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
266 .. _formatting_strings:
268 Formatting strings (the ``formatv`` function)
269 ---------------------------------------------
270 While LLVM doesn't necessarily do a lot of string manipulation and parsing, it
271 does do a lot of string formatting. From diagnostic messages, to llvm tool
272 outputs such as ``llvm-readobj`` to printing verbose disassembly listings and
273 LLDB runtime logging, the need for string formatting is pervasive.
275 The ``formatv`` is similar in spirit to ``printf``, but uses a different syntax
276 which borrows heavily from Python and C#. Unlike ``printf`` it deduces the type
277 to be formatted at compile time, so it does not need a format specifier such as
278 ``%d``. This reduces the mental overhead of trying to construct portable format
279 strings, especially for platform-specific types like ``size_t`` or pointer types.
280 Unlike both ``printf`` and Python, it additionally fails to compile if LLVM does
281 not know how to format the type. These two properties ensure that the function
282 is both safer and simpler to use than traditional formatting methods such as
283 the ``printf`` family of functions.
288 A call to ``formatv`` involves a single **format string** consisting of 0 or more
289 **replacement sequences**, followed by a variable length list of **replacement values**.
290 A replacement sequence is a string of the form ``{N[[,align]:style]}``.
292 ``N`` refers to the 0-based index of the argument from the list of replacement
293 values. Note that this means it is possible to reference the same parameter
294 multiple times, possibly with different style and/or alignment options, in any order.
296 ``align`` is an optional string specifying the width of the field to format
297 the value into, and the alignment of the value within the field. It is specified as
298 an optional **alignment style** followed by a positive integral **field width**. The
299 alignment style can be one of the characters ``-`` (left align), ``=`` (center align),
300 or ``+`` (right align). The default is right aligned.
302 ``style`` is an optional string consisting of a type specific that controls the
303 formatting of the value. For example, to format a floating point value as a percentage,
304 you can use the style option ``P``.
309 There are two ways to customize the formatting behavior for a type.
311 1. Provide a template specialization of ``llvm::format_provider<T>`` for your
312 type ``T`` with the appropriate static format method.
318 struct format_provider<MyFooBar> {
319 static void format(const MyFooBar &V, raw_ostream &Stream, StringRef Style) {
320 // Do whatever is necessary to format `V` into `Stream`
325 std::string S = formatv("{0}", X);
329 This is a useful extensibility mechanism for adding support for formatting your own
330 custom types with your own custom Style options. But it does not help when you want
331 to extend the mechanism for formatting a type that the library already knows how to
332 format. For that, we need something else.
334 2. Provide a **format adapter** with a non-static format method.
339 struct format_int_custom {
341 explicit format_int_custom(int N) : N(N) {}
342 void format(llvm::raw_ostream &Stream, StringRef Style) {
343 // Do whatever is necessary to format ``N`` into ``Stream``
349 std::string S = formatv("{0}", anything::format_int_custom(42));
353 If the search for a specialization of ``format_provider<T>`` for the given type
354 fails, ``formatv`` will subsequently check the argument for an instance method
355 named ``format`` with the signature described above. If so, it will call the
356 ``format`` method on the argument passing in the specified style. This allows
357 one to provide custom formatting of any type, including one which already has
358 a builtin format provider.
362 Below is intended to provide an incomplete set of examples demonstrating
363 the usage of ``formatv``. More information can be found by reading the
364 doxygen documentation or by looking at the unit test suite.
370 // Simple formatting of basic types and implicit string conversion.
371 S = formatv("{0} ({1:P})", 7, 0.35); // S == "7 (35.00%)"
373 // Out-of-order referencing and multi-referencing
374 outs() << formatv("{0} {2} {1} {0}", 1, "test", 3); // prints "1 3 test 1"
376 // Left, right, and center alignment
377 S = formatv("{0,7}", 'a'); // S == " a";
378 S = formatv("{0,-7}", 'a'); // S == "a ";
379 S = formatv("{0,=7}", 'a'); // S == " a ";
380 S = formatv("{0,+7}", 'a'); // S == " a";
383 S = formatv("{0:N} - {0:x} - {1:E}", 12345, 123908342); // S == "12,345 - 0x3039 - 1.24E8"
386 S = formatv("{0}", fmt_align(42, AlignStyle::Center, 7)); // S == " 42 "
387 S = formatv("{0}", fmt_repeat("hi", 3)); // S == "hihihi"
388 S = formatv("{0}", fmt_pad("hi", 2, 6)); // S == " hi "
391 std::vector<int> V = {8, 9, 10};
392 S = formatv("{0}", make_range(V.begin(), V.end())); // S == "8, 9, 10"
393 S = formatv("{0:$[+]}", make_range(V.begin(), V.end())); // S == "8+9+10"
394 S = formatv("{0:$[ + ]@[x]}", make_range(V.begin(), V.end())); // S == "0x8 + 0x9 + 0xA"
401 Proper error handling helps us identify bugs in our code, and helps end-users
402 understand errors in their tool usage. Errors fall into two broad categories:
403 *programmatic* and *recoverable*, with different strategies for handling and
409 Programmatic errors are violations of program invariants or API contracts, and
410 represent bugs within the program itself. Our aim is to document invariants, and
411 to abort quickly at the point of failure (providing some basic diagnostic) when
412 invariants are broken at runtime.
414 The fundamental tools for handling programmatic errors are assertions and the
415 llvm_unreachable function. Assertions are used to express invariant conditions,
416 and should include a message describing the invariant:
420 assert(isPhysReg(R) && "All virt regs should have been allocated already.");
422 The llvm_unreachable function can be used to document areas of control flow
423 that should never be entered if the program invariants hold:
427 enum { Foo, Bar, Baz } X = foo();
430 case Foo: /* Handle Foo */; break;
431 case Bar: /* Handle Bar */; break;
433 llvm_unreachable("X should be Foo or Bar here");
439 Recoverable errors represent an error in the program's environment, for example
440 a resource failure (a missing file, a dropped network connection, etc.), or
441 malformed input. These errors should be detected and communicated to a level of
442 the program where they can be handled appropriately. Handling the error may be
443 as simple as reporting the issue to the user, or it may involve attempts at
446 Recoverable errors are modeled using LLVM's ``Error`` scheme. This scheme
447 represents errors using function return values, similar to classic C integer
448 error codes, or C++'s ``std::error_code``. However, the ``Error`` class is
449 actually a lightweight wrapper for user-defined error types, allowing arbitrary
450 information to be attached to describe the error. This is similar to the way C++
451 exceptions allow throwing of user-defined types.
453 Success values are created by calling ``Error::success()``, E.g.:
460 return Error::success();
463 Success values are very cheap to construct and return - they have minimal
464 impact on program performance.
466 Failure values are constructed using ``make_error<T>``, where ``T`` is any class
467 that inherits from the ErrorInfo utility, E.g.:
471 class BadFileFormat : public ErrorInfo<BadFileFormat> {
476 BadFileFormat(StringRef Path) : Path(Path.str()) {}
478 void log(raw_ostream &OS) const override {
479 OS << Path << " is malformed";
482 std::error_code convertToErrorCode() const override {
483 return make_error_code(object_error::parse_failed);
487 char FileExists::ID; // This should be declared in the C++ file.
489 Error printFormattedFile(StringRef Path) {
490 if (<check for valid format>)
491 return make_error<InvalidObjectFile>(Path);
492 // print file contents.
493 return Error::success();
496 Error values can be implicitly converted to bool: true for error, false for
497 success, enabling the following idiom:
504 if (auto Err = mayFail())
506 // Success! We can proceed.
509 For functions that can fail but need to return a value the ``Expected<T>``
510 utility can be used. Values of this type can be constructed with either a
511 ``T``, or an ``Error``. Expected<T> values are also implicitly convertible to
512 boolean, but with the opposite convention to ``Error``: true for success, false
513 for error. If success, the ``T`` value can be accessed via the dereference
514 operator. If failure, the ``Error`` value can be extracted using the
515 ``takeError()`` method. Idiomatic usage looks like:
519 Expected<FormattedFile> openFormattedFile(StringRef Path) {
520 // If badly formatted, return an error.
521 if (auto Err = checkFormat(Path))
522 return std::move(Err);
523 // Otherwise return a FormattedFile instance.
524 return FormattedFile(Path);
527 Error processFormattedFile(StringRef Path) {
528 // Try to open a formatted file
529 if (auto FileOrErr = openFormattedFile(Path)) {
530 // On success, grab a reference to the file and continue.
531 auto &File = *FileOrErr;
534 // On error, extract the Error value and return it.
535 return FileOrErr.takeError();
538 If an ``Expected<T>`` value is in success mode then the ``takeError()`` method
539 will return a success value. Using this fact, the above function can be
544 Error processFormattedFile(StringRef Path) {
545 // Try to open a formatted file
546 auto FileOrErr = openFormattedFile(Path);
547 if (auto Err = FileOrErr.takeError())
548 // On error, extract the Error value and return it.
550 // On success, grab a reference to the file and continue.
551 auto &File = *FileOrErr;
555 This second form is often more readable for functions that involve multiple
556 ``Expected<T>`` values as it limits the indentation required.
558 All ``Error`` instances, whether success or failure, must be either checked or
559 moved from (via ``std::move`` or a return) before they are destructed.
560 Accidentally discarding an unchecked error will cause a program abort at the
561 point where the unchecked value's destructor is run, making it easy to identify
562 and fix violations of this rule.
564 Success values are considered checked once they have been tested (by invoking
565 the boolean conversion operator):
569 if (auto Err = canFail(...))
570 return Err; // Failure value - move error to caller.
572 // Safe to continue: Err was checked.
574 In contrast, the following code will always cause an abort, even if ``canFail``
575 returns a success value:
580 // Program will always abort here, even if canFail() returns Success, since
581 // the value is not checked.
583 Failure values are considered checked once a handler for the error type has
589 processFormattedFile(...),
590 [](const BadFileFormat &BFF) {
591 report("Unable to process " + BFF.Path + ": bad format");
593 [](const FileNotFound &FNF) {
594 report("File not found " + FNF.Path);
597 The ``handleErrors`` function takes an error as its first argument, followed by
598 a variadic list of "handlers", each of which must be a callable type (a
599 function, lambda, or class with a call operator) with one argument. The
600 ``handleErrors`` function will visit each handler in the sequence and check its
601 argument type against the dynamic type of the error, running the first handler
602 that matches. This is the same decision process that is used decide which catch
603 clause to run for a C++ exception.
605 Since the list of handlers passed to ``handleErrors`` may not cover every error
606 type that can occur, the ``handleErrors`` function also returns an Error value
607 that must be checked or propagated. If the error value that is passed to
608 ``handleErrors`` does not match any of the handlers it will be returned from
609 handleErrors. Idiomatic use of ``handleErrors`` thus looks like:
615 processFormattedFile(...),
616 [](const BadFileFormat &BFF) {
617 report("Unable to process " + BFF.Path + ": bad format");
619 [](const FileNotFound &FNF) {
620 report("File not found " + FNF.Path);
624 In cases where you truly know that the handler list is exhaustive the
625 ``handleAllErrors`` function can be used instead. This is identical to
626 ``handleErrors`` except that it will terminate the program if an unhandled
627 error is passed in, and can therefore return void. The ``handleAllErrors``
628 function should generally be avoided: the introduction of a new error type
629 elsewhere in the program can easily turn a formerly exhaustive list of errors
630 into a non-exhaustive list, risking unexpected program termination. Where
631 possible, use handleErrors and propagate unknown errors up the stack instead.
633 For tool code, where errors can be handled by printing an error message then
634 exiting with an error code, the :ref:`ExitOnError <err_exitonerr>` utility
635 may be a better choice than handleErrors, as it simplifies control flow when
636 calling fallible functions.
641 Many kinds of errors have no recovery strategy, the only action that can be
642 taken is to report them to the user so that the user can attempt to fix the
643 environment. In this case representing the error as a string makes perfect
644 sense. LLVM provides the ``StringError`` class for this purpose. It takes two
645 arguments: A string error message, and an equivalent ``std::error_code`` for
650 make_error<StringError>("Bad executable",
651 make_error_code(errc::executable_format_error"));
653 If you're certain that the error you're building will never need to be converted
654 to a ``std::error_code`` you can use the ``inconvertibleErrorCode()`` function:
658 make_error<StringError>("Bad executable", inconvertibleErrorCode());
660 This should be done only after careful consideration. If any attempt is made to
661 convert this error to a ``std::error_code`` it will trigger immediate program
662 termination. Unless you are certain that your errors will not need
663 interoperability you should look for an existing ``std::error_code`` that you
664 can convert to, and even (as painful as it is) consider introducing a new one as
667 Interoperability with std::error_code and ErrorOr
668 """""""""""""""""""""""""""""""""""""""""""""""""
670 Many existing LLVM APIs use ``std::error_code`` and its partner ``ErrorOr<T>``
671 (which plays the same role as ``Expected<T>``, but wraps a ``std::error_code``
672 rather than an ``Error``). The infectious nature of error types means that an
673 attempt to change one of these functions to return ``Error`` or ``Expected<T>``
674 instead often results in an avalanche of changes to callers, callers of callers,
675 and so on. (The first such attempt, returning an ``Error`` from
676 MachOObjectFile's constructor, was abandoned after the diff reached 3000 lines,
677 impacted half a dozen libraries, and was still growing).
679 To solve this problem, the ``Error``/``std::error_code`` interoperability requirement was
680 introduced. Two pairs of functions allow any ``Error`` value to be converted to a
681 ``std::error_code``, any ``Expected<T>`` to be converted to an ``ErrorOr<T>``, and vice
686 std::error_code errorToErrorCode(Error Err);
687 Error errorCodeToError(std::error_code EC);
689 template <typename T> ErrorOr<T> expectedToErrorOr(Expected<T> TOrErr);
690 template <typename T> Expected<T> errorOrToExpected(ErrorOr<T> TOrEC);
693 Using these APIs it is easy to make surgical patches that update individual
694 functions from ``std::error_code`` to ``Error``, and from ``ErrorOr<T>`` to
697 Returning Errors from error handlers
698 """"""""""""""""""""""""""""""""""""
700 Error recovery attempts may themselves fail. For that reason, ``handleErrors``
701 actually recognises three different forms of handler signature:
705 // Error must be handled, no new errors produced:
706 void(UserDefinedError &E);
708 // Error must be handled, new errors can be produced:
709 Error(UserDefinedError &E);
711 // Original error can be inspected, then re-wrapped and returned (or a new
712 // error can be produced):
713 Error(std::unique_ptr<UserDefinedError> E);
715 Any error returned from a handler will be returned from the ``handleErrors``
716 function so that it can be handled itself, or propagated up the stack.
720 Using ExitOnError to simplify tool code
721 """""""""""""""""""""""""""""""""""""""
723 Library code should never call ``exit`` for a recoverable error, however in tool
724 code (especially command line tools) this can be a reasonable approach. Calling
725 ``exit`` upon encountering an error dramatically simplifies control flow as the
726 error no longer needs to be propagated up the stack. This allows code to be
727 written in straight-line style, as long as each fallible call is wrapped in a
728 check and call to exit. The ``ExitOnError`` class supports this pattern by
729 providing call operators that inspect ``Error`` values, stripping the error away
730 in the success case and logging to ``stderr`` then exiting in the failure case.
732 To use this class, declare a global ``ExitOnError`` variable in your program:
736 ExitOnError ExitOnErr;
738 Calls to fallible functions can then be wrapped with a call to ``ExitOnErr``,
739 turning them into non-failing calls:
744 Expected<int> mayFail2();
747 ExitOnErr(mayFail());
748 int X = ExitOnErr(mayFail2());
751 On failure, the error's log message will be written to ``stderr``, optionally
752 preceded by a string "banner" that can be set by calling the setBanner method. A
753 mapping can also be supplied from ``Error`` values to exit codes using the
754 ``setExitCodeMapper`` method:
758 int main(int argc, char *argv[]) {
759 ExitOnErr.setBanner(std::string(argv[0]) + " error:");
760 ExitOnErr.setExitCodeMapper(
761 [](const Error &Err) {
762 if (Err.isA<BadFileFormat>())
767 Use ``ExitOnError`` in your tool code where possible as it can greatly improve
770 Fallible constructors
771 """""""""""""""""""""
773 Some classes require resource acquisition or other complex initialization that
774 can fail during construction. Unfortunately constructors can't return errors,
775 and having clients test objects after they're constructed to ensure that they're
776 valid is error prone as it's all too easy to forget the test. To work around
777 this, use the named constructor idiom and return an ``Expected<T>``:
784 static Expected<Foo> Create(Resource R1, Resource R2) {
788 return std::move(Err);
794 Foo(Resource R1, Resource R2, Error &Err) {
795 ErrorAsOutParameter EAO(&Err);
796 if (auto Err2 = R1.acquire()) {
797 Err = std::move(Err2);
805 Here, the named constructor passes an ``Error`` by reference into the actual
806 constructor, which the constructor can then use to return errors. The
807 ``ErrorAsOutParameter`` utility sets the ``Error`` value's checked flag on entry
808 to the constructor so that the error can be assigned to, then resets it on exit
809 to force the client (the named constructor) to check the error.
