llvm/ADT/IntEqClasses.h
1974 ^^^^^^^^^^^^^^^^^^^^^^^
1976 IntEqClasses provides a compact representation of equivalence classes of small
1977 integers. Initially, each integer in the range 0..n-1 has its own equivalence
1978 class. Classes can be joined by passing two class representatives to the
1979 join(a, b) method. Two integers are in the same class when findLeader() returns
1980 the same representative.
1982 Once all equivalence classes are formed, the map can be compressed so each
1983 integer 0..n-1 maps to an equivalence class number in the range 0..m-1, where m
1984 is the total number of equivalence classes. The map must be uncompressed before
1985 it can be edited again.
1987 .. _dss_immutablemap:
1989 llvm/ADT/ImmutableMap.h
1990 ^^^^^^^^^^^^^^^^^^^^^^^
1992 ImmutableMap is an immutable (functional) map implementation based on an AVL
1993 tree. Adding or removing elements is done through a Factory object and results
1994 in the creation of a new ImmutableMap object. If an ImmutableMap already exists
1995 with the given key set, then the existing one is returned; equality is compared
1996 with a FoldingSetNodeID. The time and space complexity of add or remove
1997 operations is logarithmic in the size of the original map.
2001 Other Map-Like Container Options
2002 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2004 The STL provides several other options, such as std::multimap and the various
2005 "hash_map" like containers (whether from C++ TR1 or from the SGI library). We
2006 never use hash_set and unordered_set because they are generally very expensive
2007 (each insertion requires a malloc) and very non-portable.
2009 std::multimap is useful if you want to map a key to multiple values, but has all
2010 the drawbacks of std::map. A sorted vector or some other approach is almost
2015 Bit storage containers (BitVector, SparseBitVector)
2016 ---------------------------------------------------
2018 Unlike the other containers, there are only two bit storage containers, and
2019 choosing when to use each is relatively straightforward.
2021 One additional option is ``std::vector<bool>``: we discourage its use for two
2022 reasons 1) the implementation in many common compilers (e.g. commonly
2023 available versions of GCC) is extremely inefficient and 2) the C++ standards
2024 committee is likely to deprecate this container and/or change it significantly
2025 somehow. In any case, please don't use it.
2032 The BitVector container provides a dynamic size set of bits for manipulation.
2033 It supports individual bit setting/testing, as well as set operations. The set
2034 operations take time O(size of bitvector), but operations are performed one word
2035 at a time, instead of one bit at a time. This makes the BitVector very fast for
2036 set operations compared to other containers. Use the BitVector when you expect
2037 the number of set bits to be high (i.e. a dense set).
2039 .. _dss_smallbitvector:
2044 The SmallBitVector container provides the same interface as BitVector, but it is
2045 optimized for the case where only a small number of bits, less than 25 or so,
2046 are needed. It also transparently supports larger bit counts, but slightly less
2047 efficiently than a plain BitVector, so SmallBitVector should only be used when
2048 larger counts are rare.
2050 At this time, SmallBitVector does not support set operations (and, or, xor), and
2051 its operator[] does not provide an assignable lvalue.
2053 .. _dss_sparsebitvector:
2058 The SparseBitVector container is much like BitVector, with one major difference:
2059 Only the bits that are set, are stored. This makes the SparseBitVector much
2060 more space efficient than BitVector when the set is sparse, as well as making
2061 set operations O(number of set bits) instead of O(size of universe). The
2062 downside to the SparseBitVector is that setting and testing of random bits is
2063 O(N), and on large SparseBitVectors, this can be slower than BitVector. In our
2064 implementation, setting or testing bits in sorted order (either forwards or
2065 reverse) is O(1) worst case. Testing and setting bits within 128 bits (depends
2066 on size) of the current bit is also O(1). As a general statement,
2067 testing/setting bits in a SparseBitVector is O(distance away from last set bit).
2071 Helpful Hints for Common Operations
2072 ===================================
2074 This section describes how to perform some very simple transformations of LLVM
2075 code. This is meant to give examples of common idioms used, showing the
2076 practical side of LLVM transformations.
2078 Because this is a "how-to" section, you should also read about the main classes
2079 that you will be working with. The :ref:`Core LLVM Class Hierarchy Reference
2080 <coreclasses>` contains details and descriptions of the main classes that you
2085 Basic Inspection and Traversal Routines
2086 ---------------------------------------
2088 The LLVM compiler infrastructure have many different data structures that may be
2089 traversed. Following the example of the C++ standard template library, the
2090 techniques used to traverse these various data structures are all basically the
2091 same. For a enumerable sequence of values, the ``XXXbegin()`` function (or
2092 method) returns an iterator to the start of the sequence, the ``XXXend()``
2093 function returns an iterator pointing to one past the last valid element of the
2094 sequence, and there is some ``XXXiterator`` data type that is common between the
2097 Because the pattern for iteration is common across many different aspects of the
2098 program representation, the standard template library algorithms may be used on
2099 them, and it is easier to remember how to iterate. First we show a few common
2100 examples of the data structures that need to be traversed. Other data
2101 structures are traversed in very similar ways.
2103 .. _iterate_function:
2105 Iterating over the ``BasicBlock`` in a ``Function``
2106 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2108 It's quite common to have a ``Function`` instance that you'd like to transform
2109 in some way; in particular, you'd like to manipulate its ``BasicBlock``\ s. To
2110 facilitate this, you'll need to iterate over all of the ``BasicBlock``\ s that
2111 constitute the ``Function``. The following is an example that prints the name
2112 of a ``BasicBlock`` and the number of ``Instruction``\ s it contains:
2116 // func is a pointer to a Function instance
2117 for (Function::iterator i = func->begin(), e = func->end(); i != e; ++i)
2118 // Print out the name of the basic block if it has one, and then the
2119 // number of instructions that it contains
2120 errs() << "Basic block (name=" << i->getName() << ") has "
2121 << i->size() << " instructions.\n";
2123 Note that i can be used as if it were a pointer for the purposes of invoking
2124 member functions of the ``Instruction`` class. This is because the indirection
2125 operator is overloaded for the iterator classes. In the above code, the
2126 expression ``i->size()`` is exactly equivalent to ``(*i).size()`` just like
2129 .. _iterate_basicblock:
2131 Iterating over the ``Instruction`` in a ``BasicBlock``
2132 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2134 Just like when dealing with ``BasicBlock``\ s in ``Function``\ s, it's easy to
2135 iterate over the individual instructions that make up ``BasicBlock``\ s. Here's
2136 a code snippet that prints out each instruction in a ``BasicBlock``:
2140 // blk is a pointer to a BasicBlock instance
2141 for (BasicBlock::iterator i = blk->begin(), e = blk->end(); i != e; ++i)
2142 // The next statement works since operator<<(ostream&,...)
2143 // is overloaded for Instruction&
2144 errs() << *i << "\n";
2147 However, this isn't really the best way to print out the contents of a
2148 ``BasicBlock``! Since the ostream operators are overloaded for virtually
2149 anything you'll care about, you could have just invoked the print routine on the
2150 basic block itself: ``errs() << *blk << "\n";``.
2152 .. _iterate_insiter:
2154 Iterating over the ``Instruction`` in a ``Function``
2155 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2157 If you're finding that you commonly iterate over a ``Function``'s
2158 ``BasicBlock``\ s and then that ``BasicBlock``'s ``Instruction``\ s,
2159 ``InstIterator`` should be used instead. You'll need to include
2160 ``llvm/IR/InstIterator.h`` (`doxygen
2161 <http://llvm.org/doxygen/InstIterator_8h.html>`__) and then instantiate
2162 ``InstIterator``\ s explicitly in your code. Here's a small example that shows
2163 how to dump all instructions in a function to the standard error stream:
2167 #include "llvm/IR/InstIterator.h"
2169 // F is a pointer to a Function instance
2170 for (inst_iterator I = inst_begin(F), E = inst_end(F); I != E; ++I)
2171 errs() << *I << "\n";
2173 Easy, isn't it? You can also use ``InstIterator``\ s to fill a work list with
2174 its initial contents. For example, if you wanted to initialize a work list to
2175 contain all instructions in a ``Function`` F, all you would need to do is
2180 std::set<Instruction*> worklist;
2181 // or better yet, SmallPtrSet<Instruction*, 64> worklist;
2183 for (inst_iterator I = inst_begin(F), E = inst_end(F); I != E; ++I)
2184 worklist.insert(&*I);
2186 The STL set ``worklist`` would now contain all instructions in the ``Function``
2189 .. _iterate_convert:
2191 Turning an iterator into a class pointer (and vice-versa)
2192 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2194 Sometimes, it'll be useful to grab a reference (or pointer) to a class instance
2195 when all you've got at hand is an iterator. Well, extracting a reference or a
2196 pointer from an iterator is very straight-forward. Assuming that ``i`` is a
2197 ``BasicBlock::iterator`` and ``j`` is a ``BasicBlock::const_iterator``:
2201 Instruction& inst = *i; // Grab reference to instruction reference
2202 Instruction* pinst = &*i; // Grab pointer to instruction reference
2203 const Instruction& inst = *j;
2205 However, the iterators you'll be working with in the LLVM framework are special:
2206 they will automatically convert to a ptr-to-instance type whenever they need to.
