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 <https://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 <https://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.stroustrup.com/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 <https://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
167 ``isa_and_nonnull<>``:
168 The ``isa_and_nonnull<>`` operator works just like the ``isa<>`` operator,
169 except that it allows for a null pointer as an argument (which it then
170 returns false). This can sometimes be useful, allowing you to combine several
171 null checks into one.
174 The ``cast_or_null<>`` operator works just like the ``cast<>`` operator,
175 except that it allows for a null pointer as an argument (which it then
176 propagates). This can sometimes be useful, allowing you to combine several
177 null checks into one.
179 ``dyn_cast_or_null<>``:
180 The ``dyn_cast_or_null<>`` operator works just like the ``dyn_cast<>``
181 operator, except that it allows for a null pointer as an argument (which it
182 then propagates). This can sometimes be useful, allowing you to combine
183 several null checks into one.
185 These five templates can be used with any classes, whether they have a v-table
186 or not. If you want to add support for these templates, see the document
187 :doc:`How to set up LLVM-style RTTI for your class hierarchy
188 <HowToSetUpLLVMStyleRTTI>`
192 Passing strings (the ``StringRef`` and ``Twine`` classes)
193 ---------------------------------------------------------
195 Although LLVM generally does not do much string manipulation, we do have several
196 important APIs which take strings. Two important examples are the Value class
197 -- which has names for instructions, functions, etc. -- and the ``StringMap``
198 class which is used extensively in LLVM and Clang.
200 These are generic classes, and they need to be able to accept strings which may
201 have embedded null characters. Therefore, they cannot simply take a ``const
202 char *``, and taking a ``const std::string&`` requires clients to perform a heap
203 allocation which is usually unnecessary. Instead, many LLVM APIs use a
204 ``StringRef`` or a ``const Twine&`` for passing strings efficiently.
208 The ``StringRef`` class
209 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
211 The ``StringRef`` data type represents a reference to a constant string (a
212 character array and a length) and supports the common operations available on
213 ``std::string``, but does not require heap allocation.
215 It can be implicitly constructed using a C style null-terminated string, an
216 ``std::string``, or explicitly with a character pointer and length. For
217 example, the ``StringRef`` find function is declared as:
221 iterator find(StringRef Key);
223 and clients can call it using any one of:
227 Map.find("foo"); // Lookup "foo"
228 Map.find(std::string("bar")); // Lookup "bar"
229 Map.find(StringRef("\0baz", 4)); // Lookup "\0baz"
231 Similarly, APIs which need to return a string may return a ``StringRef``
232 instance, which can be used directly or converted to an ``std::string`` using
233 the ``str`` member function. See ``llvm/ADT/StringRef.h`` (`doxygen
234 <https://llvm.org/doxygen/StringRef_8h_source.html>`__) for more
237 You should rarely use the ``StringRef`` class directly, because it contains
238 pointers to external memory it is not generally safe to store an instance of the
239 class (unless you know that the external storage will not be freed).
240 ``StringRef`` is small and pervasive enough in LLVM that it should always be
246 The ``Twine`` (`doxygen <https://llvm.org/doxygen/classllvm_1_1Twine.html>`__)
247 class is an efficient way for APIs to accept concatenated strings. For example,
248 a common LLVM paradigm is to name one instruction based on the name of another
249 instruction with a suffix, for example:
253 New = CmpInst::Create(..., SO->getName() + ".cmp");
255 The ``Twine`` class is effectively a lightweight `rope
256 <http://en.wikipedia.org/wiki/Rope_(computer_science)>`_ which points to
257 temporary (stack allocated) objects. Twines can be implicitly constructed as
258 the result of the plus operator applied to strings (i.e., a C strings, an
259 ``std::string``, or a ``StringRef``). The twine delays the actual concatenation
260 of strings until it is actually required, at which point it can be efficiently
261 rendered directly into a character array. This avoids unnecessary heap
262 allocation involved in constructing the temporary results of string
263 concatenation. See ``llvm/ADT/Twine.h`` (`doxygen
264 <https://llvm.org/doxygen/Twine_8h_source.html>`__) and :ref:`here <dss_twine>`
265 for more information.
267 As with a ``StringRef``, ``Twine`` objects point to external memory and should
268 almost never be stored or mentioned directly. They are intended solely for use
269 when defining a function which should be able to efficiently accept concatenated
272 .. _formatting_strings:
274 Formatting strings (the ``formatv`` function)
275 ---------------------------------------------
276 While LLVM doesn't necessarily do a lot of string manipulation and parsing, it
277 does do a lot of string formatting. From diagnostic messages, to llvm tool
278 outputs such as ``llvm-readobj`` to printing verbose disassembly listings and
279 LLDB runtime logging, the need for string formatting is pervasive.
281 The ``formatv`` is similar in spirit to ``printf``, but uses a different syntax
282 which borrows heavily from Python and C#. Unlike ``printf`` it deduces the type
283 to be formatted at compile time, so it does not need a format specifier such as
284 ``%d``. This reduces the mental overhead of trying to construct portable format
285 strings, especially for platform-specific types like ``size_t`` or pointer types.
286 Unlike both ``printf`` and Python, it additionally fails to compile if LLVM does
287 not know how to format the type. These two properties ensure that the function
288 is both safer and simpler to use than traditional formatting methods such as
289 the ``printf`` family of functions.
294 A call to ``formatv`` involves a single **format string** consisting of 0 or more
295 **replacement sequences**, followed by a variable length list of **replacement values**.
296 A replacement sequence is a string of the form ``{N[[,align]:style]}``.
298 ``N`` refers to the 0-based index of the argument from the list of replacement
299 values. Note that this means it is possible to reference the same parameter
300 multiple times, possibly with different style and/or alignment options, in any order.
302 ``align`` is an optional string specifying the width of the field to format
303 the value into, and the alignment of the value within the field. It is specified as
304 an optional **alignment style** followed by a positive integral **field width**. The
305 alignment style can be one of the characters ``-`` (left align), ``=`` (center align),
306 or ``+`` (right align). The default is right aligned.
308 ``style`` is an optional string consisting of a type specific that controls the
309 formatting of the value. For example, to format a floating point value as a percentage,
310 you can use the style option ``P``.
315 There are two ways to customize the formatting behavior for a type.
317 1. Provide a template specialization of ``llvm::format_provider<T>`` for your
318 type ``T`` with the appropriate static format method.
324 struct format_provider<MyFooBar> {
325 static void format(const MyFooBar &V, raw_ostream &Stream, StringRef Style) {
326 // Do whatever is necessary to format `V` into `Stream`
331 std::string S = formatv("{0}", X);
335 This is a useful extensibility mechanism for adding support for formatting your own
336 custom types with your own custom Style options. But it does not help when you want
337 to extend the mechanism for formatting a type that the library already knows how to
338 format. For that, we need something else.
340 2. Provide a **format adapter** inheriting from ``llvm::FormatAdapter<T>``.
345 struct format_int_custom : public llvm::FormatAdapter<int> {
346 explicit format_int_custom(int N) : llvm::FormatAdapter<int>(N) {}
347 void format(llvm::raw_ostream &Stream, StringRef Style) override {
348 // Do whatever is necessary to format ``this->Item`` into ``Stream``
354 std::string S = formatv("{0}", anything::format_int_custom(42));
358 If the type is detected to be derived from ``FormatAdapter<T>``, ``formatv``
360 ``format`` method on the argument passing in the specified style. This allows
361 one to provide custom formatting of any type, including one which already has
362 a builtin format provider.
366 Below is intended to provide an incomplete set of examples demonstrating
367 the usage of ``formatv``. More information can be found by reading the
368 doxygen documentation or by looking at the unit test suite.
374 // Simple formatting of basic types and implicit string conversion.
375 S = formatv("{0} ({1:P})", 7, 0.35); // S == "7 (35.00%)"
377 // Out-of-order referencing and multi-referencing
378 outs() << formatv("{0} {2} {1} {0}", 1, "test", 3); // prints "1 3 test 1"
380 // Left, right, and center alignment
381 S = formatv("{0,7}", 'a'); // S == " a";
382 S = formatv("{0,-7}", 'a'); // S == "a ";
383 S = formatv("{0,=7}", 'a'); // S == " a ";
384 S = formatv("{0,+7}", 'a'); // S == " a";
387 S = formatv("{0:N} - {0:x} - {1:E}", 12345, 123908342); // S == "12,345 - 0x3039 - 1.24E8"
390 S = formatv("{0}", fmt_align(42, AlignStyle::Center, 7)); // S == " 42 "
391 S = formatv("{0}", fmt_repeat("hi", 3)); // S == "hihihi"
392 S = formatv("{0}", fmt_pad("hi", 2, 6)); // S == " hi "
395 std::vector<int> V = {8, 9, 10};
396 S = formatv("{0}", make_range(V.begin(), V.end())); // S == "8, 9, 10"
397 S = formatv("{0:$[+]}", make_range(V.begin(), V.end())); // S == "8+9+10"
398 S = formatv("{0:$[ + ]@[x]}", make_range(V.begin(), V.end())); // S == "0x8 + 0x9 + 0xA"
405 Proper error handling helps us identify bugs in our code, and helps end-users
406 understand errors in their tool usage. Errors fall into two broad categories:
407 *programmatic* and *recoverable*, with different strategies for handling and
413 Programmatic errors are violations of program invariants or API contracts, and
414 represent bugs within the program itself. Our aim is to document invariants, and
415 to abort quickly at the point of failure (providing some basic diagnostic) when
416 invariants are broken at runtime.
418 The fundamental tools for handling programmatic errors are assertions and the
419 llvm_unreachable function. Assertions are used to express invariant conditions,
420 and should include a message describing the invariant:
424 assert(isPhysReg(R) && "All virt regs should have been allocated already.");
426 The llvm_unreachable function can be used to document areas of control flow
427 that should never be entered if the program invariants hold:
431 enum { Foo, Bar, Baz } X = foo();
434 case Foo: /* Handle Foo */; break;
435 case Bar: /* Handle Bar */; break;
437 llvm_unreachable("X should be Foo or Bar here");
443 Recoverable errors represent an error in the program's environment, for example
444 a resource failure (a missing file, a dropped network connection, etc.), or
445 malformed input. These errors should be detected and communicated to a level of
446 the program where they can be handled appropriately. Handling the error may be
447 as simple as reporting the issue to the user, or it may involve attempts at
452 While it would be ideal to use this error handling scheme throughout
453 LLVM, there are places where this hasn't been practical to apply. In
454 situations where you absolutely must emit a non-programmatic error and
455 the ``Error`` model isn't workable you can call ``report_fatal_error``,
456 which will call installed error handlers, print a message, and abort the
457 program. The use of `report_fatal_error` in this case is discouraged.
459 Recoverable errors are modeled using LLVM's ``Error`` scheme. This scheme
460 represents errors using function return values, similar to classic C integer
461 error codes, or C++'s ``std::error_code``. However, the ``Error`` class is
462 actually a lightweight wrapper for user-defined error types, allowing arbitrary
463 information to be attached to describe the error. This is similar to the way C++
464 exceptions allow throwing of user-defined types.
466 Success values are created by calling ``Error::success()``, E.g.:
473 return Error::success();
476 Success values are very cheap to construct and return - they have minimal
477 impact on program performance.
479 Failure values are constructed using ``make_error<T>``, where ``T`` is any class
480 that inherits from the ErrorInfo utility, E.g.:
484 class BadFileFormat : public ErrorInfo<BadFileFormat> {
489 BadFileFormat(StringRef Path) : Path(Path.str()) {}
491 void log(raw_ostream &OS) const override {
492 OS << Path << " is malformed";
495 std::error_code convertToErrorCode() const override {
496 return make_error_code(object_error::parse_failed);
500 char BadFileFormat::ID; // This should be declared in the C++ file.
502 Error printFormattedFile(StringRef Path) {
503 if (<check for valid format>)
504 return make_error<BadFileFormat>(Path);
505 // print file contents.
506 return Error::success();
509 Error values can be implicitly converted to bool: true for error, false for
510 success, enabling the following idiom:
517 if (auto Err = mayFail())
519 // Success! We can proceed.
522 For functions that can fail but need to return a value the ``Expected<T>``
523 utility can be used. Values of this type can be constructed with either a
524 ``T``, or an ``Error``. Expected<T> values are also implicitly convertible to
525 boolean, but with the opposite convention to ``Error``: true for success, false
526 for error. If success, the ``T`` value can be accessed via the dereference
527 operator. If failure, the ``Error`` value can be extracted using the
528 ``takeError()`` method. Idiomatic usage looks like:
532 Expected<FormattedFile> openFormattedFile(StringRef Path) {
533 // If badly formatted, return an error.
534 if (auto Err = checkFormat(Path))
535 return std::move(Err);
536 // Otherwise return a FormattedFile instance.
537 return FormattedFile(Path);
540 Error processFormattedFile(StringRef Path) {
541 // Try to open a formatted file
542 if (auto FileOrErr = openFormattedFile(Path)) {
543 // On success, grab a reference to the file and continue.
544 auto &File = *FileOrErr;
547 // On error, extract the Error value and return it.
548 return FileOrErr.takeError();
551 If an ``Expected<T>`` value is in success mode then the ``takeError()`` method
552 will return a success value. Using this fact, the above function can be
557 Error processFormattedFile(StringRef Path) {
558 // Try to open a formatted file
559 auto FileOrErr = openFormattedFile(Path);
560 if (auto Err = FileOrErr.takeError())
561 // On error, extract the Error value and return it.
563 // On success, grab a reference to the file and continue.
564 auto &File = *FileOrErr;
568 This second form is often more readable for functions that involve multiple
569 ``Expected<T>`` values as it limits the indentation required.
571 If an ``Expected<T>`` value will be moved into an existing variable then the
572 ``moveInto()`` method avoids the need to name an extra variable. This is
573 useful to enable ``operator->()`` the ``Expected<T>`` value has pointer-like
574 semantics. For example:
578 Expected<std::unique_ptr<MemoryBuffer>> openBuffer(StringRef Path);
579 Error processBuffer(StringRef Buffer);
581 Error processBufferAtPath(StringRef Path) {
582 // Try to open a buffer.