811 By using this idiom, clients attempting to construct a Foo receive either a
812 well-formed Foo or an Error, never an object in an invalid state.
814 Propagating and consuming errors based on types
815 """""""""""""""""""""""""""""""""""""""""""""""
817 In some contexts, certain types of error are known to be benign. For example,
818 when walking an archive, some clients may be happy to skip over badly formatted
819 object files rather than terminating the walk immediately. Skipping badly
820 formatted objects could be achieved using an elaborate handler method, but the
821 Error.h header provides two utilities that make this idiom much cleaner: the
822 type inspection method, ``isA``, and the ``consumeError`` function:
826 Error walkArchive(Archive A) {
827 for (unsigned I = 0; I != A.numMembers(); ++I) {
828 auto ChildOrErr = A.getMember(I);
829 if (auto Err = ChildOrErr.takeError()) {
830 if (Err.isA<BadFileFormat>())
831 consumeError(std::move(Err))
835 auto &Child = *ChildOrErr;
839 return Error::success();
842 Concatenating Errors with joinErrors
843 """"""""""""""""""""""""""""""""""""
845 In the archive walking example above ``BadFileFormat`` errors are simply
846 consumed and ignored. If the client had wanted report these errors after
847 completing the walk over the archive they could use the ``joinErrors`` utility:
851 Error walkArchive(Archive A) {
852 Error DeferredErrs = Error::success();
853 for (unsigned I = 0; I != A.numMembers(); ++I) {
854 auto ChildOrErr = A.getMember(I);
855 if (auto Err = ChildOrErr.takeError())
856 if (Err.isA<BadFileFormat>())
857 DeferredErrs = joinErrors(std::move(DeferredErrs), std::move(Err));
860 auto &Child = *ChildOrErr;
867 The ``joinErrors`` routine builds a special error type called ``ErrorList``,
868 which holds a list of user defined errors. The ``handleErrors`` routine
869 recognizes this type and will attempt to handle each of the contained erorrs in
870 order. If all contained errors can be handled, ``handleErrors`` will return
871 ``Error::success()``, otherwise ``handleErrors`` will concatenate the remaining
872 errors and return the resulting ``ErrorList``.
874 Building fallible iterators and iterator ranges
875 """""""""""""""""""""""""""""""""""""""""""""""
877 The archive walking examples above retrieve archive members by index, however
878 this requires considerable boiler-plate for iteration and error checking. We can
879 clean this up by using ``Error`` with the "fallible iterator" pattern. The usual
880 C++ iterator patterns do not allow for failure on increment, but we can
881 incorporate support for it by having iterators hold an Error reference through
882 which they can report failure. In this pattern, if an increment operation fails
883 the failure is recorded via the Error reference and the iterator value is set to
884 the end of the range in order to terminate the loop. This ensures that the
885 dereference operation is safe anywhere that an ordinary iterator dereference
886 would be safe (i.e. when the iterator is not equal to end). Where this pattern
887 is followed (as in the ``llvm::object::Archive`` class) the result is much
888 cleaner iteration idiom:
893 for (auto &Child : Ar->children(Err)) {
894 // Use Child - we only enter the loop when it's valid
897 // Check Err after the loop to ensure it didn't break due to an error.
903 More information on Error and its related utilities can be found in the
906 Passing functions and other callable objects
907 --------------------------------------------
909 Sometimes you may want a function to be passed a callback object. In order to
910 support lambda expressions and other function objects, you should not use the
911 traditional C approach of taking a function pointer and an opaque cookie:
915 void takeCallback(bool (*Callback)(Function *, void *), void *Cookie);
917 Instead, use one of the following approaches:
922 If you don't mind putting the definition of your function into a header file,
923 make it a function template that is templated on the callable type.
927 template<typename Callable>
928 void takeCallback(Callable Callback) {
932 The ``function_ref`` class template
933 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
936 (`doxygen <http://llvm.org/docs/doxygen/html/classllvm_1_1function__ref_3_01Ret_07Params_8_8_8_08_4.html>`__) class
937 template represents a reference to a callable object, templated over the type
938 of the callable. This is a good choice for passing a callback to a function,
939 if you don't need to hold onto the callback after the function returns. In this
940 way, ``function_ref`` is to ``std::function`` as ``StringRef`` is to
943 ``function_ref<Ret(Param1, Param2, ...)>`` can be implicitly constructed from
944 any callable object that can be called with arguments of type ``Param1``,
945 ``Param2``, ..., and returns a value that can be converted to type ``Ret``.
950 void visitBasicBlocks(Function *F, function_ref<bool (BasicBlock*)> Callback) {
951 for (BasicBlock &BB : *F)
960 visitBasicBlocks(F, [&](BasicBlock *BB) {
966 Note that a ``function_ref`` object contains pointers to external memory, so it
967 is not generally safe to store an instance of the class (unless you know that
968 the external storage will not be freed). If you need this ability, consider
969 using ``std::function``. ``function_ref`` is small enough that it should always
974 The ``DEBUG()`` macro and ``-debug`` option
975 -------------------------------------------
977 Often when working on your pass you will put a bunch of debugging printouts and
978 other code into your pass. After you get it working, you want to remove it, but
979 you may need it again in the future (to work out new bugs that you run across).
981 Naturally, because of this, you don't want to delete the debug printouts, but
982 you don't want them to always be noisy. A standard compromise is to comment
983 them out, allowing you to enable them if you need them in the future.
985 The ``llvm/Support/Debug.h`` (`doxygen
986 <http://llvm.org/doxygen/Debug_8h-source.html>`__) file provides a macro named
987 ``DEBUG()`` that is a much nicer solution to this problem. Basically, you can
988 put arbitrary code into the argument of the ``DEBUG`` macro, and it is only
989 executed if '``opt``' (or any other tool) is run with the '``-debug``' command
994 DEBUG(errs() << "I am here!\n");
996 Then you can run your pass like this:
1000 $ opt < a.bc > /dev/null -mypass
1002 $ opt < a.bc > /dev/null -mypass -debug
1005 Using the ``DEBUG()`` macro instead of a home-brewed solution allows you to not
1006 have to create "yet another" command line option for the debug output for your
1007 pass. Note that ``DEBUG()`` macros are disabled for non-asserts builds, so they
1008 do not cause a performance impact at all (for the same reason, they should also
1009 not contain side-effects!).
1011 One additional nice thing about the ``DEBUG()`` macro is that you can enable or
1012 disable it directly in gdb. Just use "``set DebugFlag=0``" or "``set
1013 DebugFlag=1``" from the gdb if the program is running. If the program hasn't
1014 been started yet, you can always just run it with ``-debug``.
1018 Fine grained debug info with ``DEBUG_TYPE`` and the ``-debug-only`` option
1019 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
1021 Sometimes you may find yourself in a situation where enabling ``-debug`` just
1022 turns on **too much** information (such as when working on the code generator).
1023 If you want to enable debug information with more fine-grained control, you
1024 should define the ``DEBUG_TYPE`` macro and use the ``-debug-only`` option as
1029 #define DEBUG_TYPE "foo"
1030 DEBUG(errs() << "'foo' debug type\n");
1032 #define DEBUG_TYPE "bar"
1033 DEBUG(errs() << "'bar' debug type\n"));
1036 Then you can run your pass like this:
1038 .. code-block:: none
1040 $ opt < a.bc > /dev/null -mypass
1042 $ opt < a.bc > /dev/null -mypass -debug
1045 $ opt < a.bc > /dev/null -mypass -debug-only=foo
1047 $ opt < a.bc > /dev/null -mypass -debug-only=bar
1049 $ opt < a.bc > /dev/null -mypass -debug-only=foo,bar
1053 Of course, in practice, you should only set ``DEBUG_TYPE`` at the top of a file,
1054 to specify the debug type for the entire module. Be careful that you only do
1055 this after including Debug.h and not around any #include of headers. Also, you
1056 should use names more meaningful than "foo" and "bar", because there is no
1057 system in place to ensure that names do not conflict. If two different modules
1058 use the same string, they will all be turned on when the name is specified.
1059 This allows, for example, all debug information for instruction scheduling to be
1060 enabled with ``-debug-only=InstrSched``, even if the source lives in multiple
1061 files. The name must not include a comma (,) as that is used to separate the
1062 arguments of the ``-debug-only`` option.
1064 For performance reasons, -debug-only is not available in optimized build
1065 (``--enable-optimized``) of LLVM.
1067 The ``DEBUG_WITH_TYPE`` macro is also available for situations where you would
1068 like to set ``DEBUG_TYPE``, but only for one specific ``DEBUG`` statement. It
1069 takes an additional first parameter, which is the type to use. For example, the
1070 preceding example could be written as:
1074 DEBUG_WITH_TYPE("foo", errs() << "'foo' debug type\n");
1075 DEBUG_WITH_TYPE("bar", errs() << "'bar' debug type\n"));
1079 The ``Statistic`` class & ``-stats`` option
1080 -------------------------------------------
1082 The ``llvm/ADT/Statistic.h`` (`doxygen
1083 <http://llvm.org/doxygen/Statistic_8h-source.html>`__) file provides a class
1084 named ``Statistic`` that is used as a unified way to keep track of what the LLVM
1085 compiler is doing and how effective various optimizations are. It is useful to
1086 see what optimizations are contributing to making a particular program run
1089 Often you may run your pass on some big program, and you're interested to see
1090 how many times it makes a certain transformation. Although you can do this with
1091 hand inspection, or some ad-hoc method, this is a real pain and not very useful
1092 for big programs. Using the ``Statistic`` class makes it very easy to keep
1093 track of this information, and the calculated information is presented in a
1094 uniform manner with the rest of the passes being executed.
1096 There are many examples of ``Statistic`` uses, but the basics of using it are as
1099 #. Define your statistic like this:
1103 #define DEBUG_TYPE "mypassname" // This goes before any #includes.
1104 STATISTIC(NumXForms, "The # of times I did stuff");
1106 The ``STATISTIC`` macro defines a static variable, whose name is specified by
1107 the first argument. The pass name is taken from the ``DEBUG_TYPE`` macro, and
1108 the description is taken from the second argument. The variable defined
1109 ("NumXForms" in this case) acts like an unsigned integer.
1111 #. Whenever you make a transformation, bump the counter:
1115 ++NumXForms; // I did stuff!
1117 That's all you have to do. To get '``opt``' to print out the statistics
1118 gathered, use the '``-stats``' option:
1120 .. code-block:: none
1122 $ opt -stats -mypassname < program.bc > /dev/null
1123 ... statistics output ...
1125 Note that in order to use the '``-stats``' option, LLVM must be
1126 compiled with assertions enabled.
1128 When running ``opt`` on a C file from the SPEC benchmark suite, it gives a
1129 report that looks like this:
1131 .. code-block:: none
1133 7646 bitcodewriter - Number of normal instructions
1134 725 bitcodewriter - Number of oversized instructions
1135 129996 bitcodewriter - Number of bitcode bytes written
1136 2817 raise - Number of insts DCEd or constprop'd
1137 3213 raise - Number of cast-of-self removed
1138 5046 raise - Number of expression trees converted
1139 75 raise - Number of other getelementptr's formed
1140 138 raise - Number of load/store peepholes
1141 42 deadtypeelim - Number of unused typenames removed from symtab
1142 392 funcresolve - Number of varargs functions resolved
1143 27 globaldce - Number of global variables removed
1144 2 adce - Number of basic blocks removed
1145 134 cee - Number of branches revectored
1146 49 cee - Number of setcc instruction eliminated
1147 532 gcse - Number of loads removed
1148 2919 gcse - Number of instructions removed
1149 86 indvars - Number of canonical indvars added
1150 87 indvars - Number of aux indvars removed
1151 25 instcombine - Number of dead inst eliminate
1152 434 instcombine - Number of insts combined
1153 248 licm - Number of load insts hoisted
1154 1298 licm - Number of insts hoisted to a loop pre-header
1155 3 licm - Number of insts hoisted to multiple loop preds (bad, no loop pre-header)
1156 75 mem2reg - Number of alloca's promoted
1157 1444 cfgsimplify - Number of blocks simplified
1159 Obviously, with so many optimizations, having a unified framework for this stuff
1160 is very nice. Making your pass fit well into the framework makes it more
1161 maintainable and useful.
1165 Viewing graphs while debugging code
1166 -----------------------------------
1168 Several of the important data structures in LLVM are graphs: for example CFGs
1169 made out of LLVM :ref:`BasicBlocks <BasicBlock>`, CFGs made out of LLVM
1170 :ref:`MachineBasicBlocks <MachineBasicBlock>`, and :ref:`Instruction Selection
1171 DAGs <SelectionDAG>`. In many cases, while debugging various parts of the
1172 compiler, it is nice to instantly visualize these graphs.
1174 LLVM provides several callbacks that are available in a debug build to do
1175 exactly that. If you call the ``Function::viewCFG()`` method, for example, the
1176 current LLVM tool will pop up a window containing the CFG for the function where
1177 each basic block is a node in the graph, and each node contains the instructions
1178 in the block. Similarly, there also exists ``Function::viewCFGOnly()`` (does
1179 not include the instructions), the ``MachineFunction::viewCFG()`` and
1180 ``MachineFunction::viewCFGOnly()``, and the ``SelectionDAG::viewGraph()``
1181 methods. Within GDB, for example, you can usually use something like ``call
1182 DAG.viewGraph()`` to pop up a window. Alternatively, you can sprinkle calls to
1183 these functions in your code in places you want to debug.
1185 Getting this to work requires a small amount of setup. On Unix systems
1186 with X11, install the `graphviz <http://www.graphviz.org>`_ toolkit, and make
1187 sure 'dot' and 'gv' are in your path. If you are running on Mac OS X, download
1188 and install the Mac OS X `Graphviz program
1189 <http://www.pixelglow.com/graphviz/>`_ and add
1190 ``/Applications/Graphviz.app/Contents/MacOS/`` (or wherever you install it) to
1191 your path. The programs need not be present when configuring, building or
1192 running LLVM and can simply be installed when needed during an active debug
1195 ``SelectionDAG`` has been extended to make it easier to locate *interesting*
1196 nodes in large complex graphs. From gdb, if you ``call DAG.setGraphColor(node,
1197 "color")``, then the next ``call DAG.viewGraph()`` would highlight the node in
1198 the specified color (choices of colors can be found at `colors
1199 <http://www.graphviz.org/doc/info/colors.html>`_.) More complex node attributes
1200 can be provided with ``call DAG.setGraphAttrs(node, "attributes")`` (choices can
1201 be found at `Graph attributes <http://www.graphviz.org/doc/info/attrs.html>`_.)
1202 If you want to restart and clear all the current graph attributes, then you can
1203 ``call DAG.clearGraphAttrs()``.
1205 Note that graph visualization features are compiled out of Release builds to
1206 reduce file size. This means that you need a Debug+Asserts or Release+Asserts
1207 build to use these features.
1211 Picking the Right Data Structure for a Task
1212 ===========================================
1214 LLVM has a plethora of data structures in the ``llvm/ADT/`` directory, and we
1215 commonly use STL data structures. This section describes the trade-offs you
1216 should consider when you pick one.
1218 The first step is a choose your own adventure: do you want a sequential
1219 container, a set-like container, or a map-like container? The most important
1220 thing when choosing a container is the algorithmic properties of how you plan to
1221 access the container. Based on that, you should use:
1224 * a :ref:`map-like <ds_map>` container if you need efficient look-up of a
1225 value based on another value. Map-like containers also support efficient
1226 queries for containment (whether a key is in the map). Map-like containers
1227 generally do not support efficient reverse mapping (values to keys). If you
1228 need that, use two maps. Some map-like containers also support efficient
1229 iteration through the keys in sorted order. Map-like containers are the most
1230 expensive sort, only use them if you need one of these capabilities.
1232 * a :ref:`set-like <ds_set>` container if you need to put a bunch of stuff into
1233 a container that automatically eliminates duplicates. Some set-like
1234 containers support efficient iteration through the elements in sorted order.
1235 Set-like containers are more expensive than sequential containers.
1237 * a :ref:`sequential <ds_sequential>` container provides the most efficient way
1238 to add elements and keeps track of the order they are added to the collection.
1239 They permit duplicates and support efficient iteration, but do not support
1240 efficient look-up based on a key.
1242 * a :ref:`string <ds_string>` container is a specialized sequential container or
1243 reference structure that is used for character or byte arrays.
1245 * a :ref:`bit <ds_bit>` container provides an efficient way to store and
1246 perform set operations on sets of numeric id's, while automatically
1247 eliminating duplicates. Bit containers require a maximum of 1 bit for each
1248 identifier you want to store.
1250 Once the proper category of container is determined, you can fine tune the
1251 memory use, constant factors, and cache behaviors of access by intelligently
1252 picking a member of the category. Note that constant factors and cache behavior
1253 can be a big deal. If you have a vector that usually only contains a few
1254 elements (but could contain many), for example, it's much better to use
1255 :ref:`SmallVector <dss_smallvector>` than :ref:`vector <dss_vector>`. Doing so
1256 avoids (relatively) expensive malloc/free calls, which dwarf the cost of adding
1257 the elements to the container.