2207 Instead of dereferencing the iterator and then taking the address of the result,
2208 you can simply assign the iterator to the proper pointer type and you get the
2209 dereference and address-of operation as a result of the assignment (behind the
2210 scenes, this is a result of overloading casting mechanisms). Thus the second
2211 line of the last example,
2215 Instruction *pinst = &*i;
2217 is semantically equivalent to
2221 Instruction *pinst = i;
2223 It's also possible to turn a class pointer into the corresponding iterator, and
2224 this is a constant time operation (very efficient). The following code snippet
2225 illustrates use of the conversion constructors provided by LLVM iterators. By
2226 using these, you can explicitly grab the iterator of something without actually
2227 obtaining it via iteration over some structure:
2231 void printNextInstruction(Instruction* inst) {
2232 BasicBlock::iterator it(inst);
2233 ++it; // After this line, it refers to the instruction after *inst
2234 if (it != inst->getParent()->end()) errs() << *it << "\n";
2237 Unfortunately, these implicit conversions come at a cost; they prevent these
2238 iterators from conforming to standard iterator conventions, and thus from being
2239 usable with standard algorithms and containers. For example, they prevent the
2240 following code, where ``B`` is a ``BasicBlock``, from compiling:
2244 llvm::SmallVector<llvm::Instruction *, 16>(B->begin(), B->end());
2246 Because of this, these implicit conversions may be removed some day, and
2247 ``operator*`` changed to return a pointer instead of a reference.
2249 .. _iterate_complex:
2251 Finding call sites: a slightly more complex example
2252 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2254 Say that you're writing a FunctionPass and would like to count all the locations
2255 in the entire module (that is, across every ``Function``) where a certain
2256 function (i.e., some ``Function *``) is already in scope. As you'll learn
2257 later, you may want to use an ``InstVisitor`` to accomplish this in a much more
2258 straight-forward manner, but this example will allow us to explore how you'd do
2259 it if you didn't have ``InstVisitor`` around. In pseudo-code, this is what we
2262 .. code-block:: none
2264 initialize callCounter to zero
2265 for each Function f in the Module
2266 for each BasicBlock b in f
2267 for each Instruction i in b
2268 if (i is a CallInst and calls the given function)
2269 increment callCounter
2271 And the actual code is (remember, because we're writing a ``FunctionPass``, our
2272 ``FunctionPass``-derived class simply has to override the ``runOnFunction``
2277 Function* targetFunc = ...;
2279 class OurFunctionPass : public FunctionPass {
2281 OurFunctionPass(): callCounter(0) { }
2283 virtual runOnFunction(Function& F) {
2284 for (Function::iterator b = F.begin(), be = F.end(); b != be; ++b) {
2285 for (BasicBlock::iterator i = b->begin(), ie = b->end(); i != ie; ++i) {
2286 if (CallInst* callInst = dyn_cast<CallInst>(&*i)) {
2287 // We know we've encountered a call instruction, so we
2288 // need to determine if it's a call to the
2289 // function pointed to by m_func or not.
2290 if (callInst->getCalledFunction() == targetFunc)
2298 unsigned callCounter;
2301 .. _calls_and_invokes:
2303 Treating calls and invokes the same way
2304 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2306 You may have noticed that the previous example was a bit oversimplified in that
2307 it did not deal with call sites generated by 'invoke' instructions. In this,
2308 and in other situations, you may find that you want to treat ``CallInst``\ s and
2309 ``InvokeInst``\ s the same way, even though their most-specific common base
2310 class is ``Instruction``, which includes lots of less closely-related things.
2311 For these cases, LLVM provides a handy wrapper class called ``CallSite``
2312 (`doxygen <http://llvm.org/doxygen/classllvm_1_1CallSite.html>`__) It is
2313 essentially a wrapper around an ``Instruction`` pointer, with some methods that
2314 provide functionality common to ``CallInst``\ s and ``InvokeInst``\ s.
2316 This class has "value semantics": it should be passed by value, not by reference
2317 and it should not be dynamically allocated or deallocated using ``operator new``
2318 or ``operator delete``. It is efficiently copyable, assignable and
2319 constructable, with costs equivalents to that of a bare pointer. If you look at
2320 its definition, it has only a single pointer member.
2324 Iterating over def-use & use-def chains
2325 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2327 Frequently, we might have an instance of the ``Value`` class (`doxygen
2328 <http://llvm.org/doxygen/classllvm_1_1Value.html>`__) and we want to determine
2329 which ``User`` s use the ``Value``. The list of all ``User``\ s of a particular
2330 ``Value`` is called a *def-use* chain. For example, let's say we have a
2331 ``Function*`` named ``F`` to a particular function ``foo``. Finding all of the
2332 instructions that *use* ``foo`` is as simple as iterating over the *def-use*
2339 for (User *U : F->users()) {
2340 if (Instruction *Inst = dyn_cast<Instruction>(U)) {
2341 errs() << "F is used in instruction:\n";
2342 errs() << *Inst << "\n";
2345 Alternatively, it's common to have an instance of the ``User`` Class (`doxygen
2346 <http://llvm.org/doxygen/classllvm_1_1User.html>`__) and need to know what
2347 ``Value``\ s are used by it. The list of all ``Value``\ s used by a ``User`` is
2348 known as a *use-def* chain. Instances of class ``Instruction`` are common
2349 ``User`` s, so we might want to iterate over all of the values that a particular
2350 instruction uses (that is, the operands of the particular ``Instruction``):
2354 Instruction *pi = ...;
2356 for (Use &U : pi->operands()) {
2361 Declaring objects as ``const`` is an important tool of enforcing mutation free
2362 algorithms (such as analyses, etc.). For this purpose above iterators come in
2363 constant flavors as ``Value::const_use_iterator`` and
2364 ``Value::const_op_iterator``. They automatically arise when calling
2365 ``use/op_begin()`` on ``const Value*``\ s or ``const User*``\ s respectively.
2366 Upon dereferencing, they return ``const Use*``\ s. Otherwise the above patterns
2371 Iterating over predecessors & successors of blocks
2372 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2374 Iterating over the predecessors and successors of a block is quite easy with the
2375 routines defined in ``"llvm/IR/CFG.h"``. Just use code like this to
2376 iterate over all predecessors of BB:
2380 #include "llvm/IR/CFG.h"
2381 BasicBlock *BB = ...;
2383 for (pred_iterator PI = pred_begin(BB), E = pred_end(BB); PI != E; ++PI) {
2384 BasicBlock *Pred = *PI;
2388 Similarly, to iterate over successors use ``succ_iterator/succ_begin/succ_end``.
2392 Making simple changes
2393 ---------------------
2395 There are some primitive transformation operations present in the LLVM
2396 infrastructure that are worth knowing about. When performing transformations,
2397 it's fairly common to manipulate the contents of basic blocks. This section
2398 describes some of the common methods for doing so and gives example code.
2400 .. _schanges_creating:
2402 Creating and inserting new ``Instruction``\ s
2403 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2405 *Instantiating Instructions*
2407 Creation of ``Instruction``\ s is straight-forward: simply call the constructor
2408 for the kind of instruction to instantiate and provide the necessary parameters.
2409 For example, an ``AllocaInst`` only *requires* a (const-ptr-to) ``Type``. Thus:
2413 AllocaInst* ai = new AllocaInst(Type::Int32Ty);
2415 will create an ``AllocaInst`` instance that represents the allocation of one
2416 integer in the current stack frame, at run time. Each ``Instruction`` subclass
2417 is likely to have varying default parameters which change the semantics of the
2418 instruction, so refer to the `doxygen documentation for the subclass of
2419 Instruction <http://llvm.org/doxygen/classllvm_1_1Instruction.html>`_ that
2420 you're interested in instantiating.
2424 It is very useful to name the values of instructions when you're able to, as
2425 this facilitates the debugging of your transformations. If you end up looking
2426 at generated LLVM machine code, you definitely want to have logical names
2427 associated with the results of instructions! By supplying a value for the
2428 ``Name`` (default) parameter of the ``Instruction`` constructor, you associate a
2429 logical name with the result of the instruction's execution at run time. For
2430 example, say that I'm writing a transformation that dynamically allocates space
2431 for an integer on the stack, and that integer is going to be used as some kind
2432 of index by some other code. To accomplish this, I place an ``AllocaInst`` at
2433 the first point in the first ``BasicBlock`` of some ``Function``, and I'm
2434 intending to use it within the same ``Function``. I might do:
2438 AllocaInst* pa = new AllocaInst(Type::Int32Ty, 0, "indexLoc");
2440 where ``indexLoc`` is now the logical name of the instruction's execution value,
2441 which is a pointer to an integer on the run time stack.