583 std::unique_ptr<MemoryBuffer> MB;
584 if (auto Err = openBuffer(Path).moveInto(MB))
585 // On error, return the Error value.
587 // On success, use MB.
588 return processContent(MB->getBuffer());
591 This third form works with any type that can be assigned to from ``T&&``. This
592 can be useful if the ``Expected<T>`` value needs to be stored an already-declared
593 ``Optional<T>``. For example:
597 Expected<StringRef> extractClassName(StringRef Definition);
599 StringRef Definition;
600 Optional<StringRef> LazyName;
603 if (auto Err = extractClassName(Path).moveInto(LazyName))
604 // On error, return the Error value.
606 // On success, LazyName has been initialized.
611 All ``Error`` instances, whether success or failure, must be either checked or
612 moved from (via ``std::move`` or a return) before they are destructed.
613 Accidentally discarding an unchecked error will cause a program abort at the
614 point where the unchecked value's destructor is run, making it easy to identify
615 and fix violations of this rule.
617 Success values are considered checked once they have been tested (by invoking
618 the boolean conversion operator):
622 if (auto Err = mayFail(...))
623 return Err; // Failure value - move error to caller.
625 // Safe to continue: Err was checked.
627 In contrast, the following code will always cause an abort, even if ``mayFail``
628 returns a success value:
633 // Program will always abort here, even if mayFail() returns Success, since
634 // the value is not checked.
636 Failure values are considered checked once a handler for the error type has
642 processFormattedFile(...),
643 [](const BadFileFormat &BFF) {
644 report("Unable to process " + BFF.Path + ": bad format");
646 [](const FileNotFound &FNF) {
647 report("File not found " + FNF.Path);
650 The ``handleErrors`` function takes an error as its first argument, followed by
651 a variadic list of "handlers", each of which must be a callable type (a
652 function, lambda, or class with a call operator) with one argument. The
653 ``handleErrors`` function will visit each handler in the sequence and check its
654 argument type against the dynamic type of the error, running the first handler
655 that matches. This is the same decision process that is used decide which catch
656 clause to run for a C++ exception.
658 Since the list of handlers passed to ``handleErrors`` may not cover every error
659 type that can occur, the ``handleErrors`` function also returns an Error value
660 that must be checked or propagated. If the error value that is passed to
661 ``handleErrors`` does not match any of the handlers it will be returned from
662 handleErrors. Idiomatic use of ``handleErrors`` thus looks like:
668 processFormattedFile(...),
669 [](const BadFileFormat &BFF) {
670 report("Unable to process " + BFF.Path + ": bad format");
672 [](const FileNotFound &FNF) {
673 report("File not found " + FNF.Path);
677 In cases where you truly know that the handler list is exhaustive the
678 ``handleAllErrors`` function can be used instead. This is identical to
679 ``handleErrors`` except that it will terminate the program if an unhandled
680 error is passed in, and can therefore return void. The ``handleAllErrors``
681 function should generally be avoided: the introduction of a new error type
682 elsewhere in the program can easily turn a formerly exhaustive list of errors
683 into a non-exhaustive list, risking unexpected program termination. Where
684 possible, use handleErrors and propagate unknown errors up the stack instead.
686 For tool code, where errors can be handled by printing an error message then
687 exiting with an error code, the :ref:`ExitOnError <err_exitonerr>` utility
688 may be a better choice than handleErrors, as it simplifies control flow when
689 calling fallible functions.
691 In situations where it is known that a particular call to a fallible function
692 will always succeed (for example, a call to a function that can only fail on a
693 subset of inputs with an input that is known to be safe) the
694 :ref:`cantFail <err_cantfail>` functions can be used to remove the error type,
695 simplifying control flow.
700 Many kinds of errors have no recovery strategy, the only action that can be
701 taken is to report them to the user so that the user can attempt to fix the
702 environment. In this case representing the error as a string makes perfect
703 sense. LLVM provides the ``StringError`` class for this purpose. It takes two
704 arguments: A string error message, and an equivalent ``std::error_code`` for
705 interoperability. It also provides a ``createStringError`` function to simplify
706 common usage of this class:
710 // These two lines of code are equivalent:
711 make_error<StringError>("Bad executable", errc::executable_format_error);
712 createStringError(errc::executable_format_error, "Bad executable");
714 If you're certain that the error you're building will never need to be converted
715 to a ``std::error_code`` you can use the ``inconvertibleErrorCode()`` function:
719 createStringError(inconvertibleErrorCode(), "Bad executable");
721 This should be done only after careful consideration. If any attempt is made to
722 convert this error to a ``std::error_code`` it will trigger immediate program
723 termination. Unless you are certain that your errors will not need
724 interoperability you should look for an existing ``std::error_code`` that you
725 can convert to, and even (as painful as it is) consider introducing a new one as
728 ``createStringError`` can take ``printf`` style format specifiers to provide a
733 createStringError(errc::executable_format_error,
734 "Bad executable: %s", FileName);
736 Interoperability with std::error_code and ErrorOr
737 """""""""""""""""""""""""""""""""""""""""""""""""
739 Many existing LLVM APIs use ``std::error_code`` and its partner ``ErrorOr<T>``
740 (which plays the same role as ``Expected<T>``, but wraps a ``std::error_code``
741 rather than an ``Error``). The infectious nature of error types means that an
742 attempt to change one of these functions to return ``Error`` or ``Expected<T>``
743 instead often results in an avalanche of changes to callers, callers of callers,
744 and so on. (The first such attempt, returning an ``Error`` from
745 MachOObjectFile's constructor, was abandoned after the diff reached 3000 lines,
746 impacted half a dozen libraries, and was still growing).
748 To solve this problem, the ``Error``/``std::error_code`` interoperability requirement was
749 introduced. Two pairs of functions allow any ``Error`` value to be converted to a
750 ``std::error_code``, any ``Expected<T>`` to be converted to an ``ErrorOr<T>``, and vice
755 std::error_code errorToErrorCode(Error Err);
756 Error errorCodeToError(std::error_code EC);
758 template <typename T> ErrorOr<T> expectedToErrorOr(Expected<T> TOrErr);
759 template <typename T> Expected<T> errorOrToExpected(ErrorOr<T> TOrEC);
762 Using these APIs it is easy to make surgical patches that update individual
763 functions from ``std::error_code`` to ``Error``, and from ``ErrorOr<T>`` to
766 Returning Errors from error handlers
767 """"""""""""""""""""""""""""""""""""
769 Error recovery attempts may themselves fail. For that reason, ``handleErrors``
770 actually recognises three different forms of handler signature:
774 // Error must be handled, no new errors produced:
775 void(UserDefinedError &E);
777 // Error must be handled, new errors can be produced:
778 Error(UserDefinedError &E);
780 // Original error can be inspected, then re-wrapped and returned (or a new
781 // error can be produced):
782 Error(std::unique_ptr<UserDefinedError> E);
784 Any error returned from a handler will be returned from the ``handleErrors``
785 function so that it can be handled itself, or propagated up the stack.
789 Using ExitOnError to simplify tool code
790 """""""""""""""""""""""""""""""""""""""
792 Library code should never call ``exit`` for a recoverable error, however in tool
793 code (especially command line tools) this can be a reasonable approach. Calling
794 ``exit`` upon encountering an error dramatically simplifies control flow as the
795 error no longer needs to be propagated up the stack. This allows code to be
796 written in straight-line style, as long as each fallible call is wrapped in a
797 check and call to exit. The ``ExitOnError`` class supports this pattern by
798 providing call operators that inspect ``Error`` values, stripping the error away
799 in the success case and logging to ``stderr`` then exiting in the failure case.
801 To use this class, declare a global ``ExitOnError`` variable in your program:
805 ExitOnError ExitOnErr;
807 Calls to fallible functions can then be wrapped with a call to ``ExitOnErr``,
808 turning them into non-failing calls:
813 Expected<int> mayFail2();
816 ExitOnErr(mayFail());
817 int X = ExitOnErr(mayFail2());
820 On failure, the error's log message will be written to ``stderr``, optionally
821 preceded by a string "banner" that can be set by calling the setBanner method. A
822 mapping can also be supplied from ``Error`` values to exit codes using the
823 ``setExitCodeMapper`` method:
827 int main(int argc, char *argv[]) {
828 ExitOnErr.setBanner(std::string(argv[0]) + " error:");
829 ExitOnErr.setExitCodeMapper(
830 [](const Error &Err) {
831 if (Err.isA<BadFileFormat>())
836 Use ``ExitOnError`` in your tool code where possible as it can greatly improve
841 Using cantFail to simplify safe callsites
842 """""""""""""""""""""""""""""""""""""""""
844 Some functions may only fail for a subset of their inputs, so calls using known
845 safe inputs can be assumed to succeed.
847 The cantFail functions encapsulate this by wrapping an assertion that their
848 argument is a success value and, in the case of Expected<T>, unwrapping the
853 Error onlyFailsForSomeXValues(int X);
854 Expected<int> onlyFailsForSomeXValues2(int X);
857 cantFail(onlyFailsForSomeXValues(KnownSafeValue));
858 int Y = cantFail(onlyFailsForSomeXValues2(KnownSafeValue));
862 Like the ExitOnError utility, cantFail simplifies control flow. Their treatment
863 of error cases is very different however: Where ExitOnError is guaranteed to
864 terminate the program on an error input, cantFail simply asserts that the result
865 is success. In debug builds this will result in an assertion failure if an error
866 is encountered. In release builds the behavior of cantFail for failure values is
867 undefined. As such, care must be taken in the use of cantFail: clients must be
868 certain that a cantFail wrapped call really can not fail with the given
871 Use of the cantFail functions should be rare in library code, but they are
872 likely to be of more use in tool and unit-test code where inputs and/or
873 mocked-up classes or functions may be known to be safe.
875 Fallible constructors
876 """""""""""""""""""""
878 Some classes require resource acquisition or other complex initialization that
879 can fail during construction. Unfortunately constructors can't return errors,
880 and having clients test objects after they're constructed to ensure that they're
881 valid is error prone as it's all too easy to forget the test. To work around
882 this, use the named constructor idiom and return an ``Expected<T>``:
889 static Expected<Foo> Create(Resource R1, Resource R2) {
890 Error Err = Error::success();
893 return std::move(Err);
899 Foo(Resource R1, Resource R2, Error &Err) {
900 ErrorAsOutParameter EAO(&Err);
901 if (auto Err2 = R1.acquire()) {
902 Err = std::move(Err2);
910 Here, the named constructor passes an ``Error`` by reference into the actual
911 constructor, which the constructor can then use to return errors. The
912 ``ErrorAsOutParameter`` utility sets the ``Error`` value's checked flag on entry
913 to the constructor so that the error can be assigned to, then resets it on exit
914 to force the client (the named constructor) to check the error.
916 By using this idiom, clients attempting to construct a Foo receive either a
917 well-formed Foo or an Error, never an object in an invalid state.
919 Propagating and consuming errors based on types
920 """""""""""""""""""""""""""""""""""""""""""""""
922 In some contexts, certain types of error are known to be benign. For example,
923 when walking an archive, some clients may be happy to skip over badly formatted
924 object files rather than terminating the walk immediately. Skipping badly
925 formatted objects could be achieved using an elaborate handler method, but the
926 Error.h header provides two utilities that make this idiom much cleaner: the
927 type inspection method, ``isA``, and the ``consumeError`` function:
931 Error walkArchive(Archive A) {
932 for (unsigned I = 0; I != A.numMembers(); ++I) {
933 auto ChildOrErr = A.getMember(I);
934 if (auto Err = ChildOrErr.takeError()) {
935 if (Err.isA<BadFileFormat>())
936 consumeError(std::move(Err))
940 auto &Child = *ChildOrErr;
944 return Error::success();
947 Concatenating Errors with joinErrors
948 """"""""""""""""""""""""""""""""""""
950 In the archive walking example above ``BadFileFormat`` errors are simply
951 consumed and ignored. If the client had wanted report these errors after
952 completing the walk over the archive they could use the ``joinErrors`` utility:
956 Error walkArchive(Archive A) {
957 Error DeferredErrs = Error::success();
958 for (unsigned I = 0; I != A.numMembers(); ++I) {
959 auto ChildOrErr = A.getMember(I);
960 if (auto Err = ChildOrErr.takeError())
961 if (Err.isA<BadFileFormat>())
962 DeferredErrs = joinErrors(std::move(DeferredErrs), std::move(Err));
965 auto &Child = *ChildOrErr;
972 The ``joinErrors`` routine builds a special error type called ``ErrorList``,
973 which holds a list of user defined errors. The ``handleErrors`` routine
974 recognizes this type and will attempt to handle each of the contained errors in
975 order. If all contained errors can be handled, ``handleErrors`` will return
976 ``Error::success()``, otherwise ``handleErrors`` will concatenate the remaining
977 errors and return the resulting ``ErrorList``.
979 Building fallible iterators and iterator ranges
980 """""""""""""""""""""""""""""""""""""""""""""""
982 The archive walking examples above retrieve archive members by index, however
983 this requires considerable boiler-plate for iteration and error checking. We can
984 clean this up by using the "fallible iterator" pattern, which supports the
985 following natural iteration idiom for fallible containers like Archive:
989 Error Err = Error::success();
990 for (auto &Child : Ar->children(Err)) {
991 // Use Child - only enter the loop when it's valid
993 // Allow early exit from the loop body, since we know that Err is success
994 // when we're inside the loop.
995 if (BailOutOn(Child))
1000 // Check Err after the loop to ensure it didn't break due to an error.