1261 Sequential Containers (std::vector, std::list, etc)
1262 ---------------------------------------------------
1264 There are a variety of sequential containers available for you, based on your
1265 needs. Pick the first in this section that will do what you want.
1272 The ``llvm::ArrayRef`` class is the preferred class to use in an interface that
1273 accepts a sequential list of elements in memory and just reads from them. By
1274 taking an ``ArrayRef``, the API can be passed a fixed size array, an
1275 ``std::vector``, an ``llvm::SmallVector`` and anything else that is contiguous
1278 .. _dss_fixedarrays:
1283 Fixed size arrays are very simple and very fast. They are good if you know
1284 exactly how many elements you have, or you have a (low) upper bound on how many
1289 Heap Allocated Arrays
1290 ^^^^^^^^^^^^^^^^^^^^^
1292 Heap allocated arrays (``new[]`` + ``delete[]``) are also simple. They are good
1293 if the number of elements is variable, if you know how many elements you will
1294 need before the array is allocated, and if the array is usually large (if not,
1295 consider a :ref:`SmallVector <dss_smallvector>`). The cost of a heap allocated
1296 array is the cost of the new/delete (aka malloc/free). Also note that if you
1297 are allocating an array of a type with a constructor, the constructor and
1298 destructors will be run for every element in the array (re-sizable vectors only
1299 construct those elements actually used).
1301 .. _dss_tinyptrvector:
1303 llvm/ADT/TinyPtrVector.h
1304 ^^^^^^^^^^^^^^^^^^^^^^^^
1306 ``TinyPtrVector<Type>`` is a highly specialized collection class that is
1307 optimized to avoid allocation in the case when a vector has zero or one
1308 elements. It has two major restrictions: 1) it can only hold values of pointer
1309 type, and 2) it cannot hold a null pointer.
1311 Since this container is highly specialized, it is rarely used.
1313 .. _dss_smallvector:
1315 llvm/ADT/SmallVector.h
1316 ^^^^^^^^^^^^^^^^^^^^^^
1318 ``SmallVector<Type, N>`` is a simple class that looks and smells just like
1319 ``vector<Type>``: it supports efficient iteration, lays out elements in memory
1320 order (so you can do pointer arithmetic between elements), supports efficient
1321 push_back/pop_back operations, supports efficient random access to its elements,
1324 The advantage of SmallVector is that it allocates space for some number of
1325 elements (N) **in the object itself**. Because of this, if the SmallVector is
1326 dynamically smaller than N, no malloc is performed. This can be a big win in
1327 cases where the malloc/free call is far more expensive than the code that
1328 fiddles around with the elements.
1330 This is good for vectors that are "usually small" (e.g. the number of
1331 predecessors/successors of a block is usually less than 8). On the other hand,
1332 this makes the size of the SmallVector itself large, so you don't want to
1333 allocate lots of them (doing so will waste a lot of space). As such,
1334 SmallVectors are most useful when on the stack.
1336 SmallVector also provides a nice portable and efficient replacement for
1341 Prefer to use ``SmallVectorImpl<T>`` as a parameter type.
1343 In APIs that don't care about the "small size" (most?), prefer to use
1344 the ``SmallVectorImpl<T>`` class, which is basically just the "vector
1345 header" (and methods) without the elements allocated after it. Note that
1346 ``SmallVector<T, N>`` inherits from ``SmallVectorImpl<T>`` so the
1347 conversion is implicit and costs nothing. E.g.
1351 // BAD: Clients cannot pass e.g. SmallVector<Foo, 4>.
1352 hardcodedSmallSize(SmallVector<Foo, 2> &Out);
1353 // GOOD: Clients can pass any SmallVector<Foo, N>.
1354 allowsAnySmallSize(SmallVectorImpl<Foo> &Out);
1357 SmallVector<Foo, 8> Vec;
1358 hardcodedSmallSize(Vec); // Error.
1359 allowsAnySmallSize(Vec); // Works.
1362 Even though it has "``Impl``" in the name, this is so widely used that
1363 it really isn't "private to the implementation" anymore. A name like
1364 ``SmallVectorHeader`` would be more appropriate.
1371 ``std::vector`` is well loved and respected. It is useful when SmallVector
1372 isn't: when the size of the vector is often large (thus the small optimization
1373 will rarely be a benefit) or if you will be allocating many instances of the
1374 vector itself (which would waste space for elements that aren't in the
1375 container). vector is also useful when interfacing with code that expects
1378 One worthwhile note about std::vector: avoid code like this:
1387 Instead, write this as:
1397 Doing so will save (at least) one heap allocation and free per iteration of the
1405 ``std::deque`` is, in some senses, a generalized version of ``std::vector``.
1406 Like ``std::vector``, it provides constant time random access and other similar
1407 properties, but it also provides efficient access to the front of the list. It
1408 does not guarantee continuity of elements within memory.
1410 In exchange for this extra flexibility, ``std::deque`` has significantly higher
1411 constant factor costs than ``std::vector``. If possible, use ``std::vector`` or
1419 ``std::list`` is an extremely inefficient class that is rarely useful. It
1420 performs a heap allocation for every element inserted into it, thus having an
1421 extremely high constant factor, particularly for small data types.
1422 ``std::list`` also only supports bidirectional iteration, not random access
1425 In exchange for this high cost, std::list supports efficient access to both ends
1426 of the list (like ``std::deque``, but unlike ``std::vector`` or
1427 ``SmallVector``). In addition, the iterator invalidation characteristics of
1428 std::list are stronger than that of a vector class: inserting or removing an
1429 element into the list does not invalidate iterator or pointers to other elements
1437 ``ilist<T>`` implements an 'intrusive' doubly-linked list. It is intrusive,
1438 because it requires the element to store and provide access to the prev/next
1439 pointers for the list.
1441 ``ilist`` has the same drawbacks as ``std::list``, and additionally requires an
1442 ``ilist_traits`` implementation for the element type, but it provides some novel
1443 characteristics. In particular, it can efficiently store polymorphic objects,
1444 the traits class is informed when an element is inserted or removed from the
1445 list, and ``ilist``\ s are guaranteed to support a constant-time splice
1448 These properties are exactly what we want for things like ``Instruction``\ s and
1449 basic blocks, which is why these are implemented with ``ilist``\ s.
1451 Related classes of interest are explained in the following subsections:
1453 * :ref:`ilist_traits <dss_ilist_traits>`
1455 * :ref:`iplist <dss_iplist>`
1457 * :ref:`llvm/ADT/ilist_node.h <dss_ilist_node>`
1459 * :ref:`Sentinels <dss_ilist_sentinel>`
1461 .. _dss_packedvector:
1463 llvm/ADT/PackedVector.h
1464 ^^^^^^^^^^^^^^^^^^^^^^^
1466 Useful for storing a vector of values using only a few number of bits for each
1467 value. Apart from the standard operations of a vector-like container, it can
1468 also perform an 'or' set operation.
1476 FirstCondition = 0x1,
1477 SecondCondition = 0x2,
1482 PackedVector<State, 2> Vec1;
1483 Vec1.push_back(FirstCondition);
1485 PackedVector<State, 2> Vec2;
1486 Vec2.push_back(SecondCondition);
1489 return Vec1[0]; // returns 'Both'.
1492 .. _dss_ilist_traits:
1497 ``ilist_traits<T>`` is ``ilist<T>``'s customization mechanism. ``iplist<T>``
1498 (and consequently ``ilist<T>``) publicly derive from this traits class.
1505 ``iplist<T>`` is ``ilist<T>``'s base and as such supports a slightly narrower
1506 interface. Notably, inserters from ``T&`` are absent.
1508 ``ilist_traits<T>`` is a public base of this class and can be used for a wide
1509 variety of customizations.
1513 llvm/ADT/ilist_node.h
1514 ^^^^^^^^^^^^^^^^^^^^^
1516 ``ilist_node<T>`` implements the forward and backward links that are expected
1517 by the ``ilist<T>`` (and analogous containers) in the default manner.
1519 ``ilist_node<T>``\ s are meant to be embedded in the node type ``T``, usually
1520 ``T`` publicly derives from ``ilist_node<T>``.
1522 .. _dss_ilist_sentinel:
1527 ``ilist``\ s have another specialty that must be considered. To be a good
1528 citizen in the C++ ecosystem, it needs to support the standard container
1529 operations, such as ``begin`` and ``end`` iterators, etc. Also, the
1530 ``operator--`` must work correctly on the ``end`` iterator in the case of
1531 non-empty ``ilist``\ s.
1533 The only sensible solution to this problem is to allocate a so-called *sentinel*
1534 along with the intrusive list, which serves as the ``end`` iterator, providing
1535 the back-link to the last element. However conforming to the C++ convention it
1536 is illegal to ``operator++`` beyond the sentinel and it also must not be
1539 These constraints allow for some implementation freedom to the ``ilist`` how to
1540 allocate and store the sentinel. The corresponding policy is dictated by
1541 ``ilist_traits<T>``. By default a ``T`` gets heap-allocated whenever the need
1542 for a sentinel arises.
1544 While the default policy is sufficient in most cases, it may break down when
1545 ``T`` does not provide a default constructor. Also, in the case of many
1546 instances of ``ilist``\ s, the memory overhead of the associated sentinels is
1547 wasted. To alleviate the situation with numerous and voluminous
1548 ``T``-sentinels, sometimes a trick is employed, leading to *ghostly sentinels*.
1550 Ghostly sentinels are obtained by specially-crafted ``ilist_traits<T>`` which
1551 superpose the sentinel with the ``ilist`` instance in memory. Pointer
1552 arithmetic is used to obtain the sentinel, which is relative to the ``ilist``'s
1553 ``this`` pointer. The ``ilist`` is augmented by an extra pointer, which serves
1554 as the back-link of the sentinel. This is the only field in the ghostly
1555 sentinel which can be legally accessed.
1559 Other Sequential Container options
1560 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
1562 Other STL containers are available, such as ``std::string``.
1564 There are also various STL adapter classes such as ``std::queue``,
1565 ``std::priority_queue``, ``std::stack``, etc. These provide simplified access
1566 to an underlying container but don't affect the cost of the container itself.
1570 String-like containers
1571 ----------------------
1573 There are a variety of ways to pass around and use strings in C and C++, and
1574 LLVM adds a few new options to choose from. Pick the first option on this list
1575 that will do what you need, they are ordered according to their relative cost.
1577 Note that it is generally preferred to *not* pass strings around as ``const
1578 char*``'s. These have a number of problems, including the fact that they
1579 cannot represent embedded nul ("\0") characters, and do not have a length
1580 available efficiently. The general replacement for '``const char*``' is
1583 For more information on choosing string containers for APIs, please see
1584 :ref:`Passing Strings <string_apis>`.
1588 llvm/ADT/StringRef.h
1589 ^^^^^^^^^^^^^^^^^^^^
1591 The StringRef class is a simple value class that contains a pointer to a
1592 character and a length, and is quite related to the :ref:`ArrayRef
1593 <dss_arrayref>` class (but specialized for arrays of characters). Because
1594 StringRef carries a length with it, it safely handles strings with embedded nul
1595 characters in it, getting the length does not require a strlen call, and it even
1596 has very convenient APIs for slicing and dicing the character range that it
1599 StringRef is ideal for passing simple strings around that are known to be live,
1600 either because they are C string literals, std::string, a C array, or a
1601 SmallVector. Each of these cases has an efficient implicit conversion to
1602 StringRef, which doesn't result in a dynamic strlen being executed.
1604 StringRef has a few major limitations which make more powerful string containers
1607 #. You cannot directly convert a StringRef to a 'const char*' because there is
1608 no way to add a trailing nul (unlike the .c_str() method on various stronger
1611 #. StringRef doesn't own or keep alive the underlying string bytes.
1612 As such it can easily lead to dangling pointers, and is not suitable for
1613 embedding in datastructures in most cases (instead, use an std::string or
1614 something like that).
1616 #. For the same reason, StringRef cannot be used as the return value of a
1617 method if the method "computes" the result string. Instead, use std::string.
1619 #. StringRef's do not allow you to mutate the pointed-to string bytes and it
1620 doesn't allow you to insert or remove bytes from the range. For editing
1621 operations like this, it interoperates with the :ref:`Twine <dss_twine>`
1624 Because of its strengths and limitations, it is very common for a function to
1625 take a StringRef and for a method on an object to return a StringRef that points
1626 into some string that it owns.
1633 The Twine class is used as an intermediary datatype for APIs that want to take a
1634 string that can be constructed inline with a series of concatenations. Twine
1635 works by forming recursive instances of the Twine datatype (a simple value
1636 object) on the stack as temporary objects, linking them together into a tree
1637 which is then linearized when the Twine is consumed. Twine is only safe to use
1638 as the argument to a function, and should always be a const reference, e.g.:
1642 void foo(const Twine &T);
1646 foo(X + "." + Twine(i));
1648 This example forms a string like "blarg.42" by concatenating the values
1649 together, and does not form intermediate strings containing "blarg" or "blarg.".
1651 Because Twine is constructed with temporary objects on the stack, and because
1652 these instances are destroyed at the end of the current statement, it is an
1653 inherently dangerous API. For example, this simple variant contains undefined
1654 behavior and will probably crash:
1658 void foo(const Twine &T);
1662 const Twine &Tmp = X + "." + Twine(i);
1665 ... because the temporaries are destroyed before the call. That said, Twine's
1666 are much more efficient than intermediate std::string temporaries, and they work
1667 really well with StringRef. Just be aware of their limitations.
1669 .. _dss_smallstring:
1671 llvm/ADT/SmallString.h
1672 ^^^^^^^^^^^^^^^^^^^^^^
1674 SmallString is a subclass of :ref:`SmallVector <dss_smallvector>` that adds some
1675 convenience APIs like += that takes StringRef's. SmallString avoids allocating
1676 memory in the case when the preallocated space is enough to hold its data, and
1677 it calls back to general heap allocation when required. Since it owns its data,
1678 it is very safe to use and supports full mutation of the string.
1680 Like SmallVector's, the big downside to SmallString is their sizeof. While they
1681 are optimized for small strings, they themselves are not particularly small.
1682 This means that they work great for temporary scratch buffers on the stack, but
1683 should not generally be put into the heap: it is very rare to see a SmallString
1684 as the member of a frequently-allocated heap data structure or returned
1692 The standard C++ std::string class is a very general class that (like
1693 SmallString) owns its underlying data. sizeof(std::string) is very reasonable
1694 so it can be embedded into heap data structures and returned by-value. On the
1695 other hand, std::string is highly inefficient for inline editing (e.g.
1696 concatenating a bunch of stuff together) and because it is provided by the
1697 standard library, its performance characteristics depend a lot of the host
1698 standard library (e.g. libc++ and MSVC provide a highly optimized string class,
1699 GCC contains a really slow implementation).
1701 The major disadvantage of std::string is that almost every operation that makes
1702 them larger can allocate memory, which is slow. As such, it is better to use
1703 SmallVector or Twine as a scratch buffer, but then use std::string to persist
1708 Set-Like Containers (std::set, SmallSet, SetVector, etc)
1709 --------------------------------------------------------
1711 Set-like containers are useful when you need to canonicalize multiple values
1712 into a single representation. There are several different choices for how to do
1713 this, providing various trade-offs.
1715 .. _dss_sortedvectorset:
1720 If you intend to insert a lot of elements, then do a lot of queries, a great
1721 approach is to use a vector (or other sequential container) with
1722 std::sort+std::unique to remove duplicates. This approach works really well if
1723 your usage pattern has these two distinct phases (insert then query), and can be
1724 coupled with a good choice of :ref:`sequential container <ds_sequential>`.
1726 This combination provides the several nice properties: the result data is
1727 contiguous in memory (good for cache locality), has few allocations, is easy to
1728 address (iterators in the final vector are just indices or pointers), and can be
1729 efficiently queried with a standard binary search (e.g.
1730 ``std::lower_bound``; if you want the whole range of elements comparing
1731 equal, use ``std::equal_range``).
1738 If you have a set-like data structure that is usually small and whose elements
1739 are reasonably small, a ``SmallSet<Type, N>`` is a good choice. This set has
1740 space for N elements in place (thus, if the set is dynamically smaller than N,
1741 no malloc traffic is required) and accesses them with a simple linear search.
1742 When the set grows beyond N elements, it allocates a more expensive
1743 representation that guarantees efficient access (for most types, it falls back
1744 to :ref:`std::set <dss_set>`, but for pointers it uses something far better,
1745 :ref:`SmallPtrSet <dss_smallptrset>`.
1747 The magic of this class is that it handles small sets extremely efficiently, but
1748 gracefully handles extremely large sets without loss of efficiency. The
1749 drawback is that the interface is quite small: it supports insertion, queries
1750 and erasing, but does not support iteration.