2443 *Inserting instructions*
2445 There are essentially three ways to insert an ``Instruction`` into an existing
2446 sequence of instructions that form a ``BasicBlock``:
2448 * Insertion into an explicit instruction list
2450 Given a ``BasicBlock* pb``, an ``Instruction* pi`` within that ``BasicBlock``,
2451 and a newly-created instruction we wish to insert before ``*pi``, we do the
2456 BasicBlock *pb = ...;
2457 Instruction *pi = ...;
2458 Instruction *newInst = new Instruction(...);
2460 pb->getInstList().insert(pi, newInst); // Inserts newInst before pi in pb
2462 Appending to the end of a ``BasicBlock`` is so common that the ``Instruction``
2463 class and ``Instruction``-derived classes provide constructors which take a
2464 pointer to a ``BasicBlock`` to be appended to. For example code that looked
2469 BasicBlock *pb = ...;
2470 Instruction *newInst = new Instruction(...);
2472 pb->getInstList().push_back(newInst); // Appends newInst to pb
2478 BasicBlock *pb = ...;
2479 Instruction *newInst = new Instruction(..., pb);
2481 which is much cleaner, especially if you are creating long instruction
2484 * Insertion into an implicit instruction list
2486 ``Instruction`` instances that are already in ``BasicBlock``\ s are implicitly
2487 associated with an existing instruction list: the instruction list of the
2488 enclosing basic block. Thus, we could have accomplished the same thing as the
2489 above code without being given a ``BasicBlock`` by doing:
2493 Instruction *pi = ...;
2494 Instruction *newInst = new Instruction(...);
2496 pi->getParent()->getInstList().insert(pi, newInst);
2498 In fact, this sequence of steps occurs so frequently that the ``Instruction``
2499 class and ``Instruction``-derived classes provide constructors which take (as
2500 a default parameter) a pointer to an ``Instruction`` which the newly-created
2501 ``Instruction`` should precede. That is, ``Instruction`` constructors are
2502 capable of inserting the newly-created instance into the ``BasicBlock`` of a
2503 provided instruction, immediately before that instruction. Using an
2504 ``Instruction`` constructor with a ``insertBefore`` (default) parameter, the
2509 Instruction* pi = ...;
2510 Instruction* newInst = new Instruction(..., pi);
2512 which is much cleaner, especially if you're creating a lot of instructions and
2513 adding them to ``BasicBlock``\ s.
2515 * Insertion using an instance of ``IRBuilder``
2517 Inserting several ``Instruction``\ s can be quite laborious using the previous
2518 methods. The ``IRBuilder`` is a convenience class that can be used to add
2519 several instructions to the end of a ``BasicBlock`` or before a particular
2520 ``Instruction``. It also supports constant folding and renaming named
2521 registers (see ``IRBuilder``'s template arguments).
2523 The example below demonstrates a very simple use of the ``IRBuilder`` where
2524 three instructions are inserted before the instruction ``pi``. The first two
2525 instructions are Call instructions and third instruction multiplies the return
2526 value of the two calls.
2530 Instruction *pi = ...;
2531 IRBuilder<> Builder(pi);
2532 CallInst* callOne = Builder.CreateCall(...);
2533 CallInst* callTwo = Builder.CreateCall(...);
2534 Value* result = Builder.CreateMul(callOne, callTwo);
2536 The example below is similar to the above example except that the created
2537 ``IRBuilder`` inserts instructions at the end of the ``BasicBlock`` ``pb``.
2541 BasicBlock *pb = ...;
2542 IRBuilder<> Builder(pb);
2543 CallInst* callOne = Builder.CreateCall(...);
2544 CallInst* callTwo = Builder.CreateCall(...);
2545 Value* result = Builder.CreateMul(callOne, callTwo);
2547 See :doc:`tutorial/LangImpl03` for a practical use of the ``IRBuilder``.
2550 .. _schanges_deleting:
2552 Deleting Instructions
2553 ^^^^^^^^^^^^^^^^^^^^^
2555 Deleting an instruction from an existing sequence of instructions that form a
2556 BasicBlock_ is very straight-forward: just call the instruction's
2557 ``eraseFromParent()`` method. For example:
2561 Instruction *I = .. ;
2562 I->eraseFromParent();
2564 This unlinks the instruction from its containing basic block and deletes it. If
2565 you'd just like to unlink the instruction from its containing basic block but
2566 not delete it, you can use the ``removeFromParent()`` method.
2568 .. _schanges_replacing:
2570 Replacing an Instruction with another Value
2571 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2573 Replacing individual instructions
2574 """""""""""""""""""""""""""""""""
2576 Including "`llvm/Transforms/Utils/BasicBlockUtils.h
2577 <http://llvm.org/doxygen/BasicBlockUtils_8h-source.html>`_" permits use of two
2578 very useful replace functions: ``ReplaceInstWithValue`` and
2579 ``ReplaceInstWithInst``.
2581 .. _schanges_deleting_sub:
2583 Deleting Instructions
2584 """""""""""""""""""""
2586 * ``ReplaceInstWithValue``
2588 This function replaces all uses of a given instruction with a value, and then
2589 removes the original instruction. The following example illustrates the
2590 replacement of the result of a particular ``AllocaInst`` that allocates memory
2591 for a single integer with a null pointer to an integer.
2595 AllocaInst* instToReplace = ...;
2596 BasicBlock::iterator ii(instToReplace);
2598 ReplaceInstWithValue(instToReplace->getParent()->getInstList(), ii,
2599 Constant::getNullValue(PointerType::getUnqual(Type::Int32Ty)));
2601 * ``ReplaceInstWithInst``
2603 This function replaces a particular instruction with another instruction,
2604 inserting the new instruction into the basic block at the location where the
2605 old instruction was, and replacing any uses of the old instruction with the
2606 new instruction. The following example illustrates the replacement of one
2607 ``AllocaInst`` with another.
2611 AllocaInst* instToReplace = ...;
2612 BasicBlock::iterator ii(instToReplace);
2614 ReplaceInstWithInst(instToReplace->getParent()->getInstList(), ii,
2615 new AllocaInst(Type::Int32Ty, 0, "ptrToReplacedInt"));
2618 Replacing multiple uses of Users and Values
2619 """""""""""""""""""""""""""""""""""""""""""
2621 You can use ``Value::replaceAllUsesWith`` and ``User::replaceUsesOfWith`` to
2622 change more than one use at a time. See the doxygen documentation for the
2623 `Value Class <http://llvm.org/doxygen/classllvm_1_1Value.html>`_ and `User Class
2624 <http://llvm.org/doxygen/classllvm_1_1User.html>`_, respectively, for more
2627 .. _schanges_deletingGV:
2629 Deleting GlobalVariables
2630 ^^^^^^^^^^^^^^^^^^^^^^^^
2632 Deleting a global variable from a module is just as easy as deleting an
2633 Instruction. First, you must have a pointer to the global variable that you
2634 wish to delete. You use this pointer to erase it from its parent, the module.
2639 GlobalVariable *GV = .. ;
2641 GV->eraseFromParent();
2649 In generating IR, you may need some complex types. If you know these types
2650 statically, you can use ``TypeBuilder<...>::get()``, defined in
2651 ``llvm/Support/TypeBuilder.h``, to retrieve them. ``TypeBuilder`` has two forms
2652 depending on whether you're building types for cross-compilation or native
2653 library use. ``TypeBuilder<T, true>`` requires that ``T`` be independent of the
2654 host environment, meaning that it's built out of types from the ``llvm::types``
2655 (`doxygen <http://llvm.org/doxygen/namespacellvm_1_1types.html>`__) namespace
2656 and pointers, functions, arrays, etc. built of those. ``TypeBuilder<T, false>``
2657 additionally allows native C types whose size may depend on the host compiler.
2662 FunctionType *ft = TypeBuilder<types::i<8>(types::i<32>*), true>::get();
2664 is easier to read and write than the equivalent
2668 std::vector<const Type*> params;
2669 params.push_back(PointerType::getUnqual(Type::Int32Ty));
2670 FunctionType *ft = FunctionType::get(Type::Int8Ty, params, false);
2672 See the `class comment
2673 <http://llvm.org/doxygen/TypeBuilder_8h-source.html#l00001>`_ for more details.
2680 This section describes the interaction of the LLVM APIs with multithreading,
2681 both on the part of client applications, and in the JIT, in the hosted
2684 Note that LLVM's support for multithreading is still relatively young. Up
2685 through version 2.5, the execution of threaded hosted applications was
2686 supported, but not threaded client access to the APIs. While this use case is
2687 now supported, clients *must* adhere to the guidelines specified below to ensure
2688 proper operation in multithreaded mode.
2690 Note that, on Unix-like platforms, LLVM requires the presence of GCC's atomic
2691 intrinsics in order to support threaded operation. If you need a
2692 multhreading-capable LLVM on a platform without a suitably modern system
2693 compiler, consider compiling LLVM and LLVM-GCC in single-threaded mode, and
2694 using the resultant compiler to build a copy of LLVM with multithreading
2699 Ending Execution with ``llvm_shutdown()``
2700 -----------------------------------------
2702 When you are done using the LLVM APIs, you should call ``llvm_shutdown()`` to
2703 deallocate memory used for internal structures.