1004 To enable this idiom, iterators over fallible containers are written in a
1005 natural style, with their ``++`` and ``--`` operators replaced with fallible
1006 ``Error inc()`` and ``Error dec()`` functions. E.g.:
1010 class FallibleChildIterator {
1012 FallibleChildIterator(Archive &A, unsigned ChildIdx);
1013 Archive::Child &operator*();
1014 friend bool operator==(const ArchiveIterator &LHS,
1015 const ArchiveIterator &RHS);
1017 // operator++/operator-- replaced with fallible increment / decrement:
1019 if (!A.childValid(ChildIdx + 1))
1020 return make_error<BadArchiveMember>(...);
1022 return Error::success();
1028 Instances of this kind of fallible iterator interface are then wrapped with the
1029 fallible_iterator utility which provides ``operator++`` and ``operator--``,
1030 returning any errors via a reference passed in to the wrapper at construction
1031 time. The fallible_iterator wrapper takes care of (a) jumping to the end of the
1032 range on error, and (b) marking the error as checked whenever an iterator is
1033 compared to ``end`` and found to be inequal (in particular: this marks the
1034 error as checked throughout the body of a range-based for loop), enabling early
1035 exit from the loop without redundant error checking.
1037 Instances of the fallible iterator interface (e.g. FallibleChildIterator above)
1038 are wrapped using the ``make_fallible_itr`` and ``make_fallible_end``
1045 using child_iterator = fallible_iterator<FallibleChildIterator>;
1047 child_iterator child_begin(Error &Err) {
1048 return make_fallible_itr(FallibleChildIterator(*this, 0), Err);
1051 child_iterator child_end() {
1052 return make_fallible_end(FallibleChildIterator(*this, size()));
1055 iterator_range<child_iterator> children(Error &Err) {
1056 return make_range(child_begin(Err), child_end());
1060 Using the fallible_iterator utility allows for both natural construction of
1061 fallible iterators (using failing ``inc`` and ``dec`` operations) and
1062 relatively natural use of c++ iterator/loop idioms.
1066 More information on Error and its related utilities can be found in the
1067 Error.h header file.
1069 Passing functions and other callable objects
1070 --------------------------------------------
1072 Sometimes you may want a function to be passed a callback object. In order to
1073 support lambda expressions and other function objects, you should not use the
1074 traditional C approach of taking a function pointer and an opaque cookie:
1078 void takeCallback(bool (*Callback)(Function *, void *), void *Cookie);
1080 Instead, use one of the following approaches:
1085 If you don't mind putting the definition of your function into a header file,
1086 make it a function template that is templated on the callable type.
1090 template<typename Callable>
1091 void takeCallback(Callable Callback) {
1095 The ``function_ref`` class template
1096 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
1098 The ``function_ref``
1099 (`doxygen <https://llvm.org/doxygen/classllvm_1_1function__ref_3_01Ret_07Params_8_8_8_08_4.html>`__) class
1100 template represents a reference to a callable object, templated over the type
1101 of the callable. This is a good choice for passing a callback to a function,
1102 if you don't need to hold onto the callback after the function returns. In this
1103 way, ``function_ref`` is to ``std::function`` as ``StringRef`` is to
1106 ``function_ref<Ret(Param1, Param2, ...)>`` can be implicitly constructed from
1107 any callable object that can be called with arguments of type ``Param1``,
1108 ``Param2``, ..., and returns a value that can be converted to type ``Ret``.
1113 void visitBasicBlocks(Function *F, function_ref<bool (BasicBlock*)> Callback) {
1114 for (BasicBlock &BB : *F)
1119 can be called using:
1123 visitBasicBlocks(F, [&](BasicBlock *BB) {
1129 Note that a ``function_ref`` object contains pointers to external memory, so it
1130 is not generally safe to store an instance of the class (unless you know that
1131 the external storage will not be freed). If you need this ability, consider
1132 using ``std::function``. ``function_ref`` is small enough that it should always
1137 The ``LLVM_DEBUG()`` macro and ``-debug`` option
1138 ------------------------------------------------
1140 Often when working on your pass you will put a bunch of debugging printouts and
1141 other code into your pass. After you get it working, you want to remove it, but
1142 you may need it again in the future (to work out new bugs that you run across).
1144 Naturally, because of this, you don't want to delete the debug printouts, but
1145 you don't want them to always be noisy. A standard compromise is to comment
1146 them out, allowing you to enable them if you need them in the future.
1148 The ``llvm/Support/Debug.h`` (`doxygen
1149 <https://llvm.org/doxygen/Debug_8h_source.html>`__) file provides a macro named
1150 ``LLVM_DEBUG()`` that is a much nicer solution to this problem. Basically, you can
1151 put arbitrary code into the argument of the ``LLVM_DEBUG`` macro, and it is only
1152 executed if '``opt``' (or any other tool) is run with the '``-debug``' command
1157 LLVM_DEBUG(dbgs() << "I am here!\n");
1159 Then you can run your pass like this:
1161 .. code-block:: none
1163 $ opt < a.bc > /dev/null -mypass
1165 $ opt < a.bc > /dev/null -mypass -debug
1168 Using the ``LLVM_DEBUG()`` macro instead of a home-brewed solution allows you to not
1169 have to create "yet another" command line option for the debug output for your
1170 pass. Note that ``LLVM_DEBUG()`` macros are disabled for non-asserts builds, so they
1171 do not cause a performance impact at all (for the same reason, they should also
1172 not contain side-effects!).
1174 One additional nice thing about the ``LLVM_DEBUG()`` macro is that you can enable or
1175 disable it directly in gdb. Just use "``set DebugFlag=0``" or "``set
1176 DebugFlag=1``" from the gdb if the program is running. If the program hasn't
1177 been started yet, you can always just run it with ``-debug``.
1181 Fine grained debug info with ``DEBUG_TYPE`` and the ``-debug-only`` option
1182 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
1184 Sometimes you may find yourself in a situation where enabling ``-debug`` just
1185 turns on **too much** information (such as when working on the code generator).
1186 If you want to enable debug information with more fine-grained control, you
1187 should define the ``DEBUG_TYPE`` macro and use the ``-debug-only`` option as
1192 #define DEBUG_TYPE "foo"
1193 LLVM_DEBUG(dbgs() << "'foo' debug type\n");
1195 #define DEBUG_TYPE "bar"
1196 LLVM_DEBUG(dbgs() << "'bar' debug type\n");
1199 Then you can run your pass like this:
1201 .. code-block:: none
1203 $ opt < a.bc > /dev/null -mypass
1205 $ opt < a.bc > /dev/null -mypass -debug
1208 $ opt < a.bc > /dev/null -mypass -debug-only=foo
1210 $ opt < a.bc > /dev/null -mypass -debug-only=bar
1212 $ opt < a.bc > /dev/null -mypass -debug-only=foo,bar
1216 Of course, in practice, you should only set ``DEBUG_TYPE`` at the top of a file,
1217 to specify the debug type for the entire module. Be careful that you only do
1218 this after including Debug.h and not around any #include of headers. Also, you
1219 should use names more meaningful than "foo" and "bar", because there is no
1220 system in place to ensure that names do not conflict. If two different modules
1221 use the same string, they will all be turned on when the name is specified.
1222 This allows, for example, all debug information for instruction scheduling to be
1223 enabled with ``-debug-only=InstrSched``, even if the source lives in multiple
1224 files. The name must not include a comma (,) as that is used to separate the
1225 arguments of the ``-debug-only`` option.
1227 For performance reasons, -debug-only is not available in optimized build
1228 (``--enable-optimized``) of LLVM.
1230 The ``DEBUG_WITH_TYPE`` macro is also available for situations where you would
1231 like to set ``DEBUG_TYPE``, but only for one specific ``DEBUG`` statement. It
1232 takes an additional first parameter, which is the type to use. For example, the
1233 preceding example could be written as:
1237 DEBUG_WITH_TYPE("foo", dbgs() << "'foo' debug type\n");
1238 DEBUG_WITH_TYPE("bar", dbgs() << "'bar' debug type\n");
1242 The ``Statistic`` class & ``-stats`` option
1243 -------------------------------------------
1245 The ``llvm/ADT/Statistic.h`` (`doxygen
1246 <https://llvm.org/doxygen/Statistic_8h_source.html>`__) file provides a class
1247 named ``Statistic`` that is used as a unified way to keep track of what the LLVM
1248 compiler is doing and how effective various optimizations are. It is useful to
1249 see what optimizations are contributing to making a particular program run
1252 Often you may run your pass on some big program, and you're interested to see
1253 how many times it makes a certain transformation. Although you can do this with
1254 hand inspection, or some ad-hoc method, this is a real pain and not very useful
1255 for big programs. Using the ``Statistic`` class makes it very easy to keep
1256 track of this information, and the calculated information is presented in a
1257 uniform manner with the rest of the passes being executed.
1259 There are many examples of ``Statistic`` uses, but the basics of using it are as
1262 Define your statistic like this:
1266 #define DEBUG_TYPE "mypassname" // This goes before any #includes.
1267 STATISTIC(NumXForms, "The # of times I did stuff");
1269 The ``STATISTIC`` macro defines a static variable, whose name is specified by
1270 the first argument. The pass name is taken from the ``DEBUG_TYPE`` macro, and
1271 the description is taken from the second argument. The variable defined
1272 ("NumXForms" in this case) acts like an unsigned integer.
1274 Whenever you make a transformation, bump the counter:
1278 ++NumXForms; // I did stuff!
1280 That's all you have to do. To get '``opt``' to print out the statistics
1281 gathered, use the '``-stats``' option:
1283 .. code-block:: none
1285 $ opt -stats -mypassname < program.bc > /dev/null
1286 ... statistics output ...
1288 Note that in order to use the '``-stats``' option, LLVM must be
1289 compiled with assertions enabled.
1291 When running ``opt`` on a C file from the SPEC benchmark suite, it gives a
1292 report that looks like this:
1294 .. code-block:: none
1296 7646 bitcodewriter - Number of normal instructions
1297 725 bitcodewriter - Number of oversized instructions
1298 129996 bitcodewriter - Number of bitcode bytes written
1299 2817 raise - Number of insts DCEd or constprop'd
1300 3213 raise - Number of cast-of-self removed
1301 5046 raise - Number of expression trees converted
1302 75 raise - Number of other getelementptr's formed
1303 138 raise - Number of load/store peepholes
1304 42 deadtypeelim - Number of unused typenames removed from symtab
1305 392 funcresolve - Number of varargs functions resolved
1306 27 globaldce - Number of global variables removed
1307 2 adce - Number of basic blocks removed
1308 134 cee - Number of branches revectored
1309 49 cee - Number of setcc instruction eliminated
1310 532 gcse - Number of loads removed
1311 2919 gcse - Number of instructions removed
1312 86 indvars - Number of canonical indvars added
1313 87 indvars - Number of aux indvars removed
1314 25 instcombine - Number of dead inst eliminate
1315 434 instcombine - Number of insts combined
1316 248 licm - Number of load insts hoisted
1317 1298 licm - Number of insts hoisted to a loop pre-header
1318 3 licm - Number of insts hoisted to multiple loop preds (bad, no loop pre-header)
1319 75 mem2reg - Number of alloca's promoted
1320 1444 cfgsimplify - Number of blocks simplified
1322 Obviously, with so many optimizations, having a unified framework for this stuff
1323 is very nice. Making your pass fit well into the framework makes it more
1324 maintainable and useful.
1328 Adding debug counters to aid in debugging your code
1329 ---------------------------------------------------
1331 Sometimes, when writing new passes, or trying to track down bugs, it
1332 is useful to be able to control whether certain things in your pass
1333 happen or not. For example, there are times the minimization tooling
1334 can only easily give you large testcases. You would like to narrow
1335 your bug down to a specific transformation happening or not happening,
1336 automatically, using bisection. This is where debug counters help.
1337 They provide a framework for making parts of your code only execute a
1338 certain number of times.
1340 The ``llvm/Support/DebugCounter.h`` (`doxygen
1341 <https://llvm.org/doxygen/DebugCounter_8h_source.html>`__) file
1342 provides a class named ``DebugCounter`` that can be used to create
1343 command line counter options that control execution of parts of your code.
1345 Define your DebugCounter like this:
1349 DEBUG_COUNTER(DeleteAnInstruction, "passname-delete-instruction",
1350 "Controls which instructions get delete");
1352 The ``DEBUG_COUNTER`` macro defines a static variable, whose name
1353 is specified by the first argument. The name of the counter
1354 (which is used on the command line) is specified by the second
1355 argument, and the description used in the help is specified by the
1358 Whatever code you want that control, use ``DebugCounter::shouldExecute`` to control it.
1362 if (DebugCounter::shouldExecute(DeleteAnInstruction))
1363 I->eraseFromParent();
1365 That's all you have to do. Now, using opt, you can control when this code triggers using
1366 the '``--debug-counter``' option. There are two counters provided, ``skip`` and ``count``.
1367 ``skip`` is the number of times to skip execution of the codepath. ``count`` is the number
1368 of times, once we are done skipping, to execute the codepath.
1370 .. code-block:: none
1372 $ opt --debug-counter=passname-delete-instruction-skip=1,passname-delete-instruction-count=2 -passname
1374 This will skip the above code the first time we hit it, then execute it twice, then skip the rest of the executions.
1376 So if executed on the following code:
1378 .. code-block:: llvm
1385 It would delete number ``%2`` and ``%3``.
1387 A utility is provided in `utils/bisect-skip-count` to binary search
1388 skip and count arguments. It can be used to automatically minimize the
1389 skip and count for a debug-counter variable.
1393 Viewing graphs while debugging code
1394 -----------------------------------
1396 Several of the important data structures in LLVM are graphs: for example CFGs
1397 made out of LLVM :ref:`BasicBlocks <BasicBlock>`, CFGs made out of LLVM
1398 :ref:`MachineBasicBlocks <MachineBasicBlock>`, and :ref:`Instruction Selection
1399 DAGs <SelectionDAG>`. In many cases, while debugging various parts of the
1400 compiler, it is nice to instantly visualize these graphs.
1402 LLVM provides several callbacks that are available in a debug build to do
1403 exactly that. If you call the ``Function::viewCFG()`` method, for example, the
1404 current LLVM tool will pop up a window containing the CFG for the function where
1405 each basic block is a node in the graph, and each node contains the instructions
1406 in the block. Similarly, there also exists ``Function::viewCFGOnly()`` (does
1407 not include the instructions), the ``MachineFunction::viewCFG()`` and
1408 ``MachineFunction::viewCFGOnly()``, and the ``SelectionDAG::viewGraph()``
1409 methods. Within GDB, for example, you can usually use something like ``call
1410 DAG.viewGraph()`` to pop up a window. Alternatively, you can sprinkle calls to
1411 these functions in your code in places you want to debug.