1752 .. _dss_smallptrset:
1754 llvm/ADT/SmallPtrSet.h
1755 ^^^^^^^^^^^^^^^^^^^^^^
1757 ``SmallPtrSet`` has all the advantages of ``SmallSet`` (and a ``SmallSet`` of
1758 pointers is transparently implemented with a ``SmallPtrSet``), but also supports
1759 iterators. If more than N insertions are performed, a single quadratically
1760 probed hash table is allocated and grows as needed, providing extremely
1761 efficient access (constant time insertion/deleting/queries with low constant
1762 factors) and is very stingy with malloc traffic.
1764 Note that, unlike :ref:`std::set <dss_set>`, the iterators of ``SmallPtrSet``
1765 are invalidated whenever an insertion occurs. Also, the values visited by the
1766 iterators are not visited in sorted order.
1770 llvm/ADT/StringSet.h
1771 ^^^^^^^^^^^^^^^^^^^^
1773 ``StringSet`` is a thin wrapper around :ref:`StringMap\<char\> <dss_stringmap>`,
1774 and it allows efficient storage and retrieval of unique strings.
1776 Functionally analogous to ``SmallSet<StringRef>``, ``StringSet`` also supports
1777 iteration. (The iterator dereferences to a ``StringMapEntry<char>``, so you
1778 need to call ``i->getKey()`` to access the item of the StringSet.) On the
1779 other hand, ``StringSet`` doesn't support range-insertion and
1780 copy-construction, which :ref:`SmallSet <dss_smallset>` and :ref:`SmallPtrSet
1781 <dss_smallptrset>` do support.
1788 DenseSet is a simple quadratically probed hash table. It excels at supporting
1789 small values: it uses a single allocation to hold all of the pairs that are
1790 currently inserted in the set. DenseSet is a great way to unique small values
1791 that are not simple pointers (use :ref:`SmallPtrSet <dss_smallptrset>` for
1792 pointers). Note that DenseSet has the same requirements for the value type that
1793 :ref:`DenseMap <dss_densemap>` has.
1797 llvm/ADT/SparseSet.h
1798 ^^^^^^^^^^^^^^^^^^^^
1800 SparseSet holds a small number of objects identified by unsigned keys of
1801 moderate size. It uses a lot of memory, but provides operations that are almost
1802 as fast as a vector. Typical keys are physical registers, virtual registers, or
1803 numbered basic blocks.
1805 SparseSet is useful for algorithms that need very fast clear/find/insert/erase
1806 and fast iteration over small sets. It is not intended for building composite
1809 .. _dss_sparsemultiset:
1811 llvm/ADT/SparseMultiSet.h
1812 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
1814 SparseMultiSet adds multiset behavior to SparseSet, while retaining SparseSet's
1815 desirable attributes. Like SparseSet, it typically uses a lot of memory, but
1816 provides operations that are almost as fast as a vector. Typical keys are
1817 physical registers, virtual registers, or numbered basic blocks.
1819 SparseMultiSet is useful for algorithms that need very fast
1820 clear/find/insert/erase of the entire collection, and iteration over sets of
1821 elements sharing a key. It is often a more efficient choice than using composite
1822 data structures (e.g. vector-of-vectors, map-of-vectors). It is not intended for
1823 building composite data structures.
1827 llvm/ADT/FoldingSet.h
1828 ^^^^^^^^^^^^^^^^^^^^^
1830 FoldingSet is an aggregate class that is really good at uniquing
1831 expensive-to-create or polymorphic objects. It is a combination of a chained
1832 hash table with intrusive links (uniqued objects are required to inherit from
1833 FoldingSetNode) that uses :ref:`SmallVector <dss_smallvector>` as part of its ID
1836 Consider a case where you want to implement a "getOrCreateFoo" method for a
1837 complex object (for example, a node in the code generator). The client has a
1838 description of **what** it wants to generate (it knows the opcode and all the
1839 operands), but we don't want to 'new' a node, then try inserting it into a set
1840 only to find out it already exists, at which point we would have to delete it
1841 and return the node that already exists.
1843 To support this style of client, FoldingSet perform a query with a
1844 FoldingSetNodeID (which wraps SmallVector) that can be used to describe the
1845 element that we want to query for. The query either returns the element
1846 matching the ID or it returns an opaque ID that indicates where insertion should
1847 take place. Construction of the ID usually does not require heap traffic.
1849 Because FoldingSet uses intrusive links, it can support polymorphic objects in
1850 the set (for example, you can have SDNode instances mixed with LoadSDNodes).
1851 Because the elements are individually allocated, pointers to the elements are
1852 stable: inserting or removing elements does not invalidate any pointers to other
1860 ``std::set`` is a reasonable all-around set class, which is decent at many
1861 things but great at nothing. std::set allocates memory for each element
1862 inserted (thus it is very malloc intensive) and typically stores three pointers
1863 per element in the set (thus adding a large amount of per-element space
1864 overhead). It offers guaranteed log(n) performance, which is not particularly
1865 fast from a complexity standpoint (particularly if the elements of the set are
1866 expensive to compare, like strings), and has extremely high constant factors for
1867 lookup, insertion and removal.
1869 The advantages of std::set are that its iterators are stable (deleting or
1870 inserting an element from the set does not affect iterators or pointers to other
1871 elements) and that iteration over the set is guaranteed to be in sorted order.
1872 If the elements in the set are large, then the relative overhead of the pointers
1873 and malloc traffic is not a big deal, but if the elements of the set are small,
1874 std::set is almost never a good choice.
1878 llvm/ADT/SetVector.h
1879 ^^^^^^^^^^^^^^^^^^^^
1881 LLVM's ``SetVector<Type>`` is an adapter class that combines your choice of a
1882 set-like container along with a :ref:`Sequential Container <ds_sequential>` The
1883 important property that this provides is efficient insertion with uniquing
1884 (duplicate elements are ignored) with iteration support. It implements this by
1885 inserting elements into both a set-like container and the sequential container,
1886 using the set-like container for uniquing and the sequential container for
1889 The difference between SetVector and other sets is that the order of iteration
1890 is guaranteed to match the order of insertion into the SetVector. This property
1891 is really important for things like sets of pointers. Because pointer values
1892 are non-deterministic (e.g. vary across runs of the program on different
1893 machines), iterating over the pointers in the set will not be in a well-defined
1896 The drawback of SetVector is that it requires twice as much space as a normal
1897 set and has the sum of constant factors from the set-like container and the
1898 sequential container that it uses. Use it **only** if you need to iterate over
1899 the elements in a deterministic order. SetVector is also expensive to delete
1900 elements out of (linear time), unless you use its "pop_back" method, which is
1903 ``SetVector`` is an adapter class that defaults to using ``std::vector`` and a
1904 size 16 ``SmallSet`` for the underlying containers, so it is quite expensive.
1905 However, ``"llvm/ADT/SetVector.h"`` also provides a ``SmallSetVector`` class,
1906 which defaults to using a ``SmallVector`` and ``SmallSet`` of a specified size.
1907 If you use this, and if your sets are dynamically smaller than ``N``, you will
1908 save a lot of heap traffic.
1910 .. _dss_uniquevector:
1912 llvm/ADT/UniqueVector.h
1913 ^^^^^^^^^^^^^^^^^^^^^^^
1915 UniqueVector is similar to :ref:`SetVector <dss_setvector>` but it retains a
1916 unique ID for each element inserted into the set. It internally contains a map
1917 and a vector, and it assigns a unique ID for each value inserted into the set.
1919 UniqueVector is very expensive: its cost is the sum of the cost of maintaining
1920 both the map and vector, it has high complexity, high constant factors, and
1921 produces a lot of malloc traffic. It should be avoided.
1923 .. _dss_immutableset:
1925 llvm/ADT/ImmutableSet.h
1926 ^^^^^^^^^^^^^^^^^^^^^^^
1928 ImmutableSet is an immutable (functional) set implementation based on an AVL
1929 tree. Adding or removing elements is done through a Factory object and results
1930 in the creation of a new ImmutableSet object. If an ImmutableSet already exists
1931 with the given contents, then the existing one is returned; equality is compared
1932 with a FoldingSetNodeID. The time and space complexity of add or remove
1933 operations is logarithmic in the size of the original set.
1935 There is no method for returning an element of the set, you can only check for
1940 Other Set-Like Container Options
1941 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
1943 The STL provides several other options, such as std::multiset and the various
1944 "hash_set" like containers (whether from C++ TR1 or from the SGI library). We
1945 never use hash_set and unordered_set because they are generally very expensive
1946 (each insertion requires a malloc) and very non-portable.
1948 std::multiset is useful if you're not interested in elimination of duplicates,
1949 but has all the drawbacks of :ref:`std::set <dss_set>`. A sorted vector
1950 (where you don't delete duplicate entries) or some other approach is almost
1955 Map-Like Containers (std::map, DenseMap, etc)
1956 ---------------------------------------------
1958 Map-like containers are useful when you want to associate data to a key. As
1959 usual, there are a lot of different ways to do this. :)
1961 .. _dss_sortedvectormap:
1966 If your usage pattern follows a strict insert-then-query approach, you can
1967 trivially use the same approach as :ref:`sorted vectors for set-like containers
1968 <dss_sortedvectorset>`. The only difference is that your query function (which
1969 uses std::lower_bound to get efficient log(n) lookup) should only compare the
1970 key, not both the key and value. This yields the same advantages as sorted
1975 llvm/ADT/StringMap.h
1976 ^^^^^^^^^^^^^^^^^^^^
1978 Strings are commonly used as keys in maps, and they are difficult to support
1979 efficiently: they are variable length, inefficient to hash and compare when
1980 long, expensive to copy, etc. StringMap is a specialized container designed to
1981 cope with these issues. It supports mapping an arbitrary range of bytes to an
1982 arbitrary other object.
1984 The StringMap implementation uses a quadratically-probed hash table, where the
1985 buckets store a pointer to the heap allocated entries (and some other stuff).
1986 The entries in the map must be heap allocated because the strings are variable
1987 length. The string data (key) and the element object (value) are stored in the
1988 same allocation with the string data immediately after the element object.
1989 This container guarantees the "``(char*)(&Value+1)``" points to the key string
1992 The StringMap is very fast for several reasons: quadratic probing is very cache
1993 efficient for lookups, the hash value of strings in buckets is not recomputed
1994 when looking up an element, StringMap rarely has to touch the memory for
1995 unrelated objects when looking up a value (even when hash collisions happen),
1996 hash table growth does not recompute the hash values for strings already in the
1997 table, and each pair in the map is store in a single allocation (the string data
1998 is stored in the same allocation as the Value of a pair).
2000 StringMap also provides query methods that take byte ranges, so it only ever
2001 copies a string if a value is inserted into the table.
2003 StringMap iteratation order, however, is not guaranteed to be deterministic, so
2004 any uses which require that should instead use a std::map.
2008 llvm/ADT/IndexedMap.h
2009 ^^^^^^^^^^^^^^^^^^^^^
2011 IndexedMap is a specialized container for mapping small dense integers (or
2012 values that can be mapped to small dense integers) to some other type. It is
2013 internally implemented as a vector with a mapping function that maps the keys
2014 to the dense integer range.
2016 This is useful for cases like virtual registers in the LLVM code generator: they
2017 have a dense mapping that is offset by a compile-time constant (the first
2018 virtual register ID).
2025 DenseMap is a simple quadratically probed hash table. It excels at supporting
2026 small keys and values: it uses a single allocation to hold all of the pairs
2027 that are currently inserted in the map. DenseMap is a great way to map
2028 pointers to pointers, or map other small types to each other.
2030 There are several aspects of DenseMap that you should be aware of, however.
2031 The iterators in a DenseMap are invalidated whenever an insertion occurs,
2032 unlike map. Also, because DenseMap allocates space for a large number of
2033 key/value pairs (it starts with 64 by default), it will waste a lot of space if
2034 your keys or values are large. Finally, you must implement a partial
2035 specialization of DenseMapInfo for the key that you want, if it isn't already
2036 supported. This is required to tell DenseMap about two special marker values
2037 (which can never be inserted into the map) that it needs internally.
2039 DenseMap's find_as() method supports lookup operations using an alternate key
2040 type. This is useful in cases where the normal key type is expensive to
2041 construct, but cheap to compare against. The DenseMapInfo is responsible for
2042 defining the appropriate comparison and hashing methods for each alternate key
2050 ValueMap is a wrapper around a :ref:`DenseMap <dss_densemap>` mapping
2051 ``Value*``\ s (or subclasses) to another type. When a Value is deleted or
2052 RAUW'ed, ValueMap will update itself so the new version of the key is mapped to
2053 the same value, just as if the key were a WeakVH. You can configure exactly how
2054 this happens, and what else happens on these two events, by passing a ``Config``
2055 parameter to the ValueMap template.
2057 .. _dss_intervalmap:
2059 llvm/ADT/IntervalMap.h
2060 ^^^^^^^^^^^^^^^^^^^^^^
2062 IntervalMap is a compact map for small keys and values. It maps key intervals
2063 instead of single keys, and it will automatically coalesce adjacent intervals.
2064 When the map only contains a few intervals, they are stored in the map object
2065 itself to avoid allocations.
2067 The IntervalMap iterators are quite big, so they should not be passed around as
2068 STL iterators. The heavyweight iterators allow a smaller data structure.
2075 std::map has similar characteristics to :ref:`std::set <dss_set>`: it uses a
2076 single allocation per pair inserted into the map, it offers log(n) lookup with
2077 an extremely large constant factor, imposes a space penalty of 3 pointers per
2078 pair in the map, etc.
2080 std::map is most useful when your keys or values are very large, if you need to
2081 iterate over the collection in sorted order, or if you need stable iterators
2082 into the map (i.e. they don't get invalidated if an insertion or deletion of
2083 another element takes place).
2087 llvm/ADT/MapVector.h
2088 ^^^^^^^^^^^^^^^^^^^^
2090 ``MapVector<KeyT,ValueT>`` provides a subset of the DenseMap interface. The
2091 main difference is that the iteration order is guaranteed to be the insertion
2092 order, making it an easy (but somewhat expensive) solution for non-deterministic
2093 iteration over maps of pointers.
2095 It is implemented by mapping from key to an index in a vector of key,value
2096 pairs. This provides fast lookup and iteration, but has two main drawbacks:
2097 the key is stored twice and removing elements takes linear time. If it is
2098 necessary to remove elements, it's best to remove them in bulk using
2101 .. _dss_inteqclasses:
2103 llvm/ADT/IntEqClasses.h
2104 ^^^^^^^^^^^^^^^^^^^^^^^
2106 IntEqClasses provides a compact representation of equivalence classes of small
2107 integers. Initially, each integer in the range 0..n-1 has its own equivalence
2108 class. Classes can be joined by passing two class representatives to the
2109 join(a, b) method. Two integers are in the same class when findLeader() returns
2110 the same representative.
2112 Once all equivalence classes are formed, the map can be compressed so each
2113 integer 0..n-1 maps to an equivalence class number in the range 0..m-1, where m
2114 is the total number of equivalence classes. The map must be uncompressed before
2115 it can be edited again.
2117 .. _dss_immutablemap:
2119 llvm/ADT/ImmutableMap.h
2120 ^^^^^^^^^^^^^^^^^^^^^^^
2122 ImmutableMap is an immutable (functional) map implementation based on an AVL
2123 tree. Adding or removing elements is done through a Factory object and results
2124 in the creation of a new ImmutableMap object. If an ImmutableMap already exists
2125 with the given key set, then the existing one is returned; equality is compared
2126 with a FoldingSetNodeID. The time and space complexity of add or remove
2127 operations is logarithmic in the size of the original map.
2131 Other Map-Like Container Options
2132 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2134 The STL provides several other options, such as std::multimap and the various
2135 "hash_map" like containers (whether from C++ TR1 or from the SGI library). We
2136 never use hash_set and unordered_set because they are generally very expensive
2137 (each insertion requires a malloc) and very non-portable.
2139 std::multimap is useful if you want to map a key to multiple values, but has all
2140 the drawbacks of std::map. A sorted vector or some other approach is almost
2145 Bit storage containers (BitVector, SparseBitVector)
2146 ---------------------------------------------------
2148 Unlike the other containers, there are only two bit storage containers, and
2149 choosing when to use each is relatively straightforward.
2151 One additional option is ``std::vector<bool>``: we discourage its use for two
2152 reasons 1) the implementation in many common compilers (e.g. commonly
2153 available versions of GCC) is extremely inefficient and 2) the C++ standards
2154 committee is likely to deprecate this container and/or change it significantly
2155 somehow. In any case, please don't use it.
2162 The BitVector container provides a dynamic size set of bits for manipulation.
2163 It supports individual bit setting/testing, as well as set operations. The set
2164 operations take time O(size of bitvector), but operations are performed one word
2165 at a time, instead of one bit at a time. This makes the BitVector very fast for
2166 set operations compared to other containers. Use the BitVector when you expect
2167 the number of set bits to be high (i.e. a dense set).
2169 .. _dss_smallbitvector:
2174 The SmallBitVector container provides the same interface as BitVector, but it is
2175 optimized for the case where only a small number of bits, less than 25 or so,
2176 are needed. It also transparently supports larger bit counts, but slightly less
2177 efficiently than a plain BitVector, so SmallBitVector should only be used when
2178 larger counts are rare.