2707 Lazy Initialization with ``ManagedStatic``
2708 ------------------------------------------
2710 ``ManagedStatic`` is a utility class in LLVM used to implement static
2711 initialization of static resources, such as the global type tables. In a
2712 single-threaded environment, it implements a simple lazy initialization scheme.
2713 When LLVM is compiled with support for multi-threading, however, it uses
2714 double-checked locking to implement thread-safe lazy initialization.
2718 Achieving Isolation with ``LLVMContext``
2719 ----------------------------------------
2721 ``LLVMContext`` is an opaque class in the LLVM API which clients can use to
2722 operate multiple, isolated instances of LLVM concurrently within the same
2723 address space. For instance, in a hypothetical compile-server, the compilation
2724 of an individual translation unit is conceptually independent from all the
2725 others, and it would be desirable to be able to compile incoming translation
2726 units concurrently on independent server threads. Fortunately, ``LLVMContext``
2727 exists to enable just this kind of scenario!
2729 Conceptually, ``LLVMContext`` provides isolation. Every LLVM entity
2730 (``Module``\ s, ``Value``\ s, ``Type``\ s, ``Constant``\ s, etc.) in LLVM's
2731 in-memory IR belongs to an ``LLVMContext``. Entities in different contexts
2732 *cannot* interact with each other: ``Module``\ s in different contexts cannot be
2733 linked together, ``Function``\ s cannot be added to ``Module``\ s in different
2734 contexts, etc. What this means is that is is safe to compile on multiple
2735 threads simultaneously, as long as no two threads operate on entities within the
2738 In practice, very few places in the API require the explicit specification of a
2739 ``LLVMContext``, other than the ``Type`` creation/lookup APIs. Because every
2740 ``Type`` carries a reference to its owning context, most other entities can
2741 determine what context they belong to by looking at their own ``Type``. If you
2742 are adding new entities to LLVM IR, please try to maintain this interface
2750 LLVM's "eager" JIT compiler is safe to use in threaded programs. Multiple
2751 threads can call ``ExecutionEngine::getPointerToFunction()`` or
2752 ``ExecutionEngine::runFunction()`` concurrently, and multiple threads can run
2753 code output by the JIT concurrently. The user must still ensure that only one
2754 thread accesses IR in a given ``LLVMContext`` while another thread might be
2755 modifying it. One way to do that is to always hold the JIT lock while accessing
2756 IR outside the JIT (the JIT *modifies* the IR by adding ``CallbackVH``\ s).
2757 Another way is to only call ``getPointerToFunction()`` from the
2758 ``LLVMContext``'s thread.
2760 When the JIT is configured to compile lazily (using
2761 ``ExecutionEngine::DisableLazyCompilation(false)``), there is currently a `race
2762 condition <http://llvm.org/bugs/show_bug.cgi?id=5184>`_ in updating call sites
2763 after a function is lazily-jitted. It's still possible to use the lazy JIT in a
2764 threaded program if you ensure that only one thread at a time can call any
2765 particular lazy stub and that the JIT lock guards any IR access, but we suggest
2766 using only the eager JIT in threaded programs.
2773 This section describes some of the advanced or obscure API's that most clients
2774 do not need to be aware of. These API's tend manage the inner workings of the
2775 LLVM system, and only need to be accessed in unusual circumstances.
2779 The ``ValueSymbolTable`` class
2780 ------------------------------
2782 The ``ValueSymbolTable`` (`doxygen
2783 <http://llvm.org/doxygen/classllvm_1_1ValueSymbolTable.html>`__) class provides
2784 a symbol table that the :ref:`Function <c_Function>` and Module_ classes use for
2785 naming value definitions. The symbol table can provide a name for any Value_.
2787 Note that the ``SymbolTable`` class should not be directly accessed by most
2788 clients. It should only be used when iteration over the symbol table names
2789 themselves are required, which is very special purpose. Note that not all LLVM
2790 Value_\ s have names, and those without names (i.e. they have an empty name) do
2791 not exist in the symbol table.
2793 Symbol tables support iteration over the values in the symbol table with
2794 ``begin/end/iterator`` and supports querying to see if a specific name is in the
2795 symbol table (with ``lookup``). The ``ValueSymbolTable`` class exposes no
2796 public mutator methods, instead, simply call ``setName`` on a value, which will
2797 autoinsert it into the appropriate symbol table.
2801 The ``User`` and owned ``Use`` classes' memory layout
2802 -----------------------------------------------------
2804 The ``User`` (`doxygen <http://llvm.org/doxygen/classllvm_1_1User.html>`__)
2805 class provides a basis for expressing the ownership of ``User`` towards other
2806 `Value instance <http://llvm.org/doxygen/classllvm_1_1Value.html>`_\ s. The
2807 ``Use`` (`doxygen <http://llvm.org/doxygen/classllvm_1_1Use.html>`__) helper
2808 class is employed to do the bookkeeping and to facilitate *O(1)* addition and
2813 Interaction and relationship between ``User`` and ``Use`` objects
2814 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2816 A subclass of ``User`` can choose between incorporating its ``Use`` objects or
2817 refer to them out-of-line by means of a pointer. A mixed variant (some ``Use``
2818 s inline others hung off) is impractical and breaks the invariant that the
2819 ``Use`` objects belonging to the same ``User`` form a contiguous array.
2821 We have 2 different layouts in the ``User`` (sub)classes:
2825 The ``Use`` object(s) are inside (resp. at fixed offset) of the ``User``
2826 object and there are a fixed number of them.
2830 The ``Use`` object(s) are referenced by a pointer to an array from the
2831 ``User`` object and there may be a variable number of them.
2833 As of v2.4 each layout still possesses a direct pointer to the start of the
2834 array of ``Use``\ s. Though not mandatory for layout a), we stick to this
2835 redundancy for the sake of simplicity. The ``User`` object also stores the
2836 number of ``Use`` objects it has. (Theoretically this information can also be
2837 calculated given the scheme presented below.)
2839 Special forms of allocation operators (``operator new``) enforce the following
2842 * Layout a) is modelled by prepending the ``User`` object by the ``Use[]``
2845 .. code-block:: none
2847 ...---.---.---.---.-------...
2848 | P | P | P | P | User
2849 '''---'---'---'---'-------'''
2851 * Layout b) is modelled by pointing at the ``Use[]`` array.
2853 .. code-block:: none
2864 *(In the above figures* '``P``' *stands for the* ``Use**`` *that is stored in
2865 each* ``Use`` *object in the member* ``Use::Prev`` *)*
2869 The waymarking algorithm
2870 ^^^^^^^^^^^^^^^^^^^^^^^^
2872 Since the ``Use`` objects are deprived of the direct (back)pointer to their
2873 ``User`` objects, there must be a fast and exact method to recover it. This is
2874 accomplished by the following scheme:
2876 A bit-encoding in the 2 LSBits (least significant bits) of the ``Use::Prev``
2877 allows to find the start of the ``User`` object:
2879 * ``00`` --- binary digit 0
2881 * ``01`` --- binary digit 1
2883 * ``10`` --- stop and calculate (``s``)
2885 * ``11`` --- full stop (``S``)
2887 Given a ``Use*``, all we have to do is to walk till we get a stop and we either
2888 have a ``User`` immediately behind or we have to walk to the next stop picking
2889 up digits and calculating the offset:
2891 .. code-block:: none
2893 .---.---.---.---.---.---.---.---.---.---.---.---.---.---.---.---.----------------
2894 | 1 | s | 1 | 0 | 1 | 0 | s | 1 | 1 | 0 | s | 1 | 1 | s | 1 | S | User (or User*)
2895 '---'---'---'---'---'---'---'---'---'---'---'---'---'---'---'---'----------------
2896 |+15 |+10 |+6 |+3 |+1
2899 | | | ______________________>
2900 | | ______________________________________>
2901 | __________________________________________________________>
2903 Only the significant number of bits need to be stored between the stops, so that
2904 the *worst case is 20 memory accesses* when there are 1000 ``Use`` objects
2905 associated with a ``User``.