1413 Getting this to work requires a small amount of setup. On Unix systems
1414 with X11, install the `graphviz <http://www.graphviz.org>`_ toolkit, and make
1415 sure 'dot' and 'gv' are in your path. If you are running on macOS, download
1416 and install the macOS `Graphviz program
1417 <http://www.pixelglow.com/graphviz/>`_ and add
1418 ``/Applications/Graphviz.app/Contents/MacOS/`` (or wherever you install it) to
1419 your path. The programs need not be present when configuring, building or
1420 running LLVM and can simply be installed when needed during an active debug
1423 ``SelectionDAG`` has been extended to make it easier to locate *interesting*
1424 nodes in large complex graphs. From gdb, if you ``call DAG.setGraphColor(node,
1425 "color")``, then the next ``call DAG.viewGraph()`` would highlight the node in
1426 the specified color (choices of colors can be found at `colors
1427 <http://www.graphviz.org/doc/info/colors.html>`_.) More complex node attributes
1428 can be provided with ``call DAG.setGraphAttrs(node, "attributes")`` (choices can
1429 be found at `Graph attributes <http://www.graphviz.org/doc/info/attrs.html>`_.)
1430 If you want to restart and clear all the current graph attributes, then you can
1431 ``call DAG.clearGraphAttrs()``.
1433 Note that graph visualization features are compiled out of Release builds to
1434 reduce file size. This means that you need a Debug+Asserts or Release+Asserts
1435 build to use these features.
1439 Picking the Right Data Structure for a Task
1440 ===========================================
1442 LLVM has a plethora of data structures in the ``llvm/ADT/`` directory, and we
1443 commonly use STL data structures. This section describes the trade-offs you
1444 should consider when you pick one.
1446 The first step is a choose your own adventure: do you want a sequential
1447 container, a set-like container, or a map-like container? The most important
1448 thing when choosing a container is the algorithmic properties of how you plan to
1449 access the container. Based on that, you should use:
1452 * a :ref:`map-like <ds_map>` container if you need efficient look-up of a
1453 value based on another value. Map-like containers also support efficient
1454 queries for containment (whether a key is in the map). Map-like containers
1455 generally do not support efficient reverse mapping (values to keys). If you
1456 need that, use two maps. Some map-like containers also support efficient
1457 iteration through the keys in sorted order. Map-like containers are the most
1458 expensive sort, only use them if you need one of these capabilities.
1460 * a :ref:`set-like <ds_set>` container if you need to put a bunch of stuff into
1461 a container that automatically eliminates duplicates. Some set-like
1462 containers support efficient iteration through the elements in sorted order.
1463 Set-like containers are more expensive than sequential containers.
1465 * a :ref:`sequential <ds_sequential>` container provides the most efficient way
1466 to add elements and keeps track of the order they are added to the collection.
1467 They permit duplicates and support efficient iteration, but do not support
1468 efficient look-up based on a key.
1470 * a :ref:`string <ds_string>` container is a specialized sequential container or
1471 reference structure that is used for character or byte arrays.
1473 * a :ref:`bit <ds_bit>` container provides an efficient way to store and
1474 perform set operations on sets of numeric id's, while automatically
1475 eliminating duplicates. Bit containers require a maximum of 1 bit for each
1476 identifier you want to store.
1478 Once the proper category of container is determined, you can fine tune the
1479 memory use, constant factors, and cache behaviors of access by intelligently
1480 picking a member of the category. Note that constant factors and cache behavior
1481 can be a big deal. If you have a vector that usually only contains a few
1482 elements (but could contain many), for example, it's much better to use
1483 :ref:`SmallVector <dss_smallvector>` than :ref:`vector <dss_vector>`. Doing so
1484 avoids (relatively) expensive malloc/free calls, which dwarf the cost of adding
1485 the elements to the container.
1489 Sequential Containers (std::vector, std::list, etc)
1490 ---------------------------------------------------
1492 There are a variety of sequential containers available for you, based on your
1493 needs. Pick the first in this section that will do what you want.
1500 The ``llvm::ArrayRef`` class is the preferred class to use in an interface that
1501 accepts a sequential list of elements in memory and just reads from them. By
1502 taking an ``ArrayRef``, the API can be passed a fixed size array, an
1503 ``std::vector``, an ``llvm::SmallVector`` and anything else that is contiguous
1506 .. _dss_fixedarrays:
1511 Fixed size arrays are very simple and very fast. They are good if you know
1512 exactly how many elements you have, or you have a (low) upper bound on how many
1517 Heap Allocated Arrays
1518 ^^^^^^^^^^^^^^^^^^^^^
1520 Heap allocated arrays (``new[]`` + ``delete[]``) are also simple. They are good
1521 if the number of elements is variable, if you know how many elements you will
1522 need before the array is allocated, and if the array is usually large (if not,
1523 consider a :ref:`SmallVector <dss_smallvector>`). The cost of a heap allocated
1524 array is the cost of the new/delete (aka malloc/free). Also note that if you
1525 are allocating an array of a type with a constructor, the constructor and
1526 destructors will be run for every element in the array (re-sizable vectors only
1527 construct those elements actually used).
1529 .. _dss_tinyptrvector:
1531 llvm/ADT/TinyPtrVector.h
1532 ^^^^^^^^^^^^^^^^^^^^^^^^
1534 ``TinyPtrVector<Type>`` is a highly specialized collection class that is
1535 optimized to avoid allocation in the case when a vector has zero or one
1536 elements. It has two major restrictions: 1) it can only hold values of pointer
1537 type, and 2) it cannot hold a null pointer.
1539 Since this container is highly specialized, it is rarely used.
1541 .. _dss_smallvector:
1543 llvm/ADT/SmallVector.h
1544 ^^^^^^^^^^^^^^^^^^^^^^
1546 ``SmallVector<Type, N>`` is a simple class that looks and smells just like
1547 ``vector<Type>``: it supports efficient iteration, lays out elements in memory
1548 order (so you can do pointer arithmetic between elements), supports efficient
1549 push_back/pop_back operations, supports efficient random access to its elements,
1552 The main advantage of SmallVector is that it allocates space for some number of
1553 elements (N) **in the object itself**. Because of this, if the SmallVector is
1554 dynamically smaller than N, no malloc is performed. This can be a big win in
1555 cases where the malloc/free call is far more expensive than the code that
1556 fiddles around with the elements.
1558 This is good for vectors that are "usually small" (e.g. the number of
1559 predecessors/successors of a block is usually less than 8). On the other hand,
1560 this makes the size of the SmallVector itself large, so you don't want to
1561 allocate lots of them (doing so will waste a lot of space). As such,
1562 SmallVectors are most useful when on the stack.
1564 In the absence of a well-motivated choice for the number of
1565 inlined elements ``N``, it is recommended to use ``SmallVector<T>`` (that is,
1566 omitting the ``N``). This will choose a default number of
1567 inlined elements reasonable for allocation on the stack (for example, trying
1568 to keep ``sizeof(SmallVector<T>)`` around 64 bytes).
1570 SmallVector also provides a nice portable and efficient replacement for
1573 SmallVector has grown a few other minor advantages over std::vector, causing
1574 ``SmallVector<Type, 0>`` to be preferred over ``std::vector<Type>``.
1576 #. std::vector is exception-safe, and some implementations have pessimizations
1577 that copy elements when SmallVector would move them.
1579 #. SmallVector understands ``std::is_trivially_copyable<Type>`` and uses realloc aggressively.
1581 #. Many LLVM APIs take a SmallVectorImpl as an out parameter (see the note
1584 #. SmallVector with N equal to 0 is smaller than std::vector on 64-bit
1585 platforms, since it uses ``unsigned`` (instead of ``void*``) for its size
1590 Prefer to use ``ArrayRef<T>`` or ``SmallVectorImpl<T>`` as a parameter type.
1592 It's rarely appropriate to use ``SmallVector<T, N>`` as a parameter type.
1593 If an API only reads from the vector, it should use :ref:`ArrayRef
1594 <dss_arrayref>`. Even if an API updates the vector the "small size" is
1595 unlikely to be relevant; such an API should use the ``SmallVectorImpl<T>``
1596 class, which is the "vector header" (and methods) without the elements
1597 allocated after it. Note that ``SmallVector<T, N>`` inherits from
1598 ``SmallVectorImpl<T>`` so the conversion is implicit and costs nothing. E.g.
1602 // DISCOURAGED: Clients cannot pass e.g. raw arrays.
1603 hardcodedContiguousStorage(const SmallVectorImpl<Foo> &In);
1604 // ENCOURAGED: Clients can pass any contiguous storage of Foo.
1605 allowsAnyContiguousStorage(ArrayRef<Foo> In);
1608 Foo Vec[] = { /* ... */ };
1609 hardcodedContiguousStorage(Vec); // Error.
1610 allowsAnyContiguousStorage(Vec); // Works.
1613 // DISCOURAGED: Clients cannot pass e.g. SmallVector<Foo, 8>.
1614 hardcodedSmallSize(SmallVector<Foo, 2> &Out);
1615 // ENCOURAGED: Clients can pass any SmallVector<Foo, N>.
1616 allowsAnySmallSize(SmallVectorImpl<Foo> &Out);
1619 SmallVector<Foo, 8> Vec;
1620 hardcodedSmallSize(Vec); // Error.
1621 allowsAnySmallSize(Vec); // Works.
1624 Even though it has "``Impl``" in the name, SmallVectorImpl is widely used
1625 and is no longer "private to the implementation". A name like
1626 ``SmallVectorHeader`` might be more appropriate.
1633 ``std::vector<T>`` is well loved and respected. However, ``SmallVector<T, 0>``
1634 is often a better option due to the advantages listed above. std::vector is
1635 still useful when you need to store more than ``UINT32_MAX`` elements or when
1636 interfacing with code that expects vectors :).
1638 One worthwhile note about std::vector: avoid code like this:
1647 Instead, write this as:
1657 Doing so will save (at least) one heap allocation and free per iteration of the
1665 ``std::deque`` is, in some senses, a generalized version of ``std::vector``.
1666 Like ``std::vector``, it provides constant time random access and other similar
1667 properties, but it also provides efficient access to the front of the list. It
1668 does not guarantee continuity of elements within memory.
1670 In exchange for this extra flexibility, ``std::deque`` has significantly higher
1671 constant factor costs than ``std::vector``. If possible, use ``std::vector`` or
1679 ``std::list`` is an extremely inefficient class that is rarely useful. It
1680 performs a heap allocation for every element inserted into it, thus having an
1681 extremely high constant factor, particularly for small data types.
1682 ``std::list`` also only supports bidirectional iteration, not random access
1685 In exchange for this high cost, std::list supports efficient access to both ends
1686 of the list (like ``std::deque``, but unlike ``std::vector`` or
1687 ``SmallVector``). In addition, the iterator invalidation characteristics of
1688 std::list are stronger than that of a vector class: inserting or removing an
1689 element into the list does not invalidate iterator or pointers to other elements
1697 ``ilist<T>`` implements an 'intrusive' doubly-linked list. It is intrusive,
1698 because it requires the element to store and provide access to the prev/next
1699 pointers for the list.
1701 ``ilist`` has the same drawbacks as ``std::list``, and additionally requires an
1702 ``ilist_traits`` implementation for the element type, but it provides some novel
1703 characteristics. In particular, it can efficiently store polymorphic objects,
1704 the traits class is informed when an element is inserted or removed from the
1705 list, and ``ilist``\ s are guaranteed to support a constant-time splice
1708 These properties are exactly what we want for things like ``Instruction``\ s and
1709 basic blocks, which is why these are implemented with ``ilist``\ s.
1711 Related classes of interest are explained in the following subsections:
1713 * :ref:`ilist_traits <dss_ilist_traits>`
1715 * :ref:`iplist <dss_iplist>`
1717 * :ref:`llvm/ADT/ilist_node.h <dss_ilist_node>`
1719 * :ref:`Sentinels <dss_ilist_sentinel>`
1721 .. _dss_packedvector:
1723 llvm/ADT/PackedVector.h
1724 ^^^^^^^^^^^^^^^^^^^^^^^
1726 Useful for storing a vector of values using only a few number of bits for each
1727 value. Apart from the standard operations of a vector-like container, it can
1728 also perform an 'or' set operation.
1736 FirstCondition = 0x1,
1737 SecondCondition = 0x2,
1742 PackedVector<State, 2> Vec1;
1743 Vec1.push_back(FirstCondition);
1745 PackedVector<State, 2> Vec2;
1746 Vec2.push_back(SecondCondition);
1749 return Vec1[0]; // returns 'Both'.
1752 .. _dss_ilist_traits:
1757 ``ilist_traits<T>`` is ``ilist<T>``'s customization mechanism. ``iplist<T>``
1758 (and consequently ``ilist<T>``) publicly derive from this traits class.
1765 ``iplist<T>`` is ``ilist<T>``'s base and as such supports a slightly narrower
1766 interface. Notably, inserters from ``T&`` are absent.
1768 ``ilist_traits<T>`` is a public base of this class and can be used for a wide
1769 variety of customizations.
1773 llvm/ADT/ilist_node.h
1774 ^^^^^^^^^^^^^^^^^^^^^
1776 ``ilist_node<T>`` implements the forward and backward links that are expected
1777 by the ``ilist<T>`` (and analogous containers) in the default manner.
1779 ``ilist_node<T>``\ s are meant to be embedded in the node type ``T``, usually
1780 ``T`` publicly derives from ``ilist_node<T>``.
1782 .. _dss_ilist_sentinel:
1787 ``ilist``\ s have another specialty that must be considered. To be a good
1788 citizen in the C++ ecosystem, it needs to support the standard container
1789 operations, such as ``begin`` and ``end`` iterators, etc. Also, the
1790 ``operator--`` must work correctly on the ``end`` iterator in the case of
1791 non-empty ``ilist``\ s.