2180 At this time, SmallBitVector does not support set operations (and, or, xor), and
2181 its operator[] does not provide an assignable lvalue.
2183 .. _dss_sparsebitvector:
2188 The SparseBitVector container is much like BitVector, with one major difference:
2189 Only the bits that are set, are stored. This makes the SparseBitVector much
2190 more space efficient than BitVector when the set is sparse, as well as making
2191 set operations O(number of set bits) instead of O(size of universe). The
2192 downside to the SparseBitVector is that setting and testing of random bits is
2193 O(N), and on large SparseBitVectors, this can be slower than BitVector. In our
2194 implementation, setting or testing bits in sorted order (either forwards or
2195 reverse) is O(1) worst case. Testing and setting bits within 128 bits (depends
2196 on size) of the current bit is also O(1). As a general statement,
2197 testing/setting bits in a SparseBitVector is O(distance away from last set bit).
2201 Helpful Hints for Common Operations
2202 ===================================
2204 This section describes how to perform some very simple transformations of LLVM
2205 code. This is meant to give examples of common idioms used, showing the
2206 practical side of LLVM transformations.
2208 Because this is a "how-to" section, you should also read about the main classes
2209 that you will be working with. The :ref:`Core LLVM Class Hierarchy Reference
2210 <coreclasses>` contains details and descriptions of the main classes that you
2215 Basic Inspection and Traversal Routines
2216 ---------------------------------------
2218 The LLVM compiler infrastructure have many different data structures that may be
2219 traversed. Following the example of the C++ standard template library, the
2220 techniques used to traverse these various data structures are all basically the
2221 same. For a enumerable sequence of values, the ``XXXbegin()`` function (or
2222 method) returns an iterator to the start of the sequence, the ``XXXend()``
2223 function returns an iterator pointing to one past the last valid element of the
2224 sequence, and there is some ``XXXiterator`` data type that is common between the
2227 Because the pattern for iteration is common across many different aspects of the
2228 program representation, the standard template library algorithms may be used on
2229 them, and it is easier to remember how to iterate. First we show a few common
2230 examples of the data structures that need to be traversed. Other data
2231 structures are traversed in very similar ways.
2233 .. _iterate_function:
2235 Iterating over the ``BasicBlock`` in a ``Function``
2236 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2238 It's quite common to have a ``Function`` instance that you'd like to transform
2239 in some way; in particular, you'd like to manipulate its ``BasicBlock``\ s. To
2240 facilitate this, you'll need to iterate over all of the ``BasicBlock``\ s that
2241 constitute the ``Function``. The following is an example that prints the name
2242 of a ``BasicBlock`` and the number of ``Instruction``\ s it contains:
2246 // func is a pointer to a Function instance
2247 for (Function::iterator i = func->begin(), e = func->end(); i != e; ++i)
2248 // Print out the name of the basic block if it has one, and then the
2249 // number of instructions that it contains
2250 errs() << "Basic block (name=" << i->getName() << ") has "
2251 << i->size() << " instructions.\n";
2253 Note that i can be used as if it were a pointer for the purposes of invoking
2254 member functions of the ``Instruction`` class. This is because the indirection
2255 operator is overloaded for the iterator classes. In the above code, the
2256 expression ``i->size()`` is exactly equivalent to ``(*i).size()`` just like
2259 .. _iterate_basicblock:
2261 Iterating over the ``Instruction`` in a ``BasicBlock``
2262 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2264 Just like when dealing with ``BasicBlock``\ s in ``Function``\ s, it's easy to
2265 iterate over the individual instructions that make up ``BasicBlock``\ s. Here's
2266 a code snippet that prints out each instruction in a ``BasicBlock``:
2270 // blk is a pointer to a BasicBlock instance
2271 for (BasicBlock::iterator i = blk->begin(), e = blk->end(); i != e; ++i)
2272 // The next statement works since operator<<(ostream&,...)
2273 // is overloaded for Instruction&
2274 errs() << *i << "\n";
2277 However, this isn't really the best way to print out the contents of a
2278 ``BasicBlock``! Since the ostream operators are overloaded for virtually
2279 anything you'll care about, you could have just invoked the print routine on the
2280 basic block itself: ``errs() << *blk << "\n";``.
2282 .. _iterate_insiter:
2284 Iterating over the ``Instruction`` in a ``Function``
2285 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2287 If you're finding that you commonly iterate over a ``Function``'s
2288 ``BasicBlock``\ s and then that ``BasicBlock``'s ``Instruction``\ s,
2289 ``InstIterator`` should be used instead. You'll need to include
2290 ``llvm/IR/InstIterator.h`` (`doxygen
2291 <http://llvm.org/doxygen/InstIterator_8h.html>`__) and then instantiate
2292 ``InstIterator``\ s explicitly in your code. Here's a small example that shows
2293 how to dump all instructions in a function to the standard error stream:
2297 #include "llvm/IR/InstIterator.h"
2299 // F is a pointer to a Function instance
2300 for (inst_iterator I = inst_begin(F), E = inst_end(F); I != E; ++I)
2301 errs() << *I << "\n";
2303 Easy, isn't it? You can also use ``InstIterator``\ s to fill a work list with
2304 its initial contents. For example, if you wanted to initialize a work list to
2305 contain all instructions in a ``Function`` F, all you would need to do is
2310 std::set<Instruction*> worklist;
2311 // or better yet, SmallPtrSet<Instruction*, 64> worklist;
2313 for (inst_iterator I = inst_begin(F), E = inst_end(F); I != E; ++I)
2314 worklist.insert(&*I);
2316 The STL set ``worklist`` would now contain all instructions in the ``Function``
2319 .. _iterate_convert:
2321 Turning an iterator into a class pointer (and vice-versa)
2322 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2324 Sometimes, it'll be useful to grab a reference (or pointer) to a class instance
2325 when all you've got at hand is an iterator. Well, extracting a reference or a
2326 pointer from an iterator is very straight-forward. Assuming that ``i`` is a
2327 ``BasicBlock::iterator`` and ``j`` is a ``BasicBlock::const_iterator``:
2331 Instruction& inst = *i; // Grab reference to instruction reference
2332 Instruction* pinst = &*i; // Grab pointer to instruction reference
2333 const Instruction& inst = *j;
2335 However, the iterators you'll be working with in the LLVM framework are special:
2336 they will automatically convert to a ptr-to-instance type whenever they need to.
2337 Instead of dereferencing the iterator and then taking the address of the result,
2338 you can simply assign the iterator to the proper pointer type and you get the
2339 dereference and address-of operation as a result of the assignment (behind the
2340 scenes, this is a result of overloading casting mechanisms). Thus the second
2341 line of the last example,
2345 Instruction *pinst = &*i;
2347 is semantically equivalent to
2351 Instruction *pinst = i;
2353 It's also possible to turn a class pointer into the corresponding iterator, and
2354 this is a constant time operation (very efficient). The following code snippet
2355 illustrates use of the conversion constructors provided by LLVM iterators. By
2356 using these, you can explicitly grab the iterator of something without actually
2357 obtaining it via iteration over some structure:
2361 void printNextInstruction(Instruction* inst) {
2362 BasicBlock::iterator it(inst);
2363 ++it; // After this line, it refers to the instruction after *inst
2364 if (it != inst->getParent()->end()) errs() << *it << "\n";
2367 Unfortunately, these implicit conversions come at a cost; they prevent these
2368 iterators from conforming to standard iterator conventions, and thus from being
2369 usable with standard algorithms and containers. For example, they prevent the
2370 following code, where ``B`` is a ``BasicBlock``, from compiling:
2374 llvm::SmallVector<llvm::Instruction *, 16>(B->begin(), B->end());
2376 Because of this, these implicit conversions may be removed some day, and
2377 ``operator*`` changed to return a pointer instead of a reference.
2379 .. _iterate_complex:
2381 Finding call sites: a slightly more complex example
2382 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2384 Say that you're writing a FunctionPass and would like to count all the locations
2385 in the entire module (that is, across every ``Function``) where a certain
2386 function (i.e., some ``Function *``) is already in scope. As you'll learn
2387 later, you may want to use an ``InstVisitor`` to accomplish this in a much more
2388 straight-forward manner, but this example will allow us to explore how you'd do
2389 it if you didn't have ``InstVisitor`` around. In pseudo-code, this is what we
2392 .. code-block:: none
2394 initialize callCounter to zero
2395 for each Function f in the Module
2396 for each BasicBlock b in f
2397 for each Instruction i in b
2398 if (i is a CallInst and calls the given function)
2399 increment callCounter
2401 And the actual code is (remember, because we're writing a ``FunctionPass``, our
2402 ``FunctionPass``-derived class simply has to override the ``runOnFunction``
2407 Function* targetFunc = ...;
2409 class OurFunctionPass : public FunctionPass {
2411 OurFunctionPass(): callCounter(0) { }
2413 virtual runOnFunction(Function& F) {
2414 for (Function::iterator b = F.begin(), be = F.end(); b != be; ++b) {
2415 for (BasicBlock::iterator i = b->begin(), ie = b->end(); i != ie; ++i) {
2416 if (CallInst* callInst = dyn_cast<CallInst>(&*i)) {
2417 // We know we've encountered a call instruction, so we
2418 // need to determine if it's a call to the
2419 // function pointed to by m_func or not.
2420 if (callInst->getCalledFunction() == targetFunc)
2428 unsigned callCounter;
2431 .. _calls_and_invokes:
2433 Treating calls and invokes the same way
2434 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2436 You may have noticed that the previous example was a bit oversimplified in that
2437 it did not deal with call sites generated by 'invoke' instructions. In this,
2438 and in other situations, you may find that you want to treat ``CallInst``\ s and
2439 ``InvokeInst``\ s the same way, even though their most-specific common base
2440 class is ``Instruction``, which includes lots of less closely-related things.
2441 For these cases, LLVM provides a handy wrapper class called ``CallSite``
2442 (`doxygen <http://llvm.org/doxygen/classllvm_1_1CallSite.html>`__) It is
2443 essentially a wrapper around an ``Instruction`` pointer, with some methods that
2444 provide functionality common to ``CallInst``\ s and ``InvokeInst``\ s.
2446 This class has "value semantics": it should be passed by value, not by reference
2447 and it should not be dynamically allocated or deallocated using ``operator new``
2448 or ``operator delete``. It is efficiently copyable, assignable and
2449 constructable, with costs equivalents to that of a bare pointer. If you look at
2450 its definition, it has only a single pointer member.
2454 Iterating over def-use & use-def chains
2455 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2457 Frequently, we might have an instance of the ``Value`` class (`doxygen
2458 <http://llvm.org/doxygen/classllvm_1_1Value.html>`__) and we want to determine
2459 which ``User`` s use the ``Value``. The list of all ``User``\ s of a particular
2460 ``Value`` is called a *def-use* chain. For example, let's say we have a
2461 ``Function*`` named ``F`` to a particular function ``foo``. Finding all of the
2462 instructions that *use* ``foo`` is as simple as iterating over the *def-use*
2469 for (User *U : F->users()) {
2470 if (Instruction *Inst = dyn_cast<Instruction>(U)) {
2471 errs() << "F is used in instruction:\n";
2472 errs() << *Inst << "\n";
2475 Alternatively, it's common to have an instance of the ``User`` Class (`doxygen
2476 <http://llvm.org/doxygen/classllvm_1_1User.html>`__) and need to know what
2477 ``Value``\ s are used by it. The list of all ``Value``\ s used by a ``User`` is
2478 known as a *use-def* chain. Instances of class ``Instruction`` are common
2479 ``User`` s, so we might want to iterate over all of the values that a particular
2480 instruction uses (that is, the operands of the particular ``Instruction``):
2484 Instruction *pi = ...;
2486 for (Use &U : pi->operands()) {
2491 Declaring objects as ``const`` is an important tool of enforcing mutation free
2492 algorithms (such as analyses, etc.). For this purpose above iterators come in
2493 constant flavors as ``Value::const_use_iterator`` and
2494 ``Value::const_op_iterator``. They automatically arise when calling
2495 ``use/op_begin()`` on ``const Value*``\ s or ``const User*``\ s respectively.
2496 Upon dereferencing, they return ``const Use*``\ s. Otherwise the above patterns
2501 Iterating over predecessors & successors of blocks
2502 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2504 Iterating over the predecessors and successors of a block is quite easy with the
2505 routines defined in ``"llvm/IR/CFG.h"``. Just use code like this to
2506 iterate over all predecessors of BB:
2510 #include "llvm/IR/CFG.h"
2511 BasicBlock *BB = ...;
2513 for (pred_iterator PI = pred_begin(BB), E = pred_end(BB); PI != E; ++PI) {
2514 BasicBlock *Pred = *PI;
2518 Similarly, to iterate over successors use ``succ_iterator/succ_begin/succ_end``.
2522 Making simple changes
2523 ---------------------
2525 There are some primitive transformation operations present in the LLVM
2526 infrastructure that are worth knowing about. When performing transformations,
2527 it's fairly common to manipulate the contents of basic blocks. This section
2528 describes some of the common methods for doing so and gives example code.
2530 .. _schanges_creating:
2532 Creating and inserting new ``Instruction``\ s
2533 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2535 *Instantiating Instructions*
2537 Creation of ``Instruction``\ s is straight-forward: simply call the constructor
2538 for the kind of instruction to instantiate and provide the necessary parameters.
2539 For example, an ``AllocaInst`` only *requires* a (const-ptr-to) ``Type``. Thus:
2543 AllocaInst* ai = new AllocaInst(Type::Int32Ty);
2545 will create an ``AllocaInst`` instance that represents the allocation of one
2546 integer in the current stack frame, at run time. Each ``Instruction`` subclass
2547 is likely to have varying default parameters which change the semantics of the
2548 instruction, so refer to the `doxygen documentation for the subclass of
2549 Instruction <http://llvm.org/doxygen/classllvm_1_1Instruction.html>`_ that
2550 you're interested in instantiating.
2554 It is very useful to name the values of instructions when you're able to, as
2555 this facilitates the debugging of your transformations. If you end up looking
2556 at generated LLVM machine code, you definitely want to have logical names
2557 associated with the results of instructions! By supplying a value for the
2558 ``Name`` (default) parameter of the ``Instruction`` constructor, you associate a
2559 logical name with the result of the instruction's execution at run time. For
2560 example, say that I'm writing a transformation that dynamically allocates space
2561 for an integer on the stack, and that integer is going to be used as some kind
2562 of index by some other code. To accomplish this, I place an ``AllocaInst`` at
2563 the first point in the first ``BasicBlock`` of some ``Function``, and I'm
2564 intending to use it within the same ``Function``. I might do:
2568 AllocaInst* pa = new AllocaInst(Type::Int32Ty, 0, "indexLoc");
2570 where ``indexLoc`` is now the logical name of the instruction's execution value,
2571 which is a pointer to an integer on the run time stack.
2573 *Inserting instructions*
2575 There are essentially three ways to insert an ``Instruction`` into an existing
2576 sequence of instructions that form a ``BasicBlock``:
2578 * Insertion into an explicit instruction list
2580 Given a ``BasicBlock* pb``, an ``Instruction* pi`` within that ``BasicBlock``,
2581 and a newly-created instruction we wish to insert before ``*pi``, we do the
2586 BasicBlock *pb = ...;
2587 Instruction *pi = ...;
2588 Instruction *newInst = new Instruction(...);
2590 pb->getInstList().insert(pi, newInst); // Inserts newInst before pi in pb
2592 Appending to the end of a ``BasicBlock`` is so common that the ``Instruction``
2593 class and ``Instruction``-derived classes provide constructors which take a
2594 pointer to a ``BasicBlock`` to be appended to. For example code that looked
2599 BasicBlock *pb = ...;
2600 Instruction *newInst = new Instruction(...);
2602 pb->getInstList().push_back(newInst); // Appends newInst to pb
2608 BasicBlock *pb = ...;
2609 Instruction *newInst = new Instruction(..., pb);
2611 which is much cleaner, especially if you are creating long instruction
2614 * Insertion into an implicit instruction list
2616 ``Instruction`` instances that are already in ``BasicBlock``\ s are implicitly
2617 associated with an existing instruction list: the instruction list of the
2618 enclosing basic block. Thus, we could have accomplished the same thing as the
2619 above code without being given a ``BasicBlock`` by doing:
2623 Instruction *pi = ...;
2624 Instruction *newInst = new Instruction(...);
2626 pi->getParent()->getInstList().insert(pi, newInst);
2628 In fact, this sequence of steps occurs so frequently that the ``Instruction``
2629 class and ``Instruction``-derived classes provide constructors which take (as
2630 a default parameter) a pointer to an ``Instruction`` which the newly-created
2631 ``Instruction`` should precede. That is, ``Instruction`` constructors are
2632 capable of inserting the newly-created instance into the ``BasicBlock`` of a
2633 provided instruction, immediately before that instruction. Using an
2634 ``Instruction`` constructor with a ``insertBefore`` (default) parameter, the
2639 Instruction* pi = ...;
2640 Instruction* newInst = new Instruction(..., pi);
2642 which is much cleaner, especially if you're creating a lot of instructions and
2643 adding them to ``BasicBlock``\ s.