2909 Reference implementation
2910 ^^^^^^^^^^^^^^^^^^^^^^^^
2912 The following literate Haskell fragment demonstrates the concept:
2914 .. code-block:: haskell
2916 > import Test.QuickCheck
2918 > digits :: Int -> [Char] -> [Char]
2919 > digits 0 acc = '0' : acc
2920 > digits 1 acc = '1' : acc
2921 > digits n acc = digits (n `div` 2) $ digits (n `mod` 2) acc
2923 > dist :: Int -> [Char] -> [Char]
2926 > dist 1 acc = let r = dist 0 acc in 's' : digits (length r) r
2927 > dist n acc = dist (n - 1) $ dist 1 acc
2929 > takeLast n ss = reverse $ take n $ reverse ss
2931 > test = takeLast 40 $ dist 20 []
2934 Printing <test> gives: ``"1s100000s11010s10100s1111s1010s110s11s1S"``
2936 The reverse algorithm computes the length of the string just by examining a
2939 .. code-block:: haskell
2941 > pref :: [Char] -> Int
2943 > pref ('s':'1':rest) = decode 2 1 rest
2944 > pref (_:rest) = 1 + pref rest
2946 > decode walk acc ('0':rest) = decode (walk + 1) (acc * 2) rest
2947 > decode walk acc ('1':rest) = decode (walk + 1) (acc * 2 + 1) rest
2948 > decode walk acc _ = walk + acc
2951 Now, as expected, printing <pref test> gives ``40``.
2953 We can *quickCheck* this with following property:
2955 .. code-block:: haskell
2957 > testcase = dist 2000 []
2958 > testcaseLength = length testcase
2960 > identityProp n = n > 0 && n <= testcaseLength ==> length arr == pref arr
2961 > where arr = takeLast n testcase
2964 As expected <quickCheck identityProp> gives:
2968 *Main> quickCheck identityProp
2969 OK, passed 100 tests.
2971 Let's be a bit more exhaustive:
2973 .. code-block:: haskell
2976 > deepCheck p = check (defaultConfig { configMaxTest = 500 }) p
2979 And here is the result of <deepCheck identityProp>:
2983 *Main> deepCheck identityProp
2984 OK, passed 500 tests.
2988 Tagging considerations
2989 ^^^^^^^^^^^^^^^^^^^^^^
2991 To maintain the invariant that the 2 LSBits of each ``Use**`` in ``Use`` never
2992 change after being set up, setters of ``Use::Prev`` must re-tag the new
2993 ``Use**`` on every modification. Accordingly getters must strip the tag bits.
2995 For layout b) instead of the ``User`` we find a pointer (``User*`` with LSBit
2996 set). Following this pointer brings us to the ``User``. A portable trick
2997 ensures that the first bytes of ``User`` (if interpreted as a pointer) never has
2998 the LSBit set. (Portability is relying on the fact that all known compilers
2999 place the ``vptr`` in the first word of the instances.)
3003 Designing Type Hiercharies and Polymorphic Interfaces
3004 -----------------------------------------------------
3006 There are two different design patterns that tend to result in the use of
3007 virtual dispatch for methods in a type hierarchy in C++ programs. The first is
3008 a genuine type hierarchy where different types in the hierarchy model
3009 a specific subset of the functionality and semantics, and these types nest
3010 strictly within each other. Good examples of this can be seen in the ``Value``
3011 or ``Type`` type hierarchies.
3013 A second is the desire to dispatch dynamically across a collection of
3014 polymorphic interface implementations. This latter use case can be modeled with
3015 virtual dispatch and inheritance by defining an abstract interface base class
3016 which all implementations derive from and override. However, this
3017 implementation strategy forces an **"is-a"** relationship to exist that is not
3018 actually meaningful. There is often not some nested hierarchy of useful
3019 generalizations which code might interact with and move up and down. Instead,
3020 there is a singular interface which is dispatched across a range of
3023 The preferred implementation strategy for the second use case is that of
3024 generic programming (sometimes called "compile-time duck typing" or "static
3025 polymorphism"). For example, a template over some type parameter ``T`` can be
3026 instantiated across any particular implementation that conforms to the
3027 interface or *concept*. A good example here is the highly generic properties of
3028 any type which models a node in a directed graph. LLVM models these primarily
3029 through templates and generic programming. Such templates include the
3030 ``LoopInfoBase`` and ``DominatorTreeBase``. When this type of polymorphism
3031 truly needs **dynamic** dispatch you can generalize it using a technique
3032 called *concept-based polymorphism*. This pattern emulates the interfaces and
3033 behaviors of templates using a very limited form of virtual dispatch for type
3034 erasure inside its implementation. You can find examples of this technique in
3035 the ``PassManager.h`` system, and there is a more detailed introduction to it
3036 by Sean Parent in several of his talks and papers:
3038 #. `Inheritance Is The Base Class of Evil
3039 <http://channel9.msdn.com/Events/GoingNative/2013/Inheritance-Is-The-Base-Class-of-Evil>`_
3040 - The GoingNative 2013 talk describing this technique, and probably the best
3042 #. `Value Semantics and Concepts-based Polymorphism
3043 <http://www.youtube.com/watch?v=_BpMYeUFXv8>`_ - The C++Now! 2012 talk
3044 describing this technique in more detail.
3045 #. `Sean Parent's Papers and Presentations
3046 <http://github.com/sean-parent/sean-parent.github.com/wiki/Papers-and-Presentations>`_
3047 - A Github project full of links to slides, video, and sometimes code.
3049 When deciding between creating a type hierarchy (with either tagged or virtual
3050 dispatch) and using templates or concepts-based polymorphism, consider whether
3051 there is some refinement of an abstract base class which is a semantically
3052 meaningful type on an interface boundary. If anything more refined than the
3053 root abstract interface is meaningless to talk about as a partial extension of
3054 the semantic model, then your use case likely fits better with polymorphism and
3055 you should avoid using virtual dispatch. However, there may be some exigent
3056 circumstances that require one technique or the other to be used.
3058 If you do need to introduce a type hierarchy, we prefer to use explicitly
3059 closed type hierarchies with manual tagged dispatch and/or RTTI rather than the
3060 open inheritance model and virtual dispatch that is more common in C++ code.
3061 This is because LLVM rarely encourages library consumers to extend its core
3062 types, and leverages the closed and tag-dispatched nature of its hierarchies to
3063 generate significantly more efficient code. We have also found that a large
3064 amount of our usage of type hierarchies fits better with tag-based pattern
3065 matching rather than dynamic dispatch across a common interface. Within LLVM we
3066 have built custom helpers to facilitate this design. See this document's
3067 section on :ref:`isa and dyn_cast <isa>` and our :doc:`detailed document
3068 <HowToSetUpLLVMStyleRTTI>` which describes how you can implement this
3069 pattern for use with the LLVM helpers.
3071 .. _abi_breaking_checks:
3076 Checks and asserts that alter the LLVM C++ ABI are predicated on the
3077 preprocessor symbol `LLVM_ENABLE_ABI_BREAKING_CHECKS` -- LLVM
3078 libraries built with `LLVM_ENABLE_ABI_BREAKING_CHECKS` are not ABI
3079 compatible LLVM libraries built without it defined. By default,
3080 turning on assertions also turns on `LLVM_ENABLE_ABI_BREAKING_CHECKS`
3081 so a default +Asserts build is not ABI compatible with a
3082 default -Asserts build. Clients that want ABI compatibility
3083 between +Asserts and -Asserts builds should use the CMake or autoconf
3084 build systems to set `LLVM_ENABLE_ABI_BREAKING_CHECKS` independently
3085 of `LLVM_ENABLE_ASSERTIONS`.
3089 The Core LLVM Class Hierarchy Reference
3090 =======================================
3092 ``#include "llvm/IR/Type.h"``
3094 header source: `Type.h <http://llvm.org/doxygen/Type_8h-source.html>`_
3096 doxygen info: `Type Clases <http://llvm.org/doxygen/classllvm_1_1Type.html>`_
3098 The Core LLVM classes are the primary means of representing the program being
3099 inspected or transformed. The core LLVM classes are defined in header files in
3100 the ``include/llvm/IR`` directory, and implemented in the ``lib/IR``
3101 directory. It's worth noting that, for historical reasons, this library is
3102 called ``libLLVMCore.so``, not ``libLLVMIR.so`` as you might expect.
3106 The Type class and Derived Types
3107 --------------------------------
3109 ``Type`` is a superclass of all type classes. Every ``Value`` has a ``Type``.
3110 ``Type`` cannot be instantiated directly but only through its subclasses.
3111 Certain primitive types (``VoidType``, ``LabelType``, ``FloatType`` and
3112 ``DoubleType``) have hidden subclasses. They are hidden because they offer no
3113 useful functionality beyond what the ``Type`` class offers except to distinguish
3114 themselves from other subclasses of ``Type``.
3116 All other types are subclasses of ``DerivedType``. Types can be named, but this
3117 is not a requirement. There exists exactly one instance of a given shape at any
3118 one time. This allows type equality to be performed with address equality of
3119 the Type Instance. That is, given two ``Type*`` values, the types are identical
3120 if the pointers are identical.
3124 Important Public Methods
3125 ^^^^^^^^^^^^^^^^^^^^^^^^
3127 * ``bool isIntegerTy() const``: Returns true for any integer type.
3129 * ``bool isFloatingPointTy()``: Return true if this is one of the five
3130 floating point types.
3132 * ``bool isSized()``: Return true if the type has known size. Things
3133 that don't have a size are abstract types, labels and void.
3137 Important Derived Types
3138 ^^^^^^^^^^^^^^^^^^^^^^^
3141 Subclass of DerivedType that represents integer types of any bit width. Any
3142 bit width between ``IntegerType::MIN_INT_BITS`` (1) and
3143 ``IntegerType::MAX_INT_BITS`` (~8 million) can be represented.