1793 The only sensible solution to this problem is to allocate a so-called *sentinel*
1794 along with the intrusive list, which serves as the ``end`` iterator, providing
1795 the back-link to the last element. However conforming to the C++ convention it
1796 is illegal to ``operator++`` beyond the sentinel and it also must not be
1799 These constraints allow for some implementation freedom to the ``ilist`` how to
1800 allocate and store the sentinel. The corresponding policy is dictated by
1801 ``ilist_traits<T>``. By default a ``T`` gets heap-allocated whenever the need
1802 for a sentinel arises.
1804 While the default policy is sufficient in most cases, it may break down when
1805 ``T`` does not provide a default constructor. Also, in the case of many
1806 instances of ``ilist``\ s, the memory overhead of the associated sentinels is
1807 wasted. To alleviate the situation with numerous and voluminous
1808 ``T``-sentinels, sometimes a trick is employed, leading to *ghostly sentinels*.
1810 Ghostly sentinels are obtained by specially-crafted ``ilist_traits<T>`` which
1811 superpose the sentinel with the ``ilist`` instance in memory. Pointer
1812 arithmetic is used to obtain the sentinel, which is relative to the ``ilist``'s
1813 ``this`` pointer. The ``ilist`` is augmented by an extra pointer, which serves
1814 as the back-link of the sentinel. This is the only field in the ghostly
1815 sentinel which can be legally accessed.
1819 Other Sequential Container options
1820 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
1822 Other STL containers are available, such as ``std::string``.
1824 There are also various STL adapter classes such as ``std::queue``,
1825 ``std::priority_queue``, ``std::stack``, etc. These provide simplified access
1826 to an underlying container but don't affect the cost of the container itself.
1830 String-like containers
1831 ----------------------
1833 There are a variety of ways to pass around and use strings in C and C++, and
1834 LLVM adds a few new options to choose from. Pick the first option on this list
1835 that will do what you need, they are ordered according to their relative cost.
1837 Note that it is generally preferred to *not* pass strings around as ``const
1838 char*``'s. These have a number of problems, including the fact that they
1839 cannot represent embedded nul ("\0") characters, and do not have a length
1840 available efficiently. The general replacement for '``const char*``' is
1843 For more information on choosing string containers for APIs, please see
1844 :ref:`Passing Strings <string_apis>`.
1848 llvm/ADT/StringRef.h
1849 ^^^^^^^^^^^^^^^^^^^^
1851 The StringRef class is a simple value class that contains a pointer to a
1852 character and a length, and is quite related to the :ref:`ArrayRef
1853 <dss_arrayref>` class (but specialized for arrays of characters). Because
1854 StringRef carries a length with it, it safely handles strings with embedded nul
1855 characters in it, getting the length does not require a strlen call, and it even
1856 has very convenient APIs for slicing and dicing the character range that it
1859 StringRef is ideal for passing simple strings around that are known to be live,
1860 either because they are C string literals, std::string, a C array, or a
1861 SmallVector. Each of these cases has an efficient implicit conversion to
1862 StringRef, which doesn't result in a dynamic strlen being executed.
1864 StringRef has a few major limitations which make more powerful string containers
1867 #. You cannot directly convert a StringRef to a 'const char*' because there is
1868 no way to add a trailing nul (unlike the .c_str() method on various stronger
1871 #. StringRef doesn't own or keep alive the underlying string bytes.
1872 As such it can easily lead to dangling pointers, and is not suitable for
1873 embedding in datastructures in most cases (instead, use an std::string or
1874 something like that).
1876 #. For the same reason, StringRef cannot be used as the return value of a
1877 method if the method "computes" the result string. Instead, use std::string.
1879 #. StringRef's do not allow you to mutate the pointed-to string bytes and it
1880 doesn't allow you to insert or remove bytes from the range. For editing
1881 operations like this, it interoperates with the :ref:`Twine <dss_twine>`
1884 Because of its strengths and limitations, it is very common for a function to
1885 take a StringRef and for a method on an object to return a StringRef that points
1886 into some string that it owns.
1893 The Twine class is used as an intermediary datatype for APIs that want to take a
1894 string that can be constructed inline with a series of concatenations. Twine
1895 works by forming recursive instances of the Twine datatype (a simple value
1896 object) on the stack as temporary objects, linking them together into a tree
1897 which is then linearized when the Twine is consumed. Twine is only safe to use
1898 as the argument to a function, and should always be a const reference, e.g.:
1902 void foo(const Twine &T);
1906 foo(X + "." + Twine(i));
1908 This example forms a string like "blarg.42" by concatenating the values
1909 together, and does not form intermediate strings containing "blarg" or "blarg.".
1911 Because Twine is constructed with temporary objects on the stack, and because
1912 these instances are destroyed at the end of the current statement, it is an
1913 inherently dangerous API. For example, this simple variant contains undefined
1914 behavior and will probably crash:
1918 void foo(const Twine &T);
1922 const Twine &Tmp = X + "." + Twine(i);
1925 ... because the temporaries are destroyed before the call. That said, Twine's
1926 are much more efficient than intermediate std::string temporaries, and they work
1927 really well with StringRef. Just be aware of their limitations.
1929 .. _dss_smallstring:
1931 llvm/ADT/SmallString.h
1932 ^^^^^^^^^^^^^^^^^^^^^^
1934 SmallString is a subclass of :ref:`SmallVector <dss_smallvector>` that adds some
1935 convenience APIs like += that takes StringRef's. SmallString avoids allocating
1936 memory in the case when the preallocated space is enough to hold its data, and
1937 it calls back to general heap allocation when required. Since it owns its data,
1938 it is very safe to use and supports full mutation of the string.
1940 Like SmallVector's, the big downside to SmallString is their sizeof. While they
1941 are optimized for small strings, they themselves are not particularly small.
1942 This means that they work great for temporary scratch buffers on the stack, but
1943 should not generally be put into the heap: it is very rare to see a SmallString
1944 as the member of a frequently-allocated heap data structure or returned
1952 The standard C++ std::string class is a very general class that (like
1953 SmallString) owns its underlying data. sizeof(std::string) is very reasonable
1954 so it can be embedded into heap data structures and returned by-value. On the
1955 other hand, std::string is highly inefficient for inline editing (e.g.
1956 concatenating a bunch of stuff together) and because it is provided by the
1957 standard library, its performance characteristics depend a lot of the host
1958 standard library (e.g. libc++ and MSVC provide a highly optimized string class,
1959 GCC contains a really slow implementation).
1961 The major disadvantage of std::string is that almost every operation that makes
1962 them larger can allocate memory, which is slow. As such, it is better to use
1963 SmallVector or Twine as a scratch buffer, but then use std::string to persist
1968 Set-Like Containers (std::set, SmallSet, SetVector, etc)
1969 --------------------------------------------------------
1971 Set-like containers are useful when you need to canonicalize multiple values
1972 into a single representation. There are several different choices for how to do
1973 this, providing various trade-offs.
1975 .. _dss_sortedvectorset:
1980 If you intend to insert a lot of elements, then do a lot of queries, a great
1981 approach is to use an std::vector (or other sequential container) with
1982 std::sort+std::unique to remove duplicates. This approach works really well if
1983 your usage pattern has these two distinct phases (insert then query), and can be
1984 coupled with a good choice of :ref:`sequential container <ds_sequential>`.
1986 This combination provides the several nice properties: the result data is
1987 contiguous in memory (good for cache locality), has few allocations, is easy to
1988 address (iterators in the final vector are just indices or pointers), and can be
1989 efficiently queried with a standard binary search (e.g.
1990 ``std::lower_bound``; if you want the whole range of elements comparing
1991 equal, use ``std::equal_range``).
1998 If you have a set-like data structure that is usually small and whose elements
1999 are reasonably small, a ``SmallSet<Type, N>`` is a good choice. This set has
2000 space for N elements in place (thus, if the set is dynamically smaller than N,
2001 no malloc traffic is required) and accesses them with a simple linear search.
2002 When the set grows beyond N elements, it allocates a more expensive
2003 representation that guarantees efficient access (for most types, it falls back
2004 to :ref:`std::set <dss_set>`, but for pointers it uses something far better,
2005 :ref:`SmallPtrSet <dss_smallptrset>`.
2007 The magic of this class is that it handles small sets extremely efficiently, but
2008 gracefully handles extremely large sets without loss of efficiency.
2010 .. _dss_smallptrset:
2012 llvm/ADT/SmallPtrSet.h
2013 ^^^^^^^^^^^^^^^^^^^^^^
2015 ``SmallPtrSet`` has all the advantages of ``SmallSet`` (and a ``SmallSet`` of
2016 pointers is transparently implemented with a ``SmallPtrSet``). If more than N
2017 insertions are performed, a single quadratically probed hash table is allocated
2018 and grows as needed, providing extremely efficient access (constant time
2019 insertion/deleting/queries with low constant factors) and is very stingy with
2022 Note that, unlike :ref:`std::set <dss_set>`, the iterators of ``SmallPtrSet``
2023 are invalidated whenever an insertion occurs. Also, the values visited by the
2024 iterators are not visited in sorted order.
2028 llvm/ADT/StringSet.h
2029 ^^^^^^^^^^^^^^^^^^^^
2031 ``StringSet`` is a thin wrapper around :ref:`StringMap\<char\> <dss_stringmap>`,
2032 and it allows efficient storage and retrieval of unique strings.
2034 Functionally analogous to ``SmallSet<StringRef>``, ``StringSet`` also supports
2035 iteration. (The iterator dereferences to a ``StringMapEntry<char>``, so you
2036 need to call ``i->getKey()`` to access the item of the StringSet.) On the
2037 other hand, ``StringSet`` doesn't support range-insertion and
2038 copy-construction, which :ref:`SmallSet <dss_smallset>` and :ref:`SmallPtrSet
2039 <dss_smallptrset>` do support.
2046 DenseSet is a simple quadratically probed hash table. It excels at supporting
2047 small values: it uses a single allocation to hold all of the pairs that are
2048 currently inserted in the set. DenseSet is a great way to unique small values
2049 that are not simple pointers (use :ref:`SmallPtrSet <dss_smallptrset>` for
2050 pointers). Note that DenseSet has the same requirements for the value type that
2051 :ref:`DenseMap <dss_densemap>` has.
2055 llvm/ADT/SparseSet.h
2056 ^^^^^^^^^^^^^^^^^^^^
2058 SparseSet holds a small number of objects identified by unsigned keys of
2059 moderate size. It uses a lot of memory, but provides operations that are almost
2060 as fast as a vector. Typical keys are physical registers, virtual registers, or
2061 numbered basic blocks.
2063 SparseSet is useful for algorithms that need very fast clear/find/insert/erase
2064 and fast iteration over small sets. It is not intended for building composite
2067 .. _dss_sparsemultiset:
2069 llvm/ADT/SparseMultiSet.h
2070 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2072 SparseMultiSet adds multiset behavior to SparseSet, while retaining SparseSet's
2073 desirable attributes. Like SparseSet, it typically uses a lot of memory, but
2074 provides operations that are almost as fast as a vector. Typical keys are
2075 physical registers, virtual registers, or numbered basic blocks.
2077 SparseMultiSet is useful for algorithms that need very fast
2078 clear/find/insert/erase of the entire collection, and iteration over sets of
2079 elements sharing a key. It is often a more efficient choice than using composite
2080 data structures (e.g. vector-of-vectors, map-of-vectors). It is not intended for
2081 building composite data structures.
2085 llvm/ADT/FoldingSet.h
2086 ^^^^^^^^^^^^^^^^^^^^^
2088 FoldingSet is an aggregate class that is really good at uniquing
2089 expensive-to-create or polymorphic objects. It is a combination of a chained
2090 hash table with intrusive links (uniqued objects are required to inherit from
2091 FoldingSetNode) that uses :ref:`SmallVector <dss_smallvector>` as part of its ID
2094 Consider a case where you want to implement a "getOrCreateFoo" method for a
2095 complex object (for example, a node in the code generator). The client has a
2096 description of **what** it wants to generate (it knows the opcode and all the
2097 operands), but we don't want to 'new' a node, then try inserting it into a set
2098 only to find out it already exists, at which point we would have to delete it
2099 and return the node that already exists.
2101 To support this style of client, FoldingSet perform a query with a
2102 FoldingSetNodeID (which wraps SmallVector) that can be used to describe the
2103 element that we want to query for. The query either returns the element
2104 matching the ID or it returns an opaque ID that indicates where insertion should
2105 take place. Construction of the ID usually does not require heap traffic.
2107 Because FoldingSet uses intrusive links, it can support polymorphic objects in
2108 the set (for example, you can have SDNode instances mixed with LoadSDNodes).
2109 Because the elements are individually allocated, pointers to the elements are
2110 stable: inserting or removing elements does not invalidate any pointers to other
2118 ``std::set`` is a reasonable all-around set class, which is decent at many
2119 things but great at nothing. std::set allocates memory for each element
2120 inserted (thus it is very malloc intensive) and typically stores three pointers
2121 per element in the set (thus adding a large amount of per-element space
2122 overhead). It offers guaranteed log(n) performance, which is not particularly
2123 fast from a complexity standpoint (particularly if the elements of the set are
2124 expensive to compare, like strings), and has extremely high constant factors for
2125 lookup, insertion and removal.
2127 The advantages of std::set are that its iterators are stable (deleting or
2128 inserting an element from the set does not affect iterators or pointers to other
2129 elements) and that iteration over the set is guaranteed to be in sorted order.
2130 If the elements in the set are large, then the relative overhead of the pointers
2131 and malloc traffic is not a big deal, but if the elements of the set are small,
2132 std::set is almost never a good choice.
2136 llvm/ADT/SetVector.h
2137 ^^^^^^^^^^^^^^^^^^^^
2139 LLVM's ``SetVector<Type>`` is an adapter class that combines your choice of a
2140 set-like container along with a :ref:`Sequential Container <ds_sequential>` The
2141 important property that this provides is efficient insertion with uniquing
2142 (duplicate elements are ignored) with iteration support. It implements this by
2143 inserting elements into both a set-like container and the sequential container,
2144 using the set-like container for uniquing and the sequential container for
2147 The difference between SetVector and other sets is that the order of iteration
2148 is guaranteed to match the order of insertion into the SetVector. This property
2149 is really important for things like sets of pointers. Because pointer values
2150 are non-deterministic (e.g. vary across runs of the program on different
2151 machines), iterating over the pointers in the set will not be in a well-defined
2154 The drawback of SetVector is that it requires twice as much space as a normal
2155 set and has the sum of constant factors from the set-like container and the
2156 sequential container that it uses. Use it **only** if you need to iterate over
2157 the elements in a deterministic order. SetVector is also expensive to delete
2158 elements out of (linear time), unless you use its "pop_back" method, which is
2161 ``SetVector`` is an adapter class that defaults to using ``std::vector`` and a
2162 size 16 ``SmallSet`` for the underlying containers, so it is quite expensive.