2645 * Insertion using an instance of ``IRBuilder``
2647 Inserting several ``Instruction``\ s can be quite laborious using the previous
2648 methods. The ``IRBuilder`` is a convenience class that can be used to add
2649 several instructions to the end of a ``BasicBlock`` or before a particular
2650 ``Instruction``. It also supports constant folding and renaming named
2651 registers (see ``IRBuilder``'s template arguments).
2653 The example below demonstrates a very simple use of the ``IRBuilder`` where
2654 three instructions are inserted before the instruction ``pi``. The first two
2655 instructions are Call instructions and third instruction multiplies the return
2656 value of the two calls.
2660 Instruction *pi = ...;
2661 IRBuilder<> Builder(pi);
2662 CallInst* callOne = Builder.CreateCall(...);
2663 CallInst* callTwo = Builder.CreateCall(...);
2664 Value* result = Builder.CreateMul(callOne, callTwo);
2666 The example below is similar to the above example except that the created
2667 ``IRBuilder`` inserts instructions at the end of the ``BasicBlock`` ``pb``.
2671 BasicBlock *pb = ...;
2672 IRBuilder<> Builder(pb);
2673 CallInst* callOne = Builder.CreateCall(...);
2674 CallInst* callTwo = Builder.CreateCall(...);
2675 Value* result = Builder.CreateMul(callOne, callTwo);
2677 See :doc:`tutorial/LangImpl03` for a practical use of the ``IRBuilder``.
2680 .. _schanges_deleting:
2682 Deleting Instructions
2683 ^^^^^^^^^^^^^^^^^^^^^
2685 Deleting an instruction from an existing sequence of instructions that form a
2686 BasicBlock_ is very straight-forward: just call the instruction's
2687 ``eraseFromParent()`` method. For example:
2691 Instruction *I = .. ;
2692 I->eraseFromParent();
2694 This unlinks the instruction from its containing basic block and deletes it. If
2695 you'd just like to unlink the instruction from its containing basic block but
2696 not delete it, you can use the ``removeFromParent()`` method.
2698 .. _schanges_replacing:
2700 Replacing an Instruction with another Value
2701 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2703 Replacing individual instructions
2704 """""""""""""""""""""""""""""""""
2706 Including "`llvm/Transforms/Utils/BasicBlockUtils.h
2707 <http://llvm.org/doxygen/BasicBlockUtils_8h-source.html>`_" permits use of two
2708 very useful replace functions: ``ReplaceInstWithValue`` and
2709 ``ReplaceInstWithInst``.
2711 .. _schanges_deleting_sub:
2713 Deleting Instructions
2714 """""""""""""""""""""
2716 * ``ReplaceInstWithValue``
2718 This function replaces all uses of a given instruction with a value, and then
2719 removes the original instruction. The following example illustrates the
2720 replacement of the result of a particular ``AllocaInst`` that allocates memory
2721 for a single integer with a null pointer to an integer.
2725 AllocaInst* instToReplace = ...;
2726 BasicBlock::iterator ii(instToReplace);
2728 ReplaceInstWithValue(instToReplace->getParent()->getInstList(), ii,
2729 Constant::getNullValue(PointerType::getUnqual(Type::Int32Ty)));
2731 * ``ReplaceInstWithInst``
2733 This function replaces a particular instruction with another instruction,
2734 inserting the new instruction into the basic block at the location where the
2735 old instruction was, and replacing any uses of the old instruction with the
2736 new instruction. The following example illustrates the replacement of one
2737 ``AllocaInst`` with another.
2741 AllocaInst* instToReplace = ...;
2742 BasicBlock::iterator ii(instToReplace);
2744 ReplaceInstWithInst(instToReplace->getParent()->getInstList(), ii,
2745 new AllocaInst(Type::Int32Ty, 0, "ptrToReplacedInt"));
2748 Replacing multiple uses of Users and Values
2749 """""""""""""""""""""""""""""""""""""""""""
2751 You can use ``Value::replaceAllUsesWith`` and ``User::replaceUsesOfWith`` to
2752 change more than one use at a time. See the doxygen documentation for the
2753 `Value Class <http://llvm.org/doxygen/classllvm_1_1Value.html>`_ and `User Class
2754 <http://llvm.org/doxygen/classllvm_1_1User.html>`_, respectively, for more
2757 .. _schanges_deletingGV:
2759 Deleting GlobalVariables
2760 ^^^^^^^^^^^^^^^^^^^^^^^^
2762 Deleting a global variable from a module is just as easy as deleting an
2763 Instruction. First, you must have a pointer to the global variable that you
2764 wish to delete. You use this pointer to erase it from its parent, the module.
2769 GlobalVariable *GV = .. ;
2771 GV->eraseFromParent();
2779 In generating IR, you may need some complex types. If you know these types
2780 statically, you can use ``TypeBuilder<...>::get()``, defined in
2781 ``llvm/Support/TypeBuilder.h``, to retrieve them. ``TypeBuilder`` has two forms
2782 depending on whether you're building types for cross-compilation or native
2783 library use. ``TypeBuilder<T, true>`` requires that ``T`` be independent of the
2784 host environment, meaning that it's built out of types from the ``llvm::types``
2785 (`doxygen <http://llvm.org/doxygen/namespacellvm_1_1types.html>`__) namespace
2786 and pointers, functions, arrays, etc. built of those. ``TypeBuilder<T, false>``
2787 additionally allows native C types whose size may depend on the host compiler.
2792 FunctionType *ft = TypeBuilder<types::i<8>(types::i<32>*), true>::get();
2794 is easier to read and write than the equivalent
2798 std::vector<const Type*> params;
2799 params.push_back(PointerType::getUnqual(Type::Int32Ty));
2800 FunctionType *ft = FunctionType::get(Type::Int8Ty, params, false);
2802 See the `class comment
2803 <http://llvm.org/doxygen/TypeBuilder_8h-source.html#l00001>`_ for more details.
2810 This section describes the interaction of the LLVM APIs with multithreading,
2811 both on the part of client applications, and in the JIT, in the hosted
2814 Note that LLVM's support for multithreading is still relatively young. Up
2815 through version 2.5, the execution of threaded hosted applications was
2816 supported, but not threaded client access to the APIs. While this use case is
2817 now supported, clients *must* adhere to the guidelines specified below to ensure
2818 proper operation in multithreaded mode.
2820 Note that, on Unix-like platforms, LLVM requires the presence of GCC's atomic
2821 intrinsics in order to support threaded operation. If you need a
2822 multhreading-capable LLVM on a platform without a suitably modern system
2823 compiler, consider compiling LLVM and LLVM-GCC in single-threaded mode, and
2824 using the resultant compiler to build a copy of LLVM with multithreading
2829 Ending Execution with ``llvm_shutdown()``
2830 -----------------------------------------
2832 When you are done using the LLVM APIs, you should call ``llvm_shutdown()`` to
2833 deallocate memory used for internal structures.
2837 Lazy Initialization with ``ManagedStatic``
2838 ------------------------------------------
2840 ``ManagedStatic`` is a utility class in LLVM used to implement static
2841 initialization of static resources, such as the global type tables. In a
2842 single-threaded environment, it implements a simple lazy initialization scheme.
2843 When LLVM is compiled with support for multi-threading, however, it uses
2844 double-checked locking to implement thread-safe lazy initialization.
2848 Achieving Isolation with ``LLVMContext``
2849 ----------------------------------------
2851 ``LLVMContext`` is an opaque class in the LLVM API which clients can use to
2852 operate multiple, isolated instances of LLVM concurrently within the same
2853 address space. For instance, in a hypothetical compile-server, the compilation
2854 of an individual translation unit is conceptually independent from all the
2855 others, and it would be desirable to be able to compile incoming translation
2856 units concurrently on independent server threads. Fortunately, ``LLVMContext``
2857 exists to enable just this kind of scenario!
2859 Conceptually, ``LLVMContext`` provides isolation. Every LLVM entity
2860 (``Module``\ s, ``Value``\ s, ``Type``\ s, ``Constant``\ s, etc.) in LLVM's
2861 in-memory IR belongs to an ``LLVMContext``. Entities in different contexts
2862 *cannot* interact with each other: ``Module``\ s in different contexts cannot be
2863 linked together, ``Function``\ s cannot be added to ``Module``\ s in different
2864 contexts, etc. What this means is that is is safe to compile on multiple
2865 threads simultaneously, as long as no two threads operate on entities within the
2868 In practice, very few places in the API require the explicit specification of a
2869 ``LLVMContext``, other than the ``Type`` creation/lookup APIs. Because every
2870 ``Type`` carries a reference to its owning context, most other entities can
2871 determine what context they belong to by looking at their own ``Type``. If you
2872 are adding new entities to LLVM IR, please try to maintain this interface
2880 LLVM's "eager" JIT compiler is safe to use in threaded programs. Multiple
2881 threads can call ``ExecutionEngine::getPointerToFunction()`` or
2882 ``ExecutionEngine::runFunction()`` concurrently, and multiple threads can run
2883 code output by the JIT concurrently. The user must still ensure that only one
2884 thread accesses IR in a given ``LLVMContext`` while another thread might be
2885 modifying it. One way to do that is to always hold the JIT lock while accessing
2886 IR outside the JIT (the JIT *modifies* the IR by adding ``CallbackVH``\ s).
2887 Another way is to only call ``getPointerToFunction()`` from the
2888 ``LLVMContext``'s thread.
2890 When the JIT is configured to compile lazily (using
2891 ``ExecutionEngine::DisableLazyCompilation(false)``), there is currently a `race
2892 condition <http://llvm.org/bugs/show_bug.cgi?id=5184>`_ in updating call sites
2893 after a function is lazily-jitted. It's still possible to use the lazy JIT in a
2894 threaded program if you ensure that only one thread at a time can call any
2895 particular lazy stub and that the JIT lock guards any IR access, but we suggest
2896 using only the eager JIT in threaded programs.
2903 This section describes some of the advanced or obscure API's that most clients
2904 do not need to be aware of. These API's tend manage the inner workings of the
2905 LLVM system, and only need to be accessed in unusual circumstances.
2909 The ``ValueSymbolTable`` class
2910 ------------------------------
2912 The ``ValueSymbolTable`` (`doxygen
2913 <http://llvm.org/doxygen/classllvm_1_1ValueSymbolTable.html>`__) class provides
2914 a symbol table that the :ref:`Function <c_Function>` and Module_ classes use for
2915 naming value definitions. The symbol table can provide a name for any Value_.
2917 Note that the ``SymbolTable`` class should not be directly accessed by most
2918 clients. It should only be used when iteration over the symbol table names
2919 themselves are required, which is very special purpose. Note that not all LLVM
2920 Value_\ s have names, and those without names (i.e. they have an empty name) do
2921 not exist in the symbol table.
2923 Symbol tables support iteration over the values in the symbol table with
2924 ``begin/end/iterator`` and supports querying to see if a specific name is in the
2925 symbol table (with ``lookup``). The ``ValueSymbolTable`` class exposes no
2926 public mutator methods, instead, simply call ``setName`` on a value, which will
2927 autoinsert it into the appropriate symbol table.
2931 The ``User`` and owned ``Use`` classes' memory layout
2932 -----------------------------------------------------
2934 The ``User`` (`doxygen <http://llvm.org/doxygen/classllvm_1_1User.html>`__)
2935 class provides a basis for expressing the ownership of ``User`` towards other
2936 `Value instance <http://llvm.org/doxygen/classllvm_1_1Value.html>`_\ s. The
2937 ``Use`` (`doxygen <http://llvm.org/doxygen/classllvm_1_1Use.html>`__) helper
2938 class is employed to do the bookkeeping and to facilitate *O(1)* addition and
2943 Interaction and relationship between ``User`` and ``Use`` objects
2944 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2946 A subclass of ``User`` can choose between incorporating its ``Use`` objects or
2947 refer to them out-of-line by means of a pointer. A mixed variant (some ``Use``
2948 s inline others hung off) is impractical and breaks the invariant that the
2949 ``Use`` objects belonging to the same ``User`` form a contiguous array.
2951 We have 2 different layouts in the ``User`` (sub)classes:
2955 The ``Use`` object(s) are inside (resp. at fixed offset) of the ``User``
2956 object and there are a fixed number of them.
2960 The ``Use`` object(s) are referenced by a pointer to an array from the
2961 ``User`` object and there may be a variable number of them.
2963 As of v2.4 each layout still possesses a direct pointer to the start of the
2964 array of ``Use``\ s. Though not mandatory for layout a), we stick to this
2965 redundancy for the sake of simplicity. The ``User`` object also stores the
2966 number of ``Use`` objects it has. (Theoretically this information can also be
2967 calculated given the scheme presented below.)
2969 Special forms of allocation operators (``operator new``) enforce the following
2972 * Layout a) is modelled by prepending the ``User`` object by the ``Use[]``
2975 .. code-block:: none
2977 ...---.---.---.---.-------...
2978 | P | P | P | P | User
2979 '''---'---'---'---'-------'''
2981 * Layout b) is modelled by pointing at the ``Use[]`` array.
2983 .. code-block:: none
2994 *(In the above figures* '``P``' *stands for the* ``Use**`` *that is stored in
2995 each* ``Use`` *object in the member* ``Use::Prev`` *)*
2999 The waymarking algorithm
3000 ^^^^^^^^^^^^^^^^^^^^^^^^
3002 Since the ``Use`` objects are deprived of the direct (back)pointer to their
3003 ``User`` objects, there must be a fast and exact method to recover it. This is
3004 accomplished by the following scheme:
3006 A bit-encoding in the 2 LSBits (least significant bits) of the ``Use::Prev``
3007 allows to find the start of the ``User`` object:
3009 * ``00`` --- binary digit 0
3011 * ``01`` --- binary digit 1
3013 * ``10`` --- stop and calculate (``s``)
3015 * ``11`` --- full stop (``S``)
3017 Given a ``Use*``, all we have to do is to walk till we get a stop and we either
3018 have a ``User`` immediately behind or we have to walk to the next stop picking
3019 up digits and calculating the offset:
3021 .. code-block:: none
3023 .---.---.---.---.---.---.---.---.---.---.---.---.---.---.---.---.----------------
3024 | 1 | s | 1 | 0 | 1 | 0 | s | 1 | 1 | 0 | s | 1 | 1 | s | 1 | S | User (or User*)
3025 '---'---'---'---'---'---'---'---'---'---'---'---'---'---'---'---'----------------
3026 |+15 |+10 |+6 |+3 |+1
3029 | | | ______________________>
3030 | | ______________________________________>
3031 | __________________________________________________________>
3033 Only the significant number of bits need to be stored between the stops, so that
3034 the *worst case is 20 memory accesses* when there are 1000 ``Use`` objects
3035 associated with a ``User``.
3039 Reference implementation
3040 ^^^^^^^^^^^^^^^^^^^^^^^^
3042 The following literate Haskell fragment demonstrates the concept:
3044 .. code-block:: haskell
3046 > import Test.QuickCheck
3048 > digits :: Int -> [Char] -> [Char]
3049 > digits 0 acc = '0' : acc
3050 > digits 1 acc = '1' : acc
3051 > digits n acc = digits (n `div` 2) $ digits (n `mod` 2) acc
3053 > dist :: Int -> [Char] -> [Char]
3056 > dist 1 acc = let r = dist 0 acc in 's' : digits (length r) r
3057 > dist n acc = dist (n - 1) $ dist 1 acc
3059 > takeLast n ss = reverse $ take n $ reverse ss
3061 > test = takeLast 40 $ dist 20 []
3064 Printing <test> gives: ``"1s100000s11010s10100s1111s1010s110s11s1S"``
3066 The reverse algorithm computes the length of the string just by examining a
3069 .. code-block:: haskell
3071 > pref :: [Char] -> Int
3073 > pref ('s':'1':rest) = decode 2 1 rest
3074 > pref (_:rest) = 1 + pref rest
3076 > decode walk acc ('0':rest) = decode (walk + 1) (acc * 2) rest
3077 > decode walk acc ('1':rest) = decode (walk + 1) (acc * 2 + 1) rest
3078 > decode walk acc _ = walk + acc
3081 Now, as expected, printing <pref test> gives ``40``.
3083 We can *quickCheck* this with following property:
3085 .. code-block:: haskell
3087 > testcase = dist 2000 []
3088 > testcaseLength = length testcase
3090 > identityProp n = n > 0 && n <= testcaseLength ==> length arr == pref arr
3091 > where arr = takeLast n testcase
3094 As expected <quickCheck identityProp> gives:
3098 *Main> quickCheck identityProp
3099 OK, passed 100 tests.
3101 Let's be a bit more exhaustive:
3103 .. code-block:: haskell
3106 > deepCheck p = check (defaultConfig { configMaxTest = 500 }) p
3109 And here is the result of <deepCheck identityProp>:
3113 *Main> deepCheck identityProp
3114 OK, passed 500 tests.
3118 Tagging considerations
3119 ^^^^^^^^^^^^^^^^^^^^^^
3121 To maintain the invariant that the 2 LSBits of each ``Use**`` in ``Use`` never
3122 change after being set up, setters of ``Use::Prev`` must re-tag the new
3123 ``Use**`` on every modification. Accordingly getters must strip the tag bits.