3145 * ``static const IntegerType* get(unsigned NumBits)``: get an integer
3146 type of a specific bit width.
3148 * ``unsigned getBitWidth() const``: Get the bit width of an integer type.
3151 This is subclassed by ArrayType, PointerType and VectorType.
3153 * ``const Type * getElementType() const``: Returns the type of each
3154 of the elements in the sequential type.
3157 This is a subclass of SequentialType and defines the interface for array
3160 * ``unsigned getNumElements() const``: Returns the number of elements
3164 Subclass of SequentialType for pointer types.
3167 Subclass of SequentialType for vector types. A vector type is similar to an
3168 ArrayType but is distinguished because it is a first class type whereas
3169 ArrayType is not. Vector types are used for vector operations and are usually
3170 small vectors of an integer or floating point type.
3173 Subclass of DerivedTypes for struct types.
3178 Subclass of DerivedTypes for function types.
3180 * ``bool isVarArg() const``: Returns true if it's a vararg function.
3182 * ``const Type * getReturnType() const``: Returns the return type of the
3185 * ``const Type * getParamType (unsigned i)``: Returns the type of the ith
3188 * ``const unsigned getNumParams() const``: Returns the number of formal
3193 The ``Module`` class
3194 --------------------
3196 ``#include "llvm/IR/Module.h"``
3198 header source: `Module.h <http://llvm.org/doxygen/Module_8h-source.html>`_
3200 doxygen info: `Module Class <http://llvm.org/doxygen/classllvm_1_1Module.html>`_
3202 The ``Module`` class represents the top level structure present in LLVM
3203 programs. An LLVM module is effectively either a translation unit of the
3204 original program or a combination of several translation units merged by the
3205 linker. The ``Module`` class keeps track of a list of :ref:`Function
3206 <c_Function>`\ s, a list of GlobalVariable_\ s, and a SymbolTable_.
3207 Additionally, it contains a few helpful member functions that try to make common
3212 Important Public Members of the ``Module`` class
3213 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3215 * ``Module::Module(std::string name = "")``
3217 Constructing a Module_ is easy. You can optionally provide a name for it
3218 (probably based on the name of the translation unit).
3220 * | ``Module::iterator`` - Typedef for function list iterator
3221 | ``Module::const_iterator`` - Typedef for const_iterator.
3222 | ``begin()``, ``end()``, ``size()``, ``empty()``
3224 These are forwarding methods that make it easy to access the contents of a
3225 ``Module`` object's :ref:`Function <c_Function>` list.
3227 * ``Module::FunctionListType &getFunctionList()``
3229 Returns the list of :ref:`Function <c_Function>`\ s. This is necessary to use
3230 when you need to update the list or perform a complex action that doesn't have
3231 a forwarding method.
3235 * | ``Module::global_iterator`` - Typedef for global variable list iterator
3236 | ``Module::const_global_iterator`` - Typedef for const_iterator.
3237 | ``global_begin()``, ``global_end()``, ``global_size()``, ``global_empty()``
3239 These are forwarding methods that make it easy to access the contents of a
3240 ``Module`` object's GlobalVariable_ list.
3242 * ``Module::GlobalListType &getGlobalList()``
3244 Returns the list of GlobalVariable_\ s. This is necessary to use when you
3245 need to update the list or perform a complex action that doesn't have a
3250 * ``SymbolTable *getSymbolTable()``
3252 Return a reference to the SymbolTable_ for this ``Module``.
3256 * ``Function *getFunction(StringRef Name) const``
3258 Look up the specified function in the ``Module`` SymbolTable_. If it does not
3259 exist, return ``null``.
3261 * ``Function *getOrInsertFunction(const std::string &Name, const FunctionType
3264 Look up the specified function in the ``Module`` SymbolTable_. If it does not
3265 exist, add an external declaration for the function and return it.
3267 * ``std::string getTypeName(const Type *Ty)``
3269 If there is at least one entry in the SymbolTable_ for the specified Type_,
3270 return it. Otherwise return the empty string.
3272 * ``bool addTypeName(const std::string &Name, const Type *Ty)``
3274 Insert an entry in the SymbolTable_ mapping ``Name`` to ``Ty``. If there is
3275 already an entry for this name, true is returned and the SymbolTable_ is not
3283 ``#include "llvm/IR/Value.h"``
3285 header source: `Value.h <http://llvm.org/doxygen/Value_8h-source.html>`_
3287 doxygen info: `Value Class <http://llvm.org/doxygen/classllvm_1_1Value.html>`_
3289 The ``Value`` class is the most important class in the LLVM Source base. It
3290 represents a typed value that may be used (among other things) as an operand to
3291 an instruction. There are many different types of ``Value``\ s, such as
3292 Constant_\ s, Argument_\ s. Even Instruction_\ s and :ref:`Function
3293 <c_Function>`\ s are ``Value``\ s.
3295 A particular ``Value`` may be used many times in the LLVM representation for a
3296 program. For example, an incoming argument to a function (represented with an
3297 instance of the Argument_ class) is "used" by every instruction in the function
3298 that references the argument. To keep track of this relationship, the ``Value``
3299 class keeps a list of all of the ``User``\ s that is using it (the User_ class
3300 is a base class for all nodes in the LLVM graph that can refer to ``Value``\ s).
3301 This use list is how LLVM represents def-use information in the program, and is
3302 accessible through the ``use_*`` methods, shown below.
3304 Because LLVM is a typed representation, every LLVM ``Value`` is typed, and this
3305 Type_ is available through the ``getType()`` method. In addition, all LLVM
3306 values can be named. The "name" of the ``Value`` is a symbolic string printed
3309 .. code-block:: llvm
3315 The name of this instruction is "foo". **NOTE** that the name of any value may
3316 be missing (an empty string), so names should **ONLY** be used for debugging
3317 (making the source code easier to read, debugging printouts), they should not be
3318 used to keep track of values or map between them. For this purpose, use a
3319 ``std::map`` of pointers to the ``Value`` itself instead.
3321 One important aspect of LLVM is that there is no distinction between an SSA
3322 variable and the operation that produces it. Because of this, any reference to
3323 the value produced by an instruction (or the value available as an incoming
3324 argument, for example) is represented as a direct pointer to the instance of the
3325 class that represents this value. Although this may take some getting used to,
3326 it simplifies the representation and makes it easier to manipulate.
3330 Important Public Members of the ``Value`` class
3331 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3333 * | ``Value::use_iterator`` - Typedef for iterator over the use-list
3334 | ``Value::const_use_iterator`` - Typedef for const_iterator over the
3336 | ``unsigned use_size()`` - Returns the number of users of the value.
3337 | ``bool use_empty()`` - Returns true if there are no users.
3338 | ``use_iterator use_begin()`` - Get an iterator to the start of the
3340 | ``use_iterator use_end()`` - Get an iterator to the end of the use-list.
3341 | ``User *use_back()`` - Returns the last element in the list.
3343 These methods are the interface to access the def-use information in LLVM.
3344 As with all other iterators in LLVM, the naming conventions follow the
3345 conventions defined by the STL_.
3347 * ``Type *getType() const``
3348 This method returns the Type of the Value.
3350 * | ``bool hasName() const``
3351 | ``std::string getName() const``
3352 | ``void setName(const std::string &Name)``
3354 This family of methods is used to access and assign a name to a ``Value``, be
3355 aware of the :ref:`precaution above <nameWarning>`.
3357 * ``void replaceAllUsesWith(Value *V)``
3359 This method traverses the use list of a ``Value`` changing all User_\ s of the
3360 current value to refer to "``V``" instead. For example, if you detect that an
3361 instruction always produces a constant value (for example through constant
3362 folding), you can replace all uses of the instruction with the constant like
3367 Inst->replaceAllUsesWith(ConstVal);
3374 ``#include "llvm/IR/User.h"``
3376 header source: `User.h <http://llvm.org/doxygen/User_8h-source.html>`_
3378 doxygen info: `User Class <http://llvm.org/doxygen/classllvm_1_1User.html>`_
3382 The ``User`` class is the common base class of all LLVM nodes that may refer to
3383 ``Value``\ s. It exposes a list of "Operands" that are all of the ``Value``\ s
3384 that the User is referring to. The ``User`` class itself is a subclass of
3387 The operands of a ``User`` point directly to the LLVM ``Value`` that it refers
3388 to. Because LLVM uses Static Single Assignment (SSA) form, there can only be
3389 one definition referred to, allowing this direct connection. This connection
3390 provides the use-def information in LLVM.
3394 Important Public Members of the ``User`` class
3395 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3397 The ``User`` class exposes the operand list in two ways: through an index access
3398 interface and through an iterator based interface.
3400 * | ``Value *getOperand(unsigned i)``
3401 | ``unsigned getNumOperands()``
3403 These two methods expose the operands of the ``User`` in a convenient form for
3406 * | ``User::op_iterator`` - Typedef for iterator over the operand list
3407 | ``op_iterator op_begin()`` - Get an iterator to the start of the operand
3409 | ``op_iterator op_end()`` - Get an iterator to the end of the operand list.