2163 However, ``"llvm/ADT/SetVector.h"`` also provides a ``SmallSetVector`` class,
2164 which defaults to using a ``SmallVector`` and ``SmallSet`` of a specified size.
2165 If you use this, and if your sets are dynamically smaller than ``N``, you will
2166 save a lot of heap traffic.
2168 .. _dss_uniquevector:
2170 llvm/ADT/UniqueVector.h
2171 ^^^^^^^^^^^^^^^^^^^^^^^
2173 UniqueVector is similar to :ref:`SetVector <dss_setvector>` but it retains a
2174 unique ID for each element inserted into the set. It internally contains a map
2175 and a vector, and it assigns a unique ID for each value inserted into the set.
2177 UniqueVector is very expensive: its cost is the sum of the cost of maintaining
2178 both the map and vector, it has high complexity, high constant factors, and
2179 produces a lot of malloc traffic. It should be avoided.
2181 .. _dss_immutableset:
2183 llvm/ADT/ImmutableSet.h
2184 ^^^^^^^^^^^^^^^^^^^^^^^
2186 ImmutableSet is an immutable (functional) set implementation based on an AVL
2187 tree. Adding or removing elements is done through a Factory object and results
2188 in the creation of a new ImmutableSet object. If an ImmutableSet already exists
2189 with the given contents, then the existing one is returned; equality is compared
2190 with a FoldingSetNodeID. The time and space complexity of add or remove
2191 operations is logarithmic in the size of the original set.
2193 There is no method for returning an element of the set, you can only check for
2198 Other Set-Like Container Options
2199 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2201 The STL provides several other options, such as std::multiset and the various
2202 "hash_set" like containers (whether from C++ TR1 or from the SGI library). We
2203 never use hash_set and unordered_set because they are generally very expensive
2204 (each insertion requires a malloc) and very non-portable.
2206 std::multiset is useful if you're not interested in elimination of duplicates,
2207 but has all the drawbacks of :ref:`std::set <dss_set>`. A sorted vector
2208 (where you don't delete duplicate entries) or some other approach is almost
2213 Map-Like Containers (std::map, DenseMap, etc)
2214 ---------------------------------------------
2216 Map-like containers are useful when you want to associate data to a key. As
2217 usual, there are a lot of different ways to do this. :)
2219 .. _dss_sortedvectormap:
2224 If your usage pattern follows a strict insert-then-query approach, you can
2225 trivially use the same approach as :ref:`sorted vectors for set-like containers
2226 <dss_sortedvectorset>`. The only difference is that your query function (which
2227 uses std::lower_bound to get efficient log(n) lookup) should only compare the
2228 key, not both the key and value. This yields the same advantages as sorted
2233 llvm/ADT/StringMap.h
2234 ^^^^^^^^^^^^^^^^^^^^
2236 Strings are commonly used as keys in maps, and they are difficult to support
2237 efficiently: they are variable length, inefficient to hash and compare when
2238 long, expensive to copy, etc. StringMap is a specialized container designed to
2239 cope with these issues. It supports mapping an arbitrary range of bytes to an
2240 arbitrary other object.
2242 The StringMap implementation uses a quadratically-probed hash table, where the
2243 buckets store a pointer to the heap allocated entries (and some other stuff).
2244 The entries in the map must be heap allocated because the strings are variable
2245 length. The string data (key) and the element object (value) are stored in the
2246 same allocation with the string data immediately after the element object.
2247 This container guarantees the "``(char*)(&Value+1)``" points to the key string
2250 The StringMap is very fast for several reasons: quadratic probing is very cache
2251 efficient for lookups, the hash value of strings in buckets is not recomputed
2252 when looking up an element, StringMap rarely has to touch the memory for
2253 unrelated objects when looking up a value (even when hash collisions happen),
2254 hash table growth does not recompute the hash values for strings already in the
2255 table, and each pair in the map is store in a single allocation (the string data
2256 is stored in the same allocation as the Value of a pair).
2258 StringMap also provides query methods that take byte ranges, so it only ever
2259 copies a string if a value is inserted into the table.
2261 StringMap iteration order, however, is not guaranteed to be deterministic, so
2262 any uses which require that should instead use a std::map.
2266 llvm/ADT/IndexedMap.h
2267 ^^^^^^^^^^^^^^^^^^^^^
2269 IndexedMap is a specialized container for mapping small dense integers (or
2270 values that can be mapped to small dense integers) to some other type. It is
2271 internally implemented as a vector with a mapping function that maps the keys
2272 to the dense integer range.
2274 This is useful for cases like virtual registers in the LLVM code generator: they
2275 have a dense mapping that is offset by a compile-time constant (the first
2276 virtual register ID).
2283 DenseMap is a simple quadratically probed hash table. It excels at supporting
2284 small keys and values: it uses a single allocation to hold all of the pairs
2285 that are currently inserted in the map. DenseMap is a great way to map
2286 pointers to pointers, or map other small types to each other.
2288 There are several aspects of DenseMap that you should be aware of, however.
2289 The iterators in a DenseMap are invalidated whenever an insertion occurs,
2290 unlike map. Also, because DenseMap allocates space for a large number of
2291 key/value pairs (it starts with 64 by default), it will waste a lot of space if
2292 your keys or values are large. Finally, you must implement a partial
2293 specialization of DenseMapInfo for the key that you want, if it isn't already
2294 supported. This is required to tell DenseMap about two special marker values
2295 (which can never be inserted into the map) that it needs internally.
2297 DenseMap's find_as() method supports lookup operations using an alternate key
2298 type. This is useful in cases where the normal key type is expensive to
2299 construct, but cheap to compare against. The DenseMapInfo is responsible for
2300 defining the appropriate comparison and hashing methods for each alternate key
2308 ValueMap is a wrapper around a :ref:`DenseMap <dss_densemap>` mapping
2309 ``Value*``\ s (or subclasses) to another type. When a Value is deleted or
2310 RAUW'ed, ValueMap will update itself so the new version of the key is mapped to
2311 the same value, just as if the key were a WeakVH. You can configure exactly how
2312 this happens, and what else happens on these two events, by passing a ``Config``
2313 parameter to the ValueMap template.
2315 .. _dss_intervalmap:
2317 llvm/ADT/IntervalMap.h
2318 ^^^^^^^^^^^^^^^^^^^^^^
2320 IntervalMap is a compact map for small keys and values. It maps key intervals
2321 instead of single keys, and it will automatically coalesce adjacent intervals.
2322 When the map only contains a few intervals, they are stored in the map object
2323 itself to avoid allocations.
2325 The IntervalMap iterators are quite big, so they should not be passed around as
2326 STL iterators. The heavyweight iterators allow a smaller data structure.
2333 std::map has similar characteristics to :ref:`std::set <dss_set>`: it uses a
2334 single allocation per pair inserted into the map, it offers log(n) lookup with
2335 an extremely large constant factor, imposes a space penalty of 3 pointers per
2336 pair in the map, etc.
2338 std::map is most useful when your keys or values are very large, if you need to
2339 iterate over the collection in sorted order, or if you need stable iterators
2340 into the map (i.e. they don't get invalidated if an insertion or deletion of
2341 another element takes place).
2345 llvm/ADT/MapVector.h
2346 ^^^^^^^^^^^^^^^^^^^^
2348 ``MapVector<KeyT,ValueT>`` provides a subset of the DenseMap interface. The
2349 main difference is that the iteration order is guaranteed to be the insertion
2350 order, making it an easy (but somewhat expensive) solution for non-deterministic
2351 iteration over maps of pointers.
2353 It is implemented by mapping from key to an index in a vector of key,value
2354 pairs. This provides fast lookup and iteration, but has two main drawbacks:
2355 the key is stored twice and removing elements takes linear time. If it is
2356 necessary to remove elements, it's best to remove them in bulk using
2359 .. _dss_inteqclasses:
2361 llvm/ADT/IntEqClasses.h
2362 ^^^^^^^^^^^^^^^^^^^^^^^
2364 IntEqClasses provides a compact representation of equivalence classes of small
2365 integers. Initially, each integer in the range 0..n-1 has its own equivalence
2366 class. Classes can be joined by passing two class representatives to the
2367 join(a, b) method. Two integers are in the same class when findLeader() returns
2368 the same representative.
2370 Once all equivalence classes are formed, the map can be compressed so each
2371 integer 0..n-1 maps to an equivalence class number in the range 0..m-1, where m
2372 is the total number of equivalence classes. The map must be uncompressed before
2373 it can be edited again.
2375 .. _dss_immutablemap:
2377 llvm/ADT/ImmutableMap.h
2378 ^^^^^^^^^^^^^^^^^^^^^^^
2380 ImmutableMap is an immutable (functional) map implementation based on an AVL
2381 tree. Adding or removing elements is done through a Factory object and results
2382 in the creation of a new ImmutableMap object. If an ImmutableMap already exists
2383 with the given key set, then the existing one is returned; equality is compared
2384 with a FoldingSetNodeID. The time and space complexity of add or remove
2385 operations is logarithmic in the size of the original map.
2389 Other Map-Like Container Options
2390 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2392 The STL provides several other options, such as std::multimap and the various
2393 "hash_map" like containers (whether from C++ TR1 or from the SGI library). We
2394 never use hash_set and unordered_set because they are generally very expensive
2395 (each insertion requires a malloc) and very non-portable.
2397 std::multimap is useful if you want to map a key to multiple values, but has all
2398 the drawbacks of std::map. A sorted vector or some other approach is almost
2403 Bit storage containers (BitVector, SparseBitVector, CoalescingBitVector)
2404 ------------------------------------------------------------------------
2406 There are three bit storage containers, and choosing when to use each is
2407 relatively straightforward.
2409 One additional option is ``std::vector<bool>``: we discourage its use for two
2410 reasons 1) the implementation in many common compilers (e.g. commonly
2411 available versions of GCC) is extremely inefficient and 2) the C++ standards
2412 committee is likely to deprecate this container and/or change it significantly
2413 somehow. In any case, please don't use it.
2420 The BitVector container provides a dynamic size set of bits for manipulation.
2421 It supports individual bit setting/testing, as well as set operations. The set
2422 operations take time O(size of bitvector), but operations are performed one word
2423 at a time, instead of one bit at a time. This makes the BitVector very fast for
2424 set operations compared to other containers. Use the BitVector when you expect
2425 the number of set bits to be high (i.e. a dense set).
2427 .. _dss_smallbitvector:
2432 The SmallBitVector container provides the same interface as BitVector, but it is
2433 optimized for the case where only a small number of bits, less than 25 or so,
2434 are needed. It also transparently supports larger bit counts, but slightly less
2435 efficiently than a plain BitVector, so SmallBitVector should only be used when
2436 larger counts are rare.
2438 At this time, SmallBitVector does not support set operations (and, or, xor), and
2439 its operator[] does not provide an assignable lvalue.
2441 .. _dss_sparsebitvector:
2446 The SparseBitVector container is much like BitVector, with one major difference:
2447 Only the bits that are set, are stored. This makes the SparseBitVector much
2448 more space efficient than BitVector when the set is sparse, as well as making
2449 set operations O(number of set bits) instead of O(size of universe). The
2450 downside to the SparseBitVector is that setting and testing of random bits is
2451 O(N), and on large SparseBitVectors, this can be slower than BitVector. In our
2452 implementation, setting or testing bits in sorted order (either forwards or
2453 reverse) is O(1) worst case. Testing and setting bits within 128 bits (depends
2454 on size) of the current bit is also O(1). As a general statement,
2455 testing/setting bits in a SparseBitVector is O(distance away from last set bit).
2457 .. _dss_coalescingbitvector:
2462 The CoalescingBitVector container is similar in principle to a SparseBitVector,
2463 but is optimized to represent large contiguous ranges of set bits compactly. It
2464 does this by coalescing contiguous ranges of set bits into intervals. Searching
2465 for a bit in a CoalescingBitVector is O(log(gaps between contiguous ranges)).
2467 CoalescingBitVector is a better choice than BitVector when gaps between ranges
2468 of set bits are large. It's a better choice than SparseBitVector when find()
2469 operations must have fast, predictable performance. However, it's not a good
2470 choice for representing sets which have lots of very short ranges. E.g. the set
2471 `{2*x : x \in [0, n)}` would be a pathological input.
2478 A handful of `GDB pretty printers
2479 <https://sourceware.org/gdb/onlinedocs/gdb/Pretty-Printing.html>`__ are
2480 provided for some of the core LLVM libraries. To use them, execute the
2481 following (or add it to your ``~/.gdbinit``)::
2483 source /path/to/llvm/src/utils/gdb-scripts/prettyprinters.py
2485 It also might be handy to enable the `print pretty
2486 <http://ftp.gnu.org/old-gnu/Manuals/gdb/html_node/gdb_57.html>`__ option to
2487 avoid data structures being printed as a big block of text.
2491 Helpful Hints for Common Operations
2492 ===================================
2494 This section describes how to perform some very simple transformations of LLVM
2495 code. This is meant to give examples of common idioms used, showing the
2496 practical side of LLVM transformations.