3125 For layout b) instead of the ``User`` we find a pointer (``User*`` with LSBit
3126 set). Following this pointer brings us to the ``User``. A portable trick
3127 ensures that the first bytes of ``User`` (if interpreted as a pointer) never has
3128 the LSBit set. (Portability is relying on the fact that all known compilers
3129 place the ``vptr`` in the first word of the instances.)
3133 Designing Type Hiercharies and Polymorphic Interfaces
3134 -----------------------------------------------------
3136 There are two different design patterns that tend to result in the use of
3137 virtual dispatch for methods in a type hierarchy in C++ programs. The first is
3138 a genuine type hierarchy where different types in the hierarchy model
3139 a specific subset of the functionality and semantics, and these types nest
3140 strictly within each other. Good examples of this can be seen in the ``Value``
3141 or ``Type`` type hierarchies.
3143 A second is the desire to dispatch dynamically across a collection of
3144 polymorphic interface implementations. This latter use case can be modeled with
3145 virtual dispatch and inheritance by defining an abstract interface base class
3146 which all implementations derive from and override. However, this
3147 implementation strategy forces an **"is-a"** relationship to exist that is not
3148 actually meaningful. There is often not some nested hierarchy of useful
3149 generalizations which code might interact with and move up and down. Instead,
3150 there is a singular interface which is dispatched across a range of
3153 The preferred implementation strategy for the second use case is that of
3154 generic programming (sometimes called "compile-time duck typing" or "static
3155 polymorphism"). For example, a template over some type parameter ``T`` can be
3156 instantiated across any particular implementation that conforms to the
3157 interface or *concept*. A good example here is the highly generic properties of
3158 any type which models a node in a directed graph. LLVM models these primarily
3159 through templates and generic programming. Such templates include the
3160 ``LoopInfoBase`` and ``DominatorTreeBase``. When this type of polymorphism
3161 truly needs **dynamic** dispatch you can generalize it using a technique
3162 called *concept-based polymorphism*. This pattern emulates the interfaces and
3163 behaviors of templates using a very limited form of virtual dispatch for type
3164 erasure inside its implementation. You can find examples of this technique in
3165 the ``PassManager.h`` system, and there is a more detailed introduction to it
3166 by Sean Parent in several of his talks and papers:
3168 #. `Inheritance Is The Base Class of Evil
3169 <http://channel9.msdn.com/Events/GoingNative/2013/Inheritance-Is-The-Base-Class-of-Evil>`_
3170 - The GoingNative 2013 talk describing this technique, and probably the best
3172 #. `Value Semantics and Concepts-based Polymorphism
3173 <http://www.youtube.com/watch?v=_BpMYeUFXv8>`_ - The C++Now! 2012 talk
3174 describing this technique in more detail.
3175 #. `Sean Parent's Papers and Presentations
3176 <http://github.com/sean-parent/sean-parent.github.com/wiki/Papers-and-Presentations>`_
3177 - A Github project full of links to slides, video, and sometimes code.
3179 When deciding between creating a type hierarchy (with either tagged or virtual
3180 dispatch) and using templates or concepts-based polymorphism, consider whether
3181 there is some refinement of an abstract base class which is a semantically
3182 meaningful type on an interface boundary. If anything more refined than the
3183 root abstract interface is meaningless to talk about as a partial extension of
3184 the semantic model, then your use case likely fits better with polymorphism and
3185 you should avoid using virtual dispatch. However, there may be some exigent
3186 circumstances that require one technique or the other to be used.
3188 If you do need to introduce a type hierarchy, we prefer to use explicitly
3189 closed type hierarchies with manual tagged dispatch and/or RTTI rather than the
3190 open inheritance model and virtual dispatch that is more common in C++ code.
3191 This is because LLVM rarely encourages library consumers to extend its core
3192 types, and leverages the closed and tag-dispatched nature of its hierarchies to
3193 generate significantly more efficient code. We have also found that a large
3194 amount of our usage of type hierarchies fits better with tag-based pattern
3195 matching rather than dynamic dispatch across a common interface. Within LLVM we
3196 have built custom helpers to facilitate this design. See this document's
3197 section on :ref:`isa and dyn_cast <isa>` and our :doc:`detailed document
3198 <HowToSetUpLLVMStyleRTTI>` which describes how you can implement this
3199 pattern for use with the LLVM helpers.
3201 .. _abi_breaking_checks:
3206 Checks and asserts that alter the LLVM C++ ABI are predicated on the
3207 preprocessor symbol `LLVM_ENABLE_ABI_BREAKING_CHECKS` -- LLVM
3208 libraries built with `LLVM_ENABLE_ABI_BREAKING_CHECKS` are not ABI
3209 compatible LLVM libraries built without it defined. By default,
3210 turning on assertions also turns on `LLVM_ENABLE_ABI_BREAKING_CHECKS`
3211 so a default +Asserts build is not ABI compatible with a
3212 default -Asserts build. Clients that want ABI compatibility
3213 between +Asserts and -Asserts builds should use the CMake or autoconf
3214 build systems to set `LLVM_ENABLE_ABI_BREAKING_CHECKS` independently
3215 of `LLVM_ENABLE_ASSERTIONS`.
3219 The Core LLVM Class Hierarchy Reference
3220 =======================================
3222 ``#include "llvm/IR/Type.h"``
3224 header source: `Type.h <http://llvm.org/doxygen/Type_8h-source.html>`_
3226 doxygen info: `Type Clases <http://llvm.org/doxygen/classllvm_1_1Type.html>`_
3228 The Core LLVM classes are the primary means of representing the program being
3229 inspected or transformed. The core LLVM classes are defined in header files in
3230 the ``include/llvm/IR`` directory, and implemented in the ``lib/IR``
3231 directory. It's worth noting that, for historical reasons, this library is
3232 called ``libLLVMCore.so``, not ``libLLVMIR.so`` as you might expect.
3236 The Type class and Derived Types
3237 --------------------------------
3239 ``Type`` is a superclass of all type classes. Every ``Value`` has a ``Type``.
3240 ``Type`` cannot be instantiated directly but only through its subclasses.
3241 Certain primitive types (``VoidType``, ``LabelType``, ``FloatType`` and
3242 ``DoubleType``) have hidden subclasses. They are hidden because they offer no
3243 useful functionality beyond what the ``Type`` class offers except to distinguish
3244 themselves from other subclasses of ``Type``.
3246 All other types are subclasses of ``DerivedType``. Types can be named, but this
3247 is not a requirement. There exists exactly one instance of a given shape at any
3248 one time. This allows type equality to be performed with address equality of
3249 the Type Instance. That is, given two ``Type*`` values, the types are identical
3250 if the pointers are identical.
3254 Important Public Methods
3255 ^^^^^^^^^^^^^^^^^^^^^^^^
3257 * ``bool isIntegerTy() const``: Returns true for any integer type.
3259 * ``bool isFloatingPointTy()``: Return true if this is one of the five
3260 floating point types.
3262 * ``bool isSized()``: Return true if the type has known size. Things
3263 that don't have a size are abstract types, labels and void.
3267 Important Derived Types
3268 ^^^^^^^^^^^^^^^^^^^^^^^
3271 Subclass of DerivedType that represents integer types of any bit width. Any
3272 bit width between ``IntegerType::MIN_INT_BITS`` (1) and
3273 ``IntegerType::MAX_INT_BITS`` (~8 million) can be represented.
3275 * ``static const IntegerType* get(unsigned NumBits)``: get an integer
3276 type of a specific bit width.
3278 * ``unsigned getBitWidth() const``: Get the bit width of an integer type.
3281 This is subclassed by ArrayType, PointerType and VectorType.
3283 * ``const Type * getElementType() const``: Returns the type of each
3284 of the elements in the sequential type.
3287 This is a subclass of SequentialType and defines the interface for array
3290 * ``unsigned getNumElements() const``: Returns the number of elements
3294 Subclass of SequentialType for pointer types.
3297 Subclass of SequentialType for vector types. A vector type is similar to an
3298 ArrayType but is distinguished because it is a first class type whereas
3299 ArrayType is not. Vector types are used for vector operations and are usually
3300 small vectors of an integer or floating point type.
3303 Subclass of DerivedTypes for struct types.
3308 Subclass of DerivedTypes for function types.
3310 * ``bool isVarArg() const``: Returns true if it's a vararg function.
3312 * ``const Type * getReturnType() const``: Returns the return type of the
3315 * ``const Type * getParamType (unsigned i)``: Returns the type of the ith
3318 * ``const unsigned getNumParams() const``: Returns the number of formal
3323 The ``Module`` class
3324 --------------------
3326 ``#include "llvm/IR/Module.h"``
3328 header source: `Module.h <http://llvm.org/doxygen/Module_8h-source.html>`_
3330 doxygen info: `Module Class <http://llvm.org/doxygen/classllvm_1_1Module.html>`_
3332 The ``Module`` class represents the top level structure present in LLVM
3333 programs. An LLVM module is effectively either a translation unit of the
3334 original program or a combination of several translation units merged by the
3335 linker. The ``Module`` class keeps track of a list of :ref:`Function
3336 <c_Function>`\ s, a list of GlobalVariable_\ s, and a SymbolTable_.
3337 Additionally, it contains a few helpful member functions that try to make common
3342 Important Public Members of the ``Module`` class
3343 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3345 * ``Module::Module(std::string name = "")``
3347 Constructing a Module_ is easy. You can optionally provide a name for it
3348 (probably based on the name of the translation unit).
3350 * | ``Module::iterator`` - Typedef for function list iterator
3351 | ``Module::const_iterator`` - Typedef for const_iterator.
3352 | ``begin()``, ``end()``, ``size()``, ``empty()``
3354 These are forwarding methods that make it easy to access the contents of a
3355 ``Module`` object's :ref:`Function <c_Function>` list.
3357 * ``Module::FunctionListType &getFunctionList()``
3359 Returns the list of :ref:`Function <c_Function>`\ s. This is necessary to use
3360 when you need to update the list or perform a complex action that doesn't have
3361 a forwarding method.
3365 * | ``Module::global_iterator`` - Typedef for global variable list iterator
3366 | ``Module::const_global_iterator`` - Typedef for const_iterator.
3367 | ``global_begin()``, ``global_end()``, ``global_size()``, ``global_empty()``
3369 These are forwarding methods that make it easy to access the contents of a
3370 ``Module`` object's GlobalVariable_ list.
3372 * ``Module::GlobalListType &getGlobalList()``
3374 Returns the list of GlobalVariable_\ s. This is necessary to use when you
3375 need to update the list or perform a complex action that doesn't have a
3380 * ``SymbolTable *getSymbolTable()``
3382 Return a reference to the SymbolTable_ for this ``Module``.
3386 * ``Function *getFunction(StringRef Name) const``
3388 Look up the specified function in the ``Module`` SymbolTable_. If it does not
3389 exist, return ``null``.
3391 * ``Function *getOrInsertFunction(const std::string &Name, const FunctionType
3394 Look up the specified function in the ``Module`` SymbolTable_. If it does not
3395 exist, add an external declaration for the function and return it.
3397 * ``std::string getTypeName(const Type *Ty)``
3399 If there is at least one entry in the SymbolTable_ for the specified Type_,
3400 return it. Otherwise return the empty string.
3402 * ``bool addTypeName(const std::string &Name, const Type *Ty)``
3404 Insert an entry in the SymbolTable_ mapping ``Name`` to ``Ty``. If there is
3405 already an entry for this name, true is returned and the SymbolTable_ is not
3413 ``#include "llvm/IR/Value.h"``
3415 header source: `Value.h <http://llvm.org/doxygen/Value_8h-source.html>`_
3417 doxygen info: `Value Class <http://llvm.org/doxygen/classllvm_1_1Value.html>`_
3419 The ``Value`` class is the most important class in the LLVM Source base. It
3420 represents a typed value that may be used (among other things) as an operand to
3421 an instruction. There are many different types of ``Value``\ s, such as
3422 Constant_\ s, Argument_\ s. Even Instruction_\ s and :ref:`Function
3423 <c_Function>`\ s are ``Value``\ s.
3425 A particular ``Value`` may be used many times in the LLVM representation for a
3426 program. For example, an incoming argument to a function (represented with an
3427 instance of the Argument_ class) is "used" by every instruction in the function
3428 that references the argument. To keep track of this relationship, the ``Value``
3429 class keeps a list of all of the ``User``\ s that is using it (the User_ class
3430 is a base class for all nodes in the LLVM graph that can refer to ``Value``\ s).
3431 This use list is how LLVM represents def-use information in the program, and is
3432 accessible through the ``use_*`` methods, shown below.
3434 Because LLVM is a typed representation, every LLVM ``Value`` is typed, and this
3435 Type_ is available through the ``getType()`` method. In addition, all LLVM
3436 values can be named. The "name" of the ``Value`` is a symbolic string printed
3439 .. code-block:: llvm
3445 The name of this instruction is "foo". **NOTE** that the name of any value may
3446 be missing (an empty string), so names should **ONLY** be used for debugging
3447 (making the source code easier to read, debugging printouts), they should not be
3448 used to keep track of values or map between them. For this purpose, use a
3449 ``std::map`` of pointers to the ``Value`` itself instead.
3451 One important aspect of LLVM is that there is no distinction between an SSA
3452 variable and the operation that produces it. Because of this, any reference to
3453 the value produced by an instruction (or the value available as an incoming
3454 argument, for example) is represented as a direct pointer to the instance of the
3455 class that represents this value. Although this may take some getting used to,
3456 it simplifies the representation and makes it easier to manipulate.
3460 Important Public Members of the ``Value`` class
3461 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3463 * | ``Value::use_iterator`` - Typedef for iterator over the use-list
3464 | ``Value::const_use_iterator`` - Typedef for const_iterator over the
3466 | ``unsigned use_size()`` - Returns the number of users of the value.
3467 | ``bool use_empty()`` - Returns true if there are no users.
3468 | ``use_iterator use_begin()`` - Get an iterator to the start of the
3470 | ``use_iterator use_end()`` - Get an iterator to the end of the use-list.
3471 | ``User *use_back()`` - Returns the last element in the list.
3473 These methods are the interface to access the def-use information in LLVM.
3474 As with all other iterators in LLVM, the naming conventions follow the
3475 conventions defined by the STL_.
3477 * ``Type *getType() const``
3478 This method returns the Type of the Value.
3480 * | ``bool hasName() const``
3481 | ``std::string getName() const``
3482 | ``void setName(const std::string &Name)``
3484 This family of methods is used to access and assign a name to a ``Value``, be
3485 aware of the :ref:`precaution above <nameWarning>`.
3487 * ``void replaceAllUsesWith(Value *V)``
3489 This method traverses the use list of a ``Value`` changing all User_\ s of the
3490 current value to refer to "``V``" instead. For example, if you detect that an
3491 instruction always produces a constant value (for example through constant
3492 folding), you can replace all uses of the instruction with the constant like
3497 Inst->replaceAllUsesWith(ConstVal);
3504 ``#include "llvm/IR/User.h"``
3506 header source: `User.h <http://llvm.org/doxygen/User_8h-source.html>`_
3508 doxygen info: `User Class <http://llvm.org/doxygen/classllvm_1_1User.html>`_
3512 The ``User`` class is the common base class of all LLVM nodes that may refer to
3513 ``Value``\ s. It exposes a list of "Operands" that are all of the ``Value``\ s
3514 that the User is referring to. The ``User`` class itself is a subclass of
3517 The operands of a ``User`` point directly to the LLVM ``Value`` that it refers
3518 to. Because LLVM uses Static Single Assignment (SSA) form, there can only be
3519 one definition referred to, allowing this direct connection. This connection
3520 provides the use-def information in LLVM.
3524 Important Public Members of the ``User`` class
3525 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3527 The ``User`` class exposes the operand list in two ways: through an index access
3528 interface and through an iterator based interface.
3530 * | ``Value *getOperand(unsigned i)``
3531 | ``unsigned getNumOperands()``
3533 These two methods expose the operands of the ``User`` in a convenient form for
3536 * | ``User::op_iterator`` - Typedef for iterator over the operand list
3537 | ``op_iterator op_begin()`` - Get an iterator to the start of the operand
3539 | ``op_iterator op_end()`` - Get an iterator to the end of the operand list.
3541 Together, these methods make up the iterator based interface to the operands
3547 The ``Instruction`` class
3548 -------------------------
3550 ``#include "llvm/IR/Instruction.h"``
3552 header source: `Instruction.h
3553 <http://llvm.org/doxygen/Instruction_8h-source.html>`_
3555 doxygen info: `Instruction Class
3556 <http://llvm.org/doxygen/classllvm_1_1Instruction.html>`_
3558 Superclasses: User_, Value_
3560 The ``Instruction`` class is the common base class for all LLVM instructions.
3561 It provides only a few methods, but is a very commonly used class. The primary
3562 data tracked by the ``Instruction`` class itself is the opcode (instruction
3563 type) and the parent BasicBlock_ the ``Instruction`` is embedded into. To
3564 represent a specific type of instruction, one of many subclasses of
3565 ``Instruction`` are used.