3411 Together, these methods make up the iterator based interface to the operands
3417 The ``Instruction`` class
3418 -------------------------
3420 ``#include "llvm/IR/Instruction.h"``
3422 header source: `Instruction.h
3423 <http://llvm.org/doxygen/Instruction_8h-source.html>`_
3425 doxygen info: `Instruction Class
3426 <http://llvm.org/doxygen/classllvm_1_1Instruction.html>`_
3428 Superclasses: User_, Value_
3430 The ``Instruction`` class is the common base class for all LLVM instructions.
3431 It provides only a few methods, but is a very commonly used class. The primary
3432 data tracked by the ``Instruction`` class itself is the opcode (instruction
3433 type) and the parent BasicBlock_ the ``Instruction`` is embedded into. To
3434 represent a specific type of instruction, one of many subclasses of
3435 ``Instruction`` are used.
3437 Because the ``Instruction`` class subclasses the User_ class, its operands can
3438 be accessed in the same way as for other ``User``\ s (with the
3439 ``getOperand()``/``getNumOperands()`` and ``op_begin()``/``op_end()`` methods).
3440 An important file for the ``Instruction`` class is the ``llvm/Instruction.def``
3441 file. This file contains some meta-data about the various different types of
3442 instructions in LLVM. It describes the enum values that are used as opcodes
3443 (for example ``Instruction::Add`` and ``Instruction::ICmp``), as well as the
3444 concrete sub-classes of ``Instruction`` that implement the instruction (for
3445 example BinaryOperator_ and CmpInst_). Unfortunately, the use of macros in this
3446 file confuses doxygen, so these enum values don't show up correctly in the
3447 `doxygen output <http://llvm.org/doxygen/classllvm_1_1Instruction.html>`_.
3451 Important Subclasses of the ``Instruction`` class
3452 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3456 * ``BinaryOperator``
3458 This subclasses represents all two operand instructions whose operands must be
3459 the same type, except for the comparison instructions.
3464 This subclass is the parent of the 12 casting instructions. It provides
3465 common operations on cast instructions.
3471 This subclass respresents the two comparison instructions,
3472 `ICmpInst <LangRef.html#i_icmp>`_ (integer opreands), and
3473 `FCmpInst <LangRef.html#i_fcmp>`_ (floating point operands).
3477 * ``TerminatorInst``
3479 This subclass is the parent of all terminator instructions (those which can
3484 Important Public Members of the ``Instruction`` class
3485 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3487 * ``BasicBlock *getParent()``
3489 Returns the BasicBlock_ that this
3490 ``Instruction`` is embedded into.
3492 * ``bool mayWriteToMemory()``
3494 Returns true if the instruction writes to memory, i.e. it is a ``call``,
3495 ``free``, ``invoke``, or ``store``.
3497 * ``unsigned getOpcode()``
3499 Returns the opcode for the ``Instruction``.
3501 * ``Instruction *clone() const``
3503 Returns another instance of the specified instruction, identical in all ways
3504 to the original except that the instruction has no parent (i.e. it's not
3505 embedded into a BasicBlock_), and it has no name.
3509 The ``Constant`` class and subclasses
3510 -------------------------------------
3512 Constant represents a base class for different types of constants. It is
3513 subclassed by ConstantInt, ConstantArray, etc. for representing the various
3514 types of Constants. GlobalValue_ is also a subclass, which represents the
3515 address of a global variable or function.
3519 Important Subclasses of Constant
3520 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3522 * ConstantInt : This subclass of Constant represents an integer constant of
3525 * ``const APInt& getValue() const``: Returns the underlying
3526 value of this constant, an APInt value.
3528 * ``int64_t getSExtValue() const``: Converts the underlying APInt value to an
3529 int64_t via sign extension. If the value (not the bit width) of the APInt
3530 is too large to fit in an int64_t, an assertion will result. For this
3531 reason, use of this method is discouraged.
3533 * ``uint64_t getZExtValue() const``: Converts the underlying APInt value
3534 to a uint64_t via zero extension. IF the value (not the bit width) of the
3535 APInt is too large to fit in a uint64_t, an assertion will result. For this
3536 reason, use of this method is discouraged.
3538 * ``static ConstantInt* get(const APInt& Val)``: Returns the ConstantInt
3539 object that represents the value provided by ``Val``. The type is implied
3540 as the IntegerType that corresponds to the bit width of ``Val``.
3542 * ``static ConstantInt* get(const Type *Ty, uint64_t Val)``: Returns the
3543 ConstantInt object that represents the value provided by ``Val`` for integer
3546 * ConstantFP : This class represents a floating point constant.
3548 * ``double getValue() const``: Returns the underlying value of this constant.
3550 * ConstantArray : This represents a constant array.
3552 * ``const std::vector<Use> &getValues() const``: Returns a vector of
3553 component constants that makeup this array.
3555 * ConstantStruct : This represents a constant struct.
3557 * ``const std::vector<Use> &getValues() const``: Returns a vector of
3558 component constants that makeup this array.
3560 * GlobalValue : This represents either a global variable or a function. In
3561 either case, the value is a constant fixed address (after linking).
3565 The ``GlobalValue`` class
3566 -------------------------
3568 ``#include "llvm/IR/GlobalValue.h"``
3570 header source: `GlobalValue.h
3571 <http://llvm.org/doxygen/GlobalValue_8h-source.html>`_
3573 doxygen info: `GlobalValue Class
3574 <http://llvm.org/doxygen/classllvm_1_1GlobalValue.html>`_
3576 Superclasses: Constant_, User_, Value_
3578 Global values ( GlobalVariable_\ s or :ref:`Function <c_Function>`\ s) are the
3579 only LLVM values that are visible in the bodies of all :ref:`Function
3580 <c_Function>`\ s. Because they are visible at global scope, they are also
3581 subject to linking with other globals defined in different translation units.
3582 To control the linking process, ``GlobalValue``\ s know their linkage rules.
3583 Specifically, ``GlobalValue``\ s know whether they have internal or external
3584 linkage, as defined by the ``LinkageTypes`` enumeration.
3586 If a ``GlobalValue`` has internal linkage (equivalent to being ``static`` in C),
3587 it is not visible to code outside the current translation unit, and does not
3588 participate in linking. If it has external linkage, it is visible to external
3589 code, and does participate in linking. In addition to linkage information,
3590 ``GlobalValue``\ s keep track of which Module_ they are currently part of.
3592 Because ``GlobalValue``\ s are memory objects, they are always referred to by
3593 their **address**. As such, the Type_ of a global is always a pointer to its
3594 contents. It is important to remember this when using the ``GetElementPtrInst``
3595 instruction because this pointer must be dereferenced first. For example, if
3596 you have a ``GlobalVariable`` (a subclass of ``GlobalValue)`` that is an array
3597 of 24 ints, type ``[24 x i32]``, then the ``GlobalVariable`` is a pointer to
3598 that array. Although the address of the first element of this array and the
3599 value of the ``GlobalVariable`` are the same, they have different types. The
3600 ``GlobalVariable``'s type is ``[24 x i32]``. The first element's type is
3601 ``i32.`` Because of this, accessing a global value requires you to dereference
3602 the pointer with ``GetElementPtrInst`` first, then its elements can be accessed.
3603 This is explained in the `LLVM Language Reference Manual
3604 <LangRef.html#globalvars>`_.
3608 Important Public Members of the ``GlobalValue`` class
3609 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3611 * | ``bool hasInternalLinkage() const``
3612 | ``bool hasExternalLinkage() const``
3613 | ``void setInternalLinkage(bool HasInternalLinkage)``
3615 These methods manipulate the linkage characteristics of the ``GlobalValue``.
3617 * ``Module *getParent()``
3619 This returns the Module_ that the
3620 GlobalValue is currently embedded into.
3624 The ``Function`` class
3625 ----------------------
3627 ``#include "llvm/IR/Function.h"``
3629 header source: `Function.h <http://llvm.org/doxygen/Function_8h-source.html>`_
3631 doxygen info: `Function Class
3632 <http://llvm.org/doxygen/classllvm_1_1Function.html>`_
3634 Superclasses: GlobalValue_, Constant_, User_, Value_
3636 The ``Function`` class represents a single procedure in LLVM. It is actually
3637 one of the more complex classes in the LLVM hierarchy because it must keep track
3638 of a large amount of data. The ``Function`` class keeps track of a list of
3639 BasicBlock_\ s, a list of formal Argument_\ s, and a SymbolTable_.
3641 The list of BasicBlock_\ s is the most commonly used part of ``Function``
3642 objects. The list imposes an implicit ordering of the blocks in the function,
3643 which indicate how the code will be laid out by the backend. Additionally, the
3644 first BasicBlock_ is the implicit entry node for the ``Function``. It is not
3645 legal in LLVM to explicitly branch to this initial block. There are no implicit
3646 exit nodes, and in fact there may be multiple exit nodes from a single
3647 ``Function``. If the BasicBlock_ list is empty, this indicates that the
3648 ``Function`` is actually a function declaration: the actual body of the function
3649 hasn't been linked in yet.