2498 Because this is a "how-to" section, you should also read about the main classes
2499 that you will be working with. The :ref:`Core LLVM Class Hierarchy Reference
2500 <coreclasses>` contains details and descriptions of the main classes that you
2505 Basic Inspection and Traversal Routines
2506 ---------------------------------------
2508 The LLVM compiler infrastructure have many different data structures that may be
2509 traversed. Following the example of the C++ standard template library, the
2510 techniques used to traverse these various data structures are all basically the
2511 same. For an enumerable sequence of values, the ``XXXbegin()`` function (or
2512 method) returns an iterator to the start of the sequence, the ``XXXend()``
2513 function returns an iterator pointing to one past the last valid element of the
2514 sequence, and there is some ``XXXiterator`` data type that is common between the
2517 Because the pattern for iteration is common across many different aspects of the
2518 program representation, the standard template library algorithms may be used on
2519 them, and it is easier to remember how to iterate. First we show a few common
2520 examples of the data structures that need to be traversed. Other data
2521 structures are traversed in very similar ways.
2523 .. _iterate_function:
2525 Iterating over the ``BasicBlock`` in a ``Function``
2526 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2528 It's quite common to have a ``Function`` instance that you'd like to transform
2529 in some way; in particular, you'd like to manipulate its ``BasicBlock``\ s. To
2530 facilitate this, you'll need to iterate over all of the ``BasicBlock``\ s that
2531 constitute the ``Function``. The following is an example that prints the name
2532 of a ``BasicBlock`` and the number of ``Instruction``\ s it contains:
2536 Function &Func = ...
2537 for (BasicBlock &BB : Func)
2538 // Print out the name of the basic block if it has one, and then the
2539 // number of instructions that it contains
2540 errs() << "Basic block (name=" << BB.getName() << ") has "
2541 << BB.size() << " instructions.\n";
2543 .. _iterate_basicblock:
2545 Iterating over the ``Instruction`` in a ``BasicBlock``
2546 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2548 Just like when dealing with ``BasicBlock``\ s in ``Function``\ s, it's easy to
2549 iterate over the individual instructions that make up ``BasicBlock``\ s. Here's
2550 a code snippet that prints out each instruction in a ``BasicBlock``:
2554 BasicBlock& BB = ...
2555 for (Instruction &I : BB)
2556 // The next statement works since operator<<(ostream&,...)
2557 // is overloaded for Instruction&
2558 errs() << I << "\n";
2561 However, this isn't really the best way to print out the contents of a
2562 ``BasicBlock``! Since the ostream operators are overloaded for virtually
2563 anything you'll care about, you could have just invoked the print routine on the
2564 basic block itself: ``errs() << BB << "\n";``.
2566 .. _iterate_insiter:
2568 Iterating over the ``Instruction`` in a ``Function``
2569 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2571 If you're finding that you commonly iterate over a ``Function``'s
2572 ``BasicBlock``\ s and then that ``BasicBlock``'s ``Instruction``\ s,
2573 ``InstIterator`` should be used instead. You'll need to include
2574 ``llvm/IR/InstIterator.h`` (`doxygen
2575 <https://llvm.org/doxygen/InstIterator_8h.html>`__) and then instantiate
2576 ``InstIterator``\ s explicitly in your code. Here's a small example that shows
2577 how to dump all instructions in a function to the standard error stream:
2581 #include "llvm/IR/InstIterator.h"
2583 // F is a pointer to a Function instance
2584 for (inst_iterator I = inst_begin(F), E = inst_end(F); I != E; ++I)
2585 errs() << *I << "\n";
2587 Easy, isn't it? You can also use ``InstIterator``\ s to fill a work list with
2588 its initial contents. For example, if you wanted to initialize a work list to
2589 contain all instructions in a ``Function`` F, all you would need to do is
2594 std::set<Instruction*> worklist;
2595 // or better yet, SmallPtrSet<Instruction*, 64> worklist;
2597 for (inst_iterator I = inst_begin(F), E = inst_end(F); I != E; ++I)
2598 worklist.insert(&*I);
2600 The STL set ``worklist`` would now contain all instructions in the ``Function``
2603 .. _iterate_convert:
2605 Turning an iterator into a class pointer (and vice-versa)
2606 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2608 Sometimes, it'll be useful to grab a reference (or pointer) to a class instance
2609 when all you've got at hand is an iterator. Well, extracting a reference or a
2610 pointer from an iterator is very straight-forward. Assuming that ``i`` is a
2611 ``BasicBlock::iterator`` and ``j`` is a ``BasicBlock::const_iterator``:
2615 Instruction& inst = *i; // Grab reference to instruction reference
2616 Instruction* pinst = &*i; // Grab pointer to instruction reference
2617 const Instruction& inst = *j;
2619 However, the iterators you'll be working with in the LLVM framework are special:
2620 they will automatically convert to a ptr-to-instance type whenever they need to.
2621 Instead of dereferencing the iterator and then taking the address of the result,
2622 you can simply assign the iterator to the proper pointer type and you get the
2623 dereference and address-of operation as a result of the assignment (behind the
2624 scenes, this is a result of overloading casting mechanisms). Thus the second
2625 line of the last example,
2629 Instruction *pinst = &*i;
2631 is semantically equivalent to
2635 Instruction *pinst = i;
2637 It's also possible to turn a class pointer into the corresponding iterator, and
2638 this is a constant time operation (very efficient). The following code snippet
2639 illustrates use of the conversion constructors provided by LLVM iterators. By
2640 using these, you can explicitly grab the iterator of something without actually
2641 obtaining it via iteration over some structure:
2645 void printNextInstruction(Instruction* inst) {
2646 BasicBlock::iterator it(inst);
2647 ++it; // After this line, it refers to the instruction after *inst
2648 if (it != inst->getParent()->end()) errs() << *it << "\n";
2651 Unfortunately, these implicit conversions come at a cost; they prevent these
2652 iterators from conforming to standard iterator conventions, and thus from being
2653 usable with standard algorithms and containers. For example, they prevent the
2654 following code, where ``B`` is a ``BasicBlock``, from compiling:
2658 llvm::SmallVector<llvm::Instruction *, 16>(B->begin(), B->end());
2660 Because of this, these implicit conversions may be removed some day, and
2661 ``operator*`` changed to return a pointer instead of a reference.
2663 .. _iterate_complex:
2665 Finding call sites: a slightly more complex example
2666 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2668 Say that you're writing a FunctionPass and would like to count all the locations
2669 in the entire module (that is, across every ``Function``) where a certain
2670 function (i.e., some ``Function *``) is already in scope. As you'll learn
2671 later, you may want to use an ``InstVisitor`` to accomplish this in a much more
2672 straight-forward manner, but this example will allow us to explore how you'd do
2673 it if you didn't have ``InstVisitor`` around. In pseudo-code, this is what we
2676 .. code-block:: none
2678 initialize callCounter to zero
2679 for each Function f in the Module
2680 for each BasicBlock b in f
2681 for each Instruction i in b
2682 if (i a Call and calls the given function)
2683 increment callCounter
2685 And the actual code is (remember, because we're writing a ``FunctionPass``, our
2686 ``FunctionPass``-derived class simply has to override the ``runOnFunction``
2691 Function* targetFunc = ...;
2693 class OurFunctionPass : public FunctionPass {
2695 OurFunctionPass(): callCounter(0) { }
2697 virtual runOnFunction(Function& F) {
2698 for (BasicBlock &B : F) {
2699 for (Instruction &I: B) {
2700 if (auto *CB = dyn_cast<CallBase>(&I)) {
2701 // We know we've encountered some kind of call instruction (call,
2702 // invoke, or callbr), so we need to determine if it's a call to
2703 // the function pointed to by m_func or not.
2704 if (CB->getCalledFunction() == targetFunc)
2712 unsigned callCounter;
2717 Iterating over def-use & use-def chains
2718 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2720 Frequently, we might have an instance of the ``Value`` class (`doxygen
2721 <https://llvm.org/doxygen/classllvm_1_1Value.html>`__) and we want to determine
2722 which ``User``\ s use the ``Value``. The list of all ``User``\ s of a particular
2723 ``Value`` is called a *def-use* chain. For example, let's say we have a
2724 ``Function*`` named ``F`` to a particular function ``foo``. Finding all of the
2725 instructions that *use* ``foo`` is as simple as iterating over the *def-use*
2732 for (User *U : F->users()) {
2733 if (Instruction *Inst = dyn_cast<Instruction>(U)) {
2734 errs() << "F is used in instruction:\n";
2735 errs() << *Inst << "\n";
2738 Alternatively, it's common to have an instance of the ``User`` Class (`doxygen
2739 <https://llvm.org/doxygen/classllvm_1_1User.html>`__) and need to know what
2740 ``Value``\ s are used by it. The list of all ``Value``\ s used by a ``User`` is
2741 known as a *use-def* chain. Instances of class ``Instruction`` are common
2742 ``User`` s, so we might want to iterate over all of the values that a particular
2743 instruction uses (that is, the operands of the particular ``Instruction``):
2747 Instruction *pi = ...;
2749 for (Use &U : pi->operands()) {
2754 Declaring objects as ``const`` is an important tool of enforcing mutation free
2755 algorithms (such as analyses, etc.). For this purpose above iterators come in
2756 constant flavors as ``Value::const_use_iterator`` and
2757 ``Value::const_op_iterator``. They automatically arise when calling
2758 ``use/op_begin()`` on ``const Value*``\ s or ``const User*``\ s respectively.
2759 Upon dereferencing, they return ``const Use*``\ s. Otherwise the above patterns
2764 Iterating over predecessors & successors of blocks
2765 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2767 Iterating over the predecessors and successors of a block is quite easy with the
2768 routines defined in ``"llvm/IR/CFG.h"``. Just use code like this to
2769 iterate over all predecessors of BB:
2773 #include "llvm/IR/CFG.h"
2774 BasicBlock *BB = ...;
2776 for (BasicBlock *Pred : predecessors(BB)) {
2780 Similarly, to iterate over successors use ``successors``.
2784 Making simple changes
2785 ---------------------
2787 There are some primitive transformation operations present in the LLVM
2788 infrastructure that are worth knowing about. When performing transformations,
2789 it's fairly common to manipulate the contents of basic blocks. This section
2790 describes some of the common methods for doing so and gives example code.
2792 .. _schanges_creating:
2794 Creating and inserting new ``Instruction``\ s
2795 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2797 *Instantiating Instructions*
2799 Creation of ``Instruction``\ s is straight-forward: simply call the constructor
2800 for the kind of instruction to instantiate and provide the necessary parameters.
2801 For example, an ``AllocaInst`` only *requires* a (const-ptr-to) ``Type``. Thus:
2805 auto *ai = new AllocaInst(Type::Int32Ty);
2807 will create an ``AllocaInst`` instance that represents the allocation of one
2808 integer in the current stack frame, at run time. Each ``Instruction`` subclass
2809 is likely to have varying default parameters which change the semantics of the
2810 instruction, so refer to the `doxygen documentation for the subclass of
2811 Instruction <https://llvm.org/doxygen/classllvm_1_1Instruction.html>`_ that
2812 you're interested in instantiating.
2816 It is very useful to name the values of instructions when you're able to, as
2817 this facilitates the debugging of your transformations. If you end up looking
2818 at generated LLVM machine code, you definitely want to have logical names
2819 associated with the results of instructions! By supplying a value for the
2820 ``Name`` (default) parameter of the ``Instruction`` constructor, you associate a
2821 logical name with the result of the instruction's execution at run time. For
2822 example, say that I'm writing a transformation that dynamically allocates space
2823 for an integer on the stack, and that integer is going to be used as some kind
2824 of index by some other code. To accomplish this, I place an ``AllocaInst`` at
2825 the first point in the first ``BasicBlock`` of some ``Function``, and I'm
2826 intending to use it within the same ``Function``. I might do:
2830 auto *pa = new AllocaInst(Type::Int32Ty, 0, "indexLoc");
2832 where ``indexLoc`` is now the logical name of the instruction's execution value,
2833 which is a pointer to an integer on the run time stack.
2835 *Inserting instructions*
2837 There are essentially three ways to insert an ``Instruction`` into an existing
2838 sequence of instructions that form a ``BasicBlock``:
2840 * Insertion into an explicit instruction list
2842 Given a ``BasicBlock* pb``, an ``Instruction* pi`` within that ``BasicBlock``,
2843 and a newly-created instruction we wish to insert before ``*pi``, we do the
2848 BasicBlock *pb = ...;
2849 Instruction *pi = ...;
2850 auto *newInst = new Instruction(...);
2852 pb->getInstList().insert(pi, newInst); // Inserts newInst before pi in pb
2854 Appending to the end of a ``BasicBlock`` is so common that the ``Instruction``
2855 class and ``Instruction``-derived classes provide constructors which take a
2856 pointer to a ``BasicBlock`` to be appended to. For example code that looked
2861 BasicBlock *pb = ...;
2862 auto *newInst = new Instruction(...);
2864 pb->getInstList().push_back(newInst); // Appends newInst to pb
2870 BasicBlock *pb = ...;
2871 auto *newInst = new Instruction(..., pb);
2873 which is much cleaner, especially if you are creating long instruction
2876 * Insertion into an implicit instruction list
2878 ``Instruction`` instances that are already in ``BasicBlock``\ s are implicitly
2879 associated with an existing instruction list: the instruction list of the
2880 enclosing basic block. Thus, we could have accomplished the same thing as the
2881 above code without being given a ``BasicBlock`` by doing:
2885 Instruction *pi = ...;
2886 auto *newInst = new Instruction(...);
2888 pi->getParent()->getInstList().insert(pi, newInst);
2890 In fact, this sequence of steps occurs so frequently that the ``Instruction``
2891 class and ``Instruction``-derived classes provide constructors which take (as
2892 a default parameter) a pointer to an ``Instruction`` which the newly-created
2893 ``Instruction`` should precede. That is, ``Instruction`` constructors are
2894 capable of inserting the newly-created instance into the ``BasicBlock`` of a
2895 provided instruction, immediately before that instruction. Using an
2896 ``Instruction`` constructor with a ``insertBefore`` (default) parameter, the
2901 Instruction* pi = ...;
2902 auto *newInst = new Instruction(..., pi);
2904 which is much cleaner, especially if you're creating a lot of instructions and
2905 adding them to ``BasicBlock``\ s.
2907 * Insertion using an instance of ``IRBuilder``
2909 Inserting several ``Instruction``\ s can be quite laborious using the previous
2910 methods. The ``IRBuilder`` is a convenience class that can be used to add
2911 several instructions to the end of a ``BasicBlock`` or before a particular
2912 ``Instruction``. It also supports constant folding and renaming named
2913 registers (see ``IRBuilder``'s template arguments).