3567 Because the ``Instruction`` class subclasses the User_ class, its operands can
3568 be accessed in the same way as for other ``User``\ s (with the
3569 ``getOperand()``/``getNumOperands()`` and ``op_begin()``/``op_end()`` methods).
3570 An important file for the ``Instruction`` class is the ``llvm/Instruction.def``
3571 file. This file contains some meta-data about the various different types of
3572 instructions in LLVM. It describes the enum values that are used as opcodes
3573 (for example ``Instruction::Add`` and ``Instruction::ICmp``), as well as the
3574 concrete sub-classes of ``Instruction`` that implement the instruction (for
3575 example BinaryOperator_ and CmpInst_). Unfortunately, the use of macros in this
3576 file confuses doxygen, so these enum values don't show up correctly in the
3577 `doxygen output <http://llvm.org/doxygen/classllvm_1_1Instruction.html>`_.
3581 Important Subclasses of the ``Instruction`` class
3582 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3586 * ``BinaryOperator``
3588 This subclasses represents all two operand instructions whose operands must be
3589 the same type, except for the comparison instructions.
3594 This subclass is the parent of the 12 casting instructions. It provides
3595 common operations on cast instructions.
3601 This subclass respresents the two comparison instructions,
3602 `ICmpInst <LangRef.html#i_icmp>`_ (integer opreands), and
3603 `FCmpInst <LangRef.html#i_fcmp>`_ (floating point operands).
3607 * ``TerminatorInst``
3609 This subclass is the parent of all terminator instructions (those which can
3614 Important Public Members of the ``Instruction`` class
3615 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3617 * ``BasicBlock *getParent()``
3619 Returns the BasicBlock_ that this
3620 ``Instruction`` is embedded into.
3622 * ``bool mayWriteToMemory()``
3624 Returns true if the instruction writes to memory, i.e. it is a ``call``,
3625 ``free``, ``invoke``, or ``store``.
3627 * ``unsigned getOpcode()``
3629 Returns the opcode for the ``Instruction``.
3631 * ``Instruction *clone() const``
3633 Returns another instance of the specified instruction, identical in all ways
3634 to the original except that the instruction has no parent (i.e. it's not
3635 embedded into a BasicBlock_), and it has no name.
3639 The ``Constant`` class and subclasses
3640 -------------------------------------
3642 Constant represents a base class for different types of constants. It is
3643 subclassed by ConstantInt, ConstantArray, etc. for representing the various
3644 types of Constants. GlobalValue_ is also a subclass, which represents the
3645 address of a global variable or function.
3649 Important Subclasses of Constant
3650 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3652 * ConstantInt : This subclass of Constant represents an integer constant of
3655 * ``const APInt& getValue() const``: Returns the underlying
3656 value of this constant, an APInt value.
3658 * ``int64_t getSExtValue() const``: Converts the underlying APInt value to an
3659 int64_t via sign extension. If the value (not the bit width) of the APInt
3660 is too large to fit in an int64_t, an assertion will result. For this
3661 reason, use of this method is discouraged.
3663 * ``uint64_t getZExtValue() const``: Converts the underlying APInt value
3664 to a uint64_t via zero extension. IF the value (not the bit width) of the
3665 APInt is too large to fit in a uint64_t, an assertion will result. For this
3666 reason, use of this method is discouraged.
3668 * ``static ConstantInt* get(const APInt& Val)``: Returns the ConstantInt
3669 object that represents the value provided by ``Val``. The type is implied
3670 as the IntegerType that corresponds to the bit width of ``Val``.
3672 * ``static ConstantInt* get(const Type *Ty, uint64_t Val)``: Returns the
3673 ConstantInt object that represents the value provided by ``Val`` for integer
3676 * ConstantFP : This class represents a floating point constant.
3678 * ``double getValue() const``: Returns the underlying value of this constant.
3680 * ConstantArray : This represents a constant array.
3682 * ``const std::vector<Use> &getValues() const``: Returns a vector of
3683 component constants that makeup this array.
3685 * ConstantStruct : This represents a constant struct.
3687 * ``const std::vector<Use> &getValues() const``: Returns a vector of
3688 component constants that makeup this array.
3690 * GlobalValue : This represents either a global variable or a function. In
3691 either case, the value is a constant fixed address (after linking).
3695 The ``GlobalValue`` class
3696 -------------------------
3698 ``#include "llvm/IR/GlobalValue.h"``
3700 header source: `GlobalValue.h
3701 <http://llvm.org/doxygen/GlobalValue_8h-source.html>`_
3703 doxygen info: `GlobalValue Class
3704 <http://llvm.org/doxygen/classllvm_1_1GlobalValue.html>`_
3706 Superclasses: Constant_, User_, Value_
3708 Global values ( GlobalVariable_\ s or :ref:`Function <c_Function>`\ s) are the
3709 only LLVM values that are visible in the bodies of all :ref:`Function
3710 <c_Function>`\ s. Because they are visible at global scope, they are also
3711 subject to linking with other globals defined in different translation units.
3712 To control the linking process, ``GlobalValue``\ s know their linkage rules.
3713 Specifically, ``GlobalValue``\ s know whether they have internal or external
3714 linkage, as defined by the ``LinkageTypes`` enumeration.
3716 If a ``GlobalValue`` has internal linkage (equivalent to being ``static`` in C),
3717 it is not visible to code outside the current translation unit, and does not
3718 participate in linking. If it has external linkage, it is visible to external
3719 code, and does participate in linking. In addition to linkage information,
3720 ``GlobalValue``\ s keep track of which Module_ they are currently part of.
3722 Because ``GlobalValue``\ s are memory objects, they are always referred to by
3723 their **address**. As such, the Type_ of a global is always a pointer to its
3724 contents. It is important to remember this when using the ``GetElementPtrInst``
3725 instruction because this pointer must be dereferenced first. For example, if
3726 you have a ``GlobalVariable`` (a subclass of ``GlobalValue)`` that is an array
3727 of 24 ints, type ``[24 x i32]``, then the ``GlobalVariable`` is a pointer to
3728 that array. Although the address of the first element of this array and the
3729 value of the ``GlobalVariable`` are the same, they have different types. The
3730 ``GlobalVariable``'s type is ``[24 x i32]``. The first element's type is
3731 ``i32.`` Because of this, accessing a global value requires you to dereference
3732 the pointer with ``GetElementPtrInst`` first, then its elements can be accessed.
3733 This is explained in the `LLVM Language Reference Manual
3734 <LangRef.html#globalvars>`_.
3738 Important Public Members of the ``GlobalValue`` class
3739 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3741 * | ``bool hasInternalLinkage() const``
3742 | ``bool hasExternalLinkage() const``
3743 | ``void setInternalLinkage(bool HasInternalLinkage)``
3745 These methods manipulate the linkage characteristics of the ``GlobalValue``.
3747 * ``Module *getParent()``
3749 This returns the Module_ that the
3750 GlobalValue is currently embedded into.
3754 The ``Function`` class
3755 ----------------------
3757 ``#include "llvm/IR/Function.h"``
3759 header source: `Function.h <http://llvm.org/doxygen/Function_8h-source.html>`_
3761 doxygen info: `Function Class
3762 <http://llvm.org/doxygen/classllvm_1_1Function.html>`_
3764 Superclasses: GlobalValue_, Constant_, User_, Value_
3766 The ``Function`` class represents a single procedure in LLVM. It is actually
3767 one of the more complex classes in the LLVM hierarchy because it must keep track
3768 of a large amount of data. The ``Function`` class keeps track of a list of
3769 BasicBlock_\ s, a list of formal Argument_\ s, and a SymbolTable_.
3771 The list of BasicBlock_\ s is the most commonly used part of ``Function``
3772 objects. The list imposes an implicit ordering of the blocks in the function,
3773 which indicate how the code will be laid out by the backend. Additionally, the
3774 first BasicBlock_ is the implicit entry node for the ``Function``. It is not
3775 legal in LLVM to explicitly branch to this initial block. There are no implicit
3776 exit nodes, and in fact there may be multiple exit nodes from a single
3777 ``Function``. If the BasicBlock_ list is empty, this indicates that the
3778 ``Function`` is actually a function declaration: the actual body of the function
3779 hasn't been linked in yet.
3781 In addition to a list of BasicBlock_\ s, the ``Function`` class also keeps track
3782 of the list of formal Argument_\ s that the function receives. This container
3783 manages the lifetime of the Argument_ nodes, just like the BasicBlock_ list does
3784 for the BasicBlock_\ s.
3786 The SymbolTable_ is a very rarely used LLVM feature that is only used when you
3787 have to look up a value by name. Aside from that, the SymbolTable_ is used
3788 internally to make sure that there are not conflicts between the names of
3789 Instruction_\ s, BasicBlock_\ s, or Argument_\ s in the function body.
3791 Note that ``Function`` is a GlobalValue_ and therefore also a Constant_. The
3792 value of the function is its address (after linking) which is guaranteed to be
3797 Important Public Members of the ``Function``
3798 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3800 * ``Function(const FunctionType *Ty, LinkageTypes Linkage,
3801 const std::string &N = "", Module* Parent = 0)``
3803 Constructor used when you need to create new ``Function``\ s to add the
3804 program. The constructor must specify the type of the function to create and
3805 what type of linkage the function should have. The FunctionType_ argument
3806 specifies the formal arguments and return value for the function. The same
3807 FunctionType_ value can be used to create multiple functions. The ``Parent``
3808 argument specifies the Module in which the function is defined. If this
3809 argument is provided, the function will automatically be inserted into that
3810 module's list of functions.
3812 * ``bool isDeclaration()``
3814 Return whether or not the ``Function`` has a body defined. If the function is
3815 "external", it does not have a body, and thus must be resolved by linking with
3816 a function defined in a different translation unit.
3818 * | ``Function::iterator`` - Typedef for basic block list iterator
3819 | ``Function::const_iterator`` - Typedef for const_iterator.
3820 | ``begin()``, ``end()``, ``size()``, ``empty()``
3822 These are forwarding methods that make it easy to access the contents of a
3823 ``Function`` object's BasicBlock_ list.
3825 * ``Function::BasicBlockListType &getBasicBlockList()``
3827 Returns the list of BasicBlock_\ s. This is necessary to use when you need to
3828 update the list or perform a complex action that doesn't have a forwarding
3831 * | ``Function::arg_iterator`` - Typedef for the argument list iterator
3832 | ``Function::const_arg_iterator`` - Typedef for const_iterator.
3833 | ``arg_begin()``, ``arg_end()``, ``arg_size()``, ``arg_empty()``
3835 These are forwarding methods that make it easy to access the contents of a
3836 ``Function`` object's Argument_ list.
3838 * ``Function::ArgumentListType &getArgumentList()``
3840 Returns the list of Argument_. This is necessary to use when you need to
3841 update the list or perform a complex action that doesn't have a forwarding
3844 * ``BasicBlock &getEntryBlock()``
3846 Returns the entry ``BasicBlock`` for the function. Because the entry block
3847 for the function is always the first block, this returns the first block of
3850 * | ``Type *getReturnType()``
3851 | ``FunctionType *getFunctionType()``
3853 This traverses the Type_ of the ``Function`` and returns the return type of
3854 the function, or the FunctionType_ of the actual function.
3856 * ``SymbolTable *getSymbolTable()``
3858 Return a pointer to the SymbolTable_ for this ``Function``.
3862 The ``GlobalVariable`` class
3863 ----------------------------
3865 ``#include "llvm/IR/GlobalVariable.h"``
3867 header source: `GlobalVariable.h
3868 <http://llvm.org/doxygen/GlobalVariable_8h-source.html>`_
3870 doxygen info: `GlobalVariable Class
3871 <http://llvm.org/doxygen/classllvm_1_1GlobalVariable.html>`_
3873 Superclasses: GlobalValue_, Constant_, User_, Value_
3875 Global variables are represented with the (surprise surprise) ``GlobalVariable``
3876 class. Like functions, ``GlobalVariable``\ s are also subclasses of
3877 GlobalValue_, and as such are always referenced by their address (global values
3878 must live in memory, so their "name" refers to their constant address). See
3879 GlobalValue_ for more on this. Global variables may have an initial value
3880 (which must be a Constant_), and if they have an initializer, they may be marked
3881 as "constant" themselves (indicating that their contents never change at
3884 .. _m_GlobalVariable:
3886 Important Public Members of the ``GlobalVariable`` class
3887 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3889 * ``GlobalVariable(const Type *Ty, bool isConstant, LinkageTypes &Linkage,
3890 Constant *Initializer = 0, const std::string &Name = "", Module* Parent = 0)``
3892 Create a new global variable of the specified type. If ``isConstant`` is true
3893 then the global variable will be marked as unchanging for the program. The
3894 Linkage parameter specifies the type of linkage (internal, external, weak,
3895 linkonce, appending) for the variable. If the linkage is InternalLinkage,
3896 WeakAnyLinkage, WeakODRLinkage, LinkOnceAnyLinkage or LinkOnceODRLinkage, then
3897 the resultant global variable will have internal linkage. AppendingLinkage
3898 concatenates together all instances (in different translation units) of the
3899 variable into a single variable but is only applicable to arrays. See the
3900 `LLVM Language Reference <LangRef.html#modulestructure>`_ for further details
3901 on linkage types. Optionally an initializer, a name, and the module to put
3902 the variable into may be specified for the global variable as well.
3904 * ``bool isConstant() const``
3906 Returns true if this is a global variable that is known not to be modified at
3909 * ``bool hasInitializer()``
3911 Returns true if this ``GlobalVariable`` has an intializer.
3913 * ``Constant *getInitializer()``
3915 Returns the initial value for a ``GlobalVariable``. It is not legal to call
3916 this method if there is no initializer.
3920 The ``BasicBlock`` class
3921 ------------------------
3923 ``#include "llvm/IR/BasicBlock.h"``
3925 header source: `BasicBlock.h
3926 <http://llvm.org/doxygen/BasicBlock_8h-source.html>`_
3928 doxygen info: `BasicBlock Class
3929 <http://llvm.org/doxygen/classllvm_1_1BasicBlock.html>`_
3933 This class represents a single entry single exit section of the code, commonly
3934 known as a basic block by the compiler community. The ``BasicBlock`` class
3935 maintains a list of Instruction_\ s, which form the body of the block. Matching
3936 the language definition, the last element of this list of instructions is always
3937 a terminator instruction (a subclass of the TerminatorInst_ class).
3939 In addition to tracking the list of instructions that make up the block, the
3940 ``BasicBlock`` class also keeps track of the :ref:`Function <c_Function>` that
3941 it is embedded into.
3943 Note that ``BasicBlock``\ s themselves are Value_\ s, because they are
3944 referenced by instructions like branches and can go in the switch tables.
3945 ``BasicBlock``\ s have type ``label``.
3949 Important Public Members of the ``BasicBlock`` class
3950 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3952 * ``BasicBlock(const std::string &Name = "", Function *Parent = 0)``
3954 The ``BasicBlock`` constructor is used to create new basic blocks for
3955 insertion into a function. The constructor optionally takes a name for the
3956 new block, and a :ref:`Function <c_Function>` to insert it into. If the
3957 ``Parent`` parameter is specified, the new ``BasicBlock`` is automatically
3958 inserted at the end of the specified :ref:`Function <c_Function>`, if not
3959 specified, the BasicBlock must be manually inserted into the :ref:`Function
3962 * | ``BasicBlock::iterator`` - Typedef for instruction list iterator
3963 | ``BasicBlock::const_iterator`` - Typedef for const_iterator.
3964 | ``begin()``, ``end()``, ``front()``, ``back()``,
3965 ``size()``, ``empty()``
3966 STL-style functions for accessing the instruction list.
3968 These methods and typedefs are forwarding functions that have the same
3969 semantics as the standard library methods of the same names. These methods
3970 expose the underlying instruction list of a basic block in a way that is easy
3971 to manipulate. To get the full complement of container operations (including
3972 operations to update the list), you must use the ``getInstList()`` method.
3974 * ``BasicBlock::InstListType &getInstList()``
3976 This method is used to get access to the underlying container that actually
3977 holds the Instructions. This method must be used when there isn't a
3978 forwarding function in the ``BasicBlock`` class for the operation that you
3979 would like to perform. Because there are no forwarding functions for
3980 "updating" operations, you need to use this if you want to update the contents
3981 of a ``BasicBlock``.
3983 * ``Function *getParent()``
3985 Returns a pointer to :ref:`Function <c_Function>` the block is embedded into,
3986 or a null pointer if it is homeless.
3988 * ``TerminatorInst *getTerminator()``
3990 Returns a pointer to the terminator instruction that appears at the end of the
3991 ``BasicBlock``. If there is no terminator instruction, or if the last
3992 instruction in the block is not a terminator, then a null pointer is returned.
3996 The ``Argument`` class
3997 ----------------------
3999 This subclass of Value defines the interface for incoming formal arguments to a
4000 function. A Function maintains a list of its formal arguments. An argument has
4001 a pointer to the parent Function.