3651 In addition to a list of BasicBlock_\ s, the ``Function`` class also keeps track
3652 of the list of formal Argument_\ s that the function receives. This container
3653 manages the lifetime of the Argument_ nodes, just like the BasicBlock_ list does
3654 for the BasicBlock_\ s.
3656 The SymbolTable_ is a very rarely used LLVM feature that is only used when you
3657 have to look up a value by name. Aside from that, the SymbolTable_ is used
3658 internally to make sure that there are not conflicts between the names of
3659 Instruction_\ s, BasicBlock_\ s, or Argument_\ s in the function body.
3661 Note that ``Function`` is a GlobalValue_ and therefore also a Constant_. The
3662 value of the function is its address (after linking) which is guaranteed to be
3667 Important Public Members of the ``Function``
3668 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3670 * ``Function(const FunctionType *Ty, LinkageTypes Linkage,
3671 const std::string &N = "", Module* Parent = 0)``
3673 Constructor used when you need to create new ``Function``\ s to add the
3674 program. The constructor must specify the type of the function to create and
3675 what type of linkage the function should have. The FunctionType_ argument
3676 specifies the formal arguments and return value for the function. The same
3677 FunctionType_ value can be used to create multiple functions. The ``Parent``
3678 argument specifies the Module in which the function is defined. If this
3679 argument is provided, the function will automatically be inserted into that
3680 module's list of functions.
3682 * ``bool isDeclaration()``
3684 Return whether or not the ``Function`` has a body defined. If the function is
3685 "external", it does not have a body, and thus must be resolved by linking with
3686 a function defined in a different translation unit.
3688 * | ``Function::iterator`` - Typedef for basic block list iterator
3689 | ``Function::const_iterator`` - Typedef for const_iterator.
3690 | ``begin()``, ``end()``, ``size()``, ``empty()``
3692 These are forwarding methods that make it easy to access the contents of a
3693 ``Function`` object's BasicBlock_ list.
3695 * ``Function::BasicBlockListType &getBasicBlockList()``
3697 Returns the list of BasicBlock_\ s. This is necessary to use when you need to
3698 update the list or perform a complex action that doesn't have a forwarding
3701 * | ``Function::arg_iterator`` - Typedef for the argument list iterator
3702 | ``Function::const_arg_iterator`` - Typedef for const_iterator.
3703 | ``arg_begin()``, ``arg_end()``, ``arg_size()``, ``arg_empty()``
3705 These are forwarding methods that make it easy to access the contents of a
3706 ``Function`` object's Argument_ list.
3708 * ``Function::ArgumentListType &getArgumentList()``
3710 Returns the list of Argument_. This is necessary to use when you need to
3711 update the list or perform a complex action that doesn't have a forwarding
3714 * ``BasicBlock &getEntryBlock()``
3716 Returns the entry ``BasicBlock`` for the function. Because the entry block
3717 for the function is always the first block, this returns the first block of
3720 * | ``Type *getReturnType()``
3721 | ``FunctionType *getFunctionType()``
3723 This traverses the Type_ of the ``Function`` and returns the return type of
3724 the function, or the FunctionType_ of the actual function.
3726 * ``SymbolTable *getSymbolTable()``
3728 Return a pointer to the SymbolTable_ for this ``Function``.
3732 The ``GlobalVariable`` class
3733 ----------------------------
3735 ``#include "llvm/IR/GlobalVariable.h"``
3737 header source: `GlobalVariable.h
3738 <http://llvm.org/doxygen/GlobalVariable_8h-source.html>`_
3740 doxygen info: `GlobalVariable Class
3741 <http://llvm.org/doxygen/classllvm_1_1GlobalVariable.html>`_
3743 Superclasses: GlobalValue_, Constant_, User_, Value_
3745 Global variables are represented with the (surprise surprise) ``GlobalVariable``
3746 class. Like functions, ``GlobalVariable``\ s are also subclasses of
3747 GlobalValue_, and as such are always referenced by their address (global values
3748 must live in memory, so their "name" refers to their constant address). See
3749 GlobalValue_ for more on this. Global variables may have an initial value
3750 (which must be a Constant_), and if they have an initializer, they may be marked
3751 as "constant" themselves (indicating that their contents never change at
3754 .. _m_GlobalVariable:
3756 Important Public Members of the ``GlobalVariable`` class
3757 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3759 * ``GlobalVariable(const Type *Ty, bool isConstant, LinkageTypes &Linkage,
3760 Constant *Initializer = 0, const std::string &Name = "", Module* Parent = 0)``
3762 Create a new global variable of the specified type. If ``isConstant`` is true
3763 then the global variable will be marked as unchanging for the program. The
3764 Linkage parameter specifies the type of linkage (internal, external, weak,
3765 linkonce, appending) for the variable. If the linkage is InternalLinkage,
3766 WeakAnyLinkage, WeakODRLinkage, LinkOnceAnyLinkage or LinkOnceODRLinkage, then
3767 the resultant global variable will have internal linkage. AppendingLinkage
3768 concatenates together all instances (in different translation units) of the
3769 variable into a single variable but is only applicable to arrays. See the
3770 `LLVM Language Reference <LangRef.html#modulestructure>`_ for further details
3771 on linkage types. Optionally an initializer, a name, and the module to put
3772 the variable into may be specified for the global variable as well.
3774 * ``bool isConstant() const``
3776 Returns true if this is a global variable that is known not to be modified at
3779 * ``bool hasInitializer()``
3781 Returns true if this ``GlobalVariable`` has an intializer.
3783 * ``Constant *getInitializer()``
3785 Returns the initial value for a ``GlobalVariable``. It is not legal to call
3786 this method if there is no initializer.
3790 The ``BasicBlock`` class
3791 ------------------------
3793 ``#include "llvm/IR/BasicBlock.h"``
3795 header source: `BasicBlock.h
3796 <http://llvm.org/doxygen/BasicBlock_8h-source.html>`_
3798 doxygen info: `BasicBlock Class
3799 <http://llvm.org/doxygen/classllvm_1_1BasicBlock.html>`_
3803 This class represents a single entry single exit section of the code, commonly
3804 known as a basic block by the compiler community. The ``BasicBlock`` class
3805 maintains a list of Instruction_\ s, which form the body of the block. Matching
3806 the language definition, the last element of this list of instructions is always
3807 a terminator instruction (a subclass of the TerminatorInst_ class).
3809 In addition to tracking the list of instructions that make up the block, the
3810 ``BasicBlock`` class also keeps track of the :ref:`Function <c_Function>` that
3811 it is embedded into.
3813 Note that ``BasicBlock``\ s themselves are Value_\ s, because they are
3814 referenced by instructions like branches and can go in the switch tables.
3815 ``BasicBlock``\ s have type ``label``.
3819 Important Public Members of the ``BasicBlock`` class
3820 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3822 * ``BasicBlock(const std::string &Name = "", Function *Parent = 0)``
3824 The ``BasicBlock`` constructor is used to create new basic blocks for
3825 insertion into a function. The constructor optionally takes a name for the
3826 new block, and a :ref:`Function <c_Function>` to insert it into. If the
3827 ``Parent`` parameter is specified, the new ``BasicBlock`` is automatically
3828 inserted at the end of the specified :ref:`Function <c_Function>`, if not
3829 specified, the BasicBlock must be manually inserted into the :ref:`Function
3832 * | ``BasicBlock::iterator`` - Typedef for instruction list iterator
3833 | ``BasicBlock::const_iterator`` - Typedef for const_iterator.
3834 | ``begin()``, ``end()``, ``front()``, ``back()``,
3835 ``size()``, ``empty()``
3836 STL-style functions for accessing the instruction list.
3838 These methods and typedefs are forwarding functions that have the same
3839 semantics as the standard library methods of the same names. These methods
3840 expose the underlying instruction list of a basic block in a way that is easy
3841 to manipulate. To get the full complement of container operations (including
3842 operations to update the list), you must use the ``getInstList()`` method.
3844 * ``BasicBlock::InstListType &getInstList()``
3846 This method is used to get access to the underlying container that actually
3847 holds the Instructions. This method must be used when there isn't a
3848 forwarding function in the ``BasicBlock`` class for the operation that you
3849 would like to perform. Because there are no forwarding functions for
3850 "updating" operations, you need to use this if you want to update the contents
3851 of a ``BasicBlock``.
3853 * ``Function *getParent()``
3855 Returns a pointer to :ref:`Function <c_Function>` the block is embedded into,
3856 or a null pointer if it is homeless.
3858 * ``TerminatorInst *getTerminator()``
3860 Returns a pointer to the terminator instruction that appears at the end of the
3861 ``BasicBlock``. If there is no terminator instruction, or if the last
3862 instruction in the block is not a terminator, then a null pointer is returned.
3866 The ``Argument`` class
3867 ----------------------
3869 This subclass of Value defines the interface for incoming formal arguments to a
3870 function. A Function maintains a list of its formal arguments. An argument has
3871 a pointer to the parent Function.