2915 The example below demonstrates a very simple use of the ``IRBuilder`` where
2916 three instructions are inserted before the instruction ``pi``. The first two
2917 instructions are Call instructions and third instruction multiplies the return
2918 value of the two calls.
2922 Instruction *pi = ...;
2923 IRBuilder<> Builder(pi);
2924 CallInst* callOne = Builder.CreateCall(...);
2925 CallInst* callTwo = Builder.CreateCall(...);
2926 Value* result = Builder.CreateMul(callOne, callTwo);
2928 The example below is similar to the above example except that the created
2929 ``IRBuilder`` inserts instructions at the end of the ``BasicBlock`` ``pb``.
2933 BasicBlock *pb = ...;
2934 IRBuilder<> Builder(pb);
2935 CallInst* callOne = Builder.CreateCall(...);
2936 CallInst* callTwo = Builder.CreateCall(...);
2937 Value* result = Builder.CreateMul(callOne, callTwo);
2939 See :doc:`tutorial/LangImpl03` for a practical use of the ``IRBuilder``.
2942 .. _schanges_deleting:
2944 Deleting Instructions
2945 ^^^^^^^^^^^^^^^^^^^^^
2947 Deleting an instruction from an existing sequence of instructions that form a
2948 BasicBlock_ is very straight-forward: just call the instruction's
2949 ``eraseFromParent()`` method. For example:
2953 Instruction *I = .. ;
2954 I->eraseFromParent();
2956 This unlinks the instruction from its containing basic block and deletes it. If
2957 you'd just like to unlink the instruction from its containing basic block but
2958 not delete it, you can use the ``removeFromParent()`` method.
2960 .. _schanges_replacing:
2962 Replacing an Instruction with another Value
2963 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2965 Replacing individual instructions
2966 """""""""""""""""""""""""""""""""
2968 Including "`llvm/Transforms/Utils/BasicBlockUtils.h
2969 <https://llvm.org/doxygen/BasicBlockUtils_8h_source.html>`_" permits use of two
2970 very useful replace functions: ``ReplaceInstWithValue`` and
2971 ``ReplaceInstWithInst``.
2973 .. _schanges_deleting_sub:
2975 Deleting Instructions
2976 """""""""""""""""""""
2978 * ``ReplaceInstWithValue``
2980 This function replaces all uses of a given instruction with a value, and then
2981 removes the original instruction. The following example illustrates the
2982 replacement of the result of a particular ``AllocaInst`` that allocates memory
2983 for a single integer with a null pointer to an integer.
2987 AllocaInst* instToReplace = ...;
2988 BasicBlock::iterator ii(instToReplace);
2990 ReplaceInstWithValue(instToReplace->getParent()->getInstList(), ii,
2991 Constant::getNullValue(PointerType::getUnqual(Type::Int32Ty)));
2993 * ``ReplaceInstWithInst``
2995 This function replaces a particular instruction with another instruction,
2996 inserting the new instruction into the basic block at the location where the
2997 old instruction was, and replacing any uses of the old instruction with the
2998 new instruction. The following example illustrates the replacement of one
2999 ``AllocaInst`` with another.
3003 AllocaInst* instToReplace = ...;
3004 BasicBlock::iterator ii(instToReplace);
3006 ReplaceInstWithInst(instToReplace->getParent()->getInstList(), ii,
3007 new AllocaInst(Type::Int32Ty, 0, "ptrToReplacedInt"));
3010 Replacing multiple uses of Users and Values
3011 """""""""""""""""""""""""""""""""""""""""""
3013 You can use ``Value::replaceAllUsesWith`` and ``User::replaceUsesOfWith`` to
3014 change more than one use at a time. See the doxygen documentation for the
3015 `Value Class <https://llvm.org/doxygen/classllvm_1_1Value.html>`_ and `User Class
3016 <https://llvm.org/doxygen/classllvm_1_1User.html>`_, respectively, for more
3019 .. _schanges_deletingGV:
3021 Deleting GlobalVariables
3022 ^^^^^^^^^^^^^^^^^^^^^^^^
3024 Deleting a global variable from a module is just as easy as deleting an
3025 Instruction. First, you must have a pointer to the global variable that you
3026 wish to delete. You use this pointer to erase it from its parent, the module.
3031 GlobalVariable *GV = .. ;
3033 GV->eraseFromParent();
3041 This section describes the interaction of the LLVM APIs with multithreading,
3042 both on the part of client applications, and in the JIT, in the hosted
3045 Note that LLVM's support for multithreading is still relatively young. Up
3046 through version 2.5, the execution of threaded hosted applications was
3047 supported, but not threaded client access to the APIs. While this use case is
3048 now supported, clients *must* adhere to the guidelines specified below to ensure
3049 proper operation in multithreaded mode.
3051 Note that, on Unix-like platforms, LLVM requires the presence of GCC's atomic
3052 intrinsics in order to support threaded operation. If you need a
3053 multithreading-capable LLVM on a platform without a suitably modern system
3054 compiler, consider compiling LLVM and LLVM-GCC in single-threaded mode, and
3055 using the resultant compiler to build a copy of LLVM with multithreading
3060 Ending Execution with ``llvm_shutdown()``
3061 -----------------------------------------
3063 When you are done using the LLVM APIs, you should call ``llvm_shutdown()`` to
3064 deallocate memory used for internal structures.
3068 Lazy Initialization with ``ManagedStatic``
3069 ------------------------------------------
3071 ``ManagedStatic`` is a utility class in LLVM used to implement static
3072 initialization of static resources, such as the global type tables. In a
3073 single-threaded environment, it implements a simple lazy initialization scheme.
3074 When LLVM is compiled with support for multi-threading, however, it uses
3075 double-checked locking to implement thread-safe lazy initialization.
3079 Achieving Isolation with ``LLVMContext``
3080 ----------------------------------------
3082 ``LLVMContext`` is an opaque class in the LLVM API which clients can use to
3083 operate multiple, isolated instances of LLVM concurrently within the same
3084 address space. For instance, in a hypothetical compile-server, the compilation
3085 of an individual translation unit is conceptually independent from all the
3086 others, and it would be desirable to be able to compile incoming translation
3087 units concurrently on independent server threads. Fortunately, ``LLVMContext``
3088 exists to enable just this kind of scenario!
3090 Conceptually, ``LLVMContext`` provides isolation. Every LLVM entity
3091 (``Module``\ s, ``Value``\ s, ``Type``\ s, ``Constant``\ s, etc.) in LLVM's
3092 in-memory IR belongs to an ``LLVMContext``. Entities in different contexts
3093 *cannot* interact with each other: ``Module``\ s in different contexts cannot be
3094 linked together, ``Function``\ s cannot be added to ``Module``\ s in different
3095 contexts, etc. What this means is that is safe to compile on multiple
3096 threads simultaneously, as long as no two threads operate on entities within the
3099 In practice, very few places in the API require the explicit specification of a
3100 ``LLVMContext``, other than the ``Type`` creation/lookup APIs. Because every
3101 ``Type`` carries a reference to its owning context, most other entities can
3102 determine what context they belong to by looking at their own ``Type``. If you
3103 are adding new entities to LLVM IR, please try to maintain this interface
3111 LLVM's "eager" JIT compiler is safe to use in threaded programs. Multiple
3112 threads can call ``ExecutionEngine::getPointerToFunction()`` or
3113 ``ExecutionEngine::runFunction()`` concurrently, and multiple threads can run
3114 code output by the JIT concurrently. The user must still ensure that only one
3115 thread accesses IR in a given ``LLVMContext`` while another thread might be
3116 modifying it. One way to do that is to always hold the JIT lock while accessing
3117 IR outside the JIT (the JIT *modifies* the IR by adding ``CallbackVH``\ s).
3118 Another way is to only call ``getPointerToFunction()`` from the
3119 ``LLVMContext``'s thread.
3121 When the JIT is configured to compile lazily (using
3122 ``ExecutionEngine::DisableLazyCompilation(false)``), there is currently a `race
3123 condition <https://bugs.llvm.org/show_bug.cgi?id=5184>`_ in updating call sites
3124 after a function is lazily-jitted. It's still possible to use the lazy JIT in a
3125 threaded program if you ensure that only one thread at a time can call any
3126 particular lazy stub and that the JIT lock guards any IR access, but we suggest
3127 using only the eager JIT in threaded programs.
3134 This section describes some of the advanced or obscure API's that most clients
3135 do not need to be aware of. These API's tend manage the inner workings of the
3136 LLVM system, and only need to be accessed in unusual circumstances.
3140 The ``ValueSymbolTable`` class
3141 ------------------------------
3143 The ``ValueSymbolTable`` (`doxygen
3144 <https://llvm.org/doxygen/classllvm_1_1ValueSymbolTable.html>`__) class provides
3145 a symbol table that the :ref:`Function <c_Function>` and Module_ classes use for
3146 naming value definitions. The symbol table can provide a name for any Value_.
3148 Note that the ``SymbolTable`` class should not be directly accessed by most
3149 clients. It should only be used when iteration over the symbol table names
3150 themselves are required, which is very special purpose. Note that not all LLVM
3151 Value_\ s have names, and those without names (i.e. they have an empty name) do
3152 not exist in the symbol table.
3154 Symbol tables support iteration over the values in the symbol table with
3155 ``begin/end/iterator`` and supports querying to see if a specific name is in the
3156 symbol table (with ``lookup``). The ``ValueSymbolTable`` class exposes no
3157 public mutator methods, instead, simply call ``setName`` on a value, which will
3158 autoinsert it into the appropriate symbol table.
3162 The ``User`` and owned ``Use`` classes' memory layout
3163 -----------------------------------------------------
3165 The ``User`` (`doxygen <https://llvm.org/doxygen/classllvm_1_1User.html>`__)
3166 class provides a basis for expressing the ownership of ``User`` towards other
3167 `Value instance <https://llvm.org/doxygen/classllvm_1_1Value.html>`_\ s. The
3168 ``Use`` (`doxygen <https://llvm.org/doxygen/classllvm_1_1Use.html>`__) helper
3169 class is employed to do the bookkeeping and to facilitate *O(1)* addition and
3174 Interaction and relationship between ``User`` and ``Use`` objects
3175 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3177 A subclass of ``User`` can choose between incorporating its ``Use`` objects or
3178 refer to them out-of-line by means of a pointer. A mixed variant (some ``Use``
3179 s inline others hung off) is impractical and breaks the invariant that the
3180 ``Use`` objects belonging to the same ``User`` form a contiguous array.
3182 We have 2 different layouts in the ``User`` (sub)classes:
3186 The ``Use`` object(s) are inside (resp. at fixed offset) of the ``User``
3187 object and there are a fixed number of them.
3191 The ``Use`` object(s) are referenced by a pointer to an array from the
3192 ``User`` object and there may be a variable number of them.
3194 As of v2.4 each layout still possesses a direct pointer to the start of the
3195 array of ``Use``\ s. Though not mandatory for layout a), we stick to this
3196 redundancy for the sake of simplicity. The ``User`` object also stores the
3197 number of ``Use`` objects it has. (Theoretically this information can also be
3198 calculated given the scheme presented below.)
3200 Special forms of allocation operators (``operator new``) enforce the following
3203 * Layout a) is modelled by prepending the ``User`` object by the ``Use[]``
3206 .. code-block:: none
3208 ...---.---.---.---.-------...
3209 | P | P | P | P | User
3210 '''---'---'---'---'-------'''
3212 * Layout b) is modelled by pointing at the ``Use[]`` array.
3214 .. code-block:: none
3225 *(In the above figures* '``P``' *stands for the* ``Use**`` *that is stored in
3226 each* ``Use`` *object in the member* ``Use::Prev`` *)*
3230 Designing Type Hierarchies and Polymorphic Interfaces
3231 -----------------------------------------------------
3233 There are two different design patterns that tend to result in the use of
3234 virtual dispatch for methods in a type hierarchy in C++ programs. The first is
3235 a genuine type hierarchy where different types in the hierarchy model
3236 a specific subset of the functionality and semantics, and these types nest
3237 strictly within each other. Good examples of this can be seen in the ``Value``
3238 or ``Type`` type hierarchies.
3240 A second is the desire to dispatch dynamically across a collection of
3241 polymorphic interface implementations. This latter use case can be modeled with
3242 virtual dispatch and inheritance by defining an abstract interface base class
3243 which all implementations derive from and override. However, this
3244 implementation strategy forces an **"is-a"** relationship to exist that is not
3245 actually meaningful. There is often not some nested hierarchy of useful
3246 generalizations which code might interact with and move up and down. Instead,
3247 there is a singular interface which is dispatched across a range of
3250 The preferred implementation strategy for the second use case is that of
3251 generic programming (sometimes called "compile-time duck typing" or "static
3252 polymorphism"). For example, a template over some type parameter ``T`` can be
3253 instantiated across any particular implementation that conforms to the
3254 interface or *concept*. A good example here is the highly generic properties of
3255 any type which models a node in a directed graph. LLVM models these primarily
3256 through templates and generic programming. Such templates include the
3257 ``LoopInfoBase`` and ``DominatorTreeBase``. When this type of polymorphism
3258 truly needs **dynamic** dispatch you can generalize it using a technique
3259 called *concept-based polymorphism*. This pattern emulates the interfaces and
3260 behaviors of templates using a very limited form of virtual dispatch for type
3261 erasure inside its implementation. You can find examples of this technique in
3262 the ``PassManager.h`` system, and there is a more detailed introduction to it
3263 by Sean Parent in several of his talks and papers:
3265 #. `Inheritance Is The Base Class of Evil
3266 <http://channel9.msdn.com/Events/GoingNative/2013/Inheritance-Is-The-Base-Class-of-Evil>`_
3267 - The GoingNative 2013 talk describing this technique, and probably the best
3269 #. `Value Semantics and Concepts-based Polymorphism
3270 <http://www.youtube.com/watch?v=_BpMYeUFXv8>`_ - The C++Now! 2012 talk
3271 describing this technique in more detail.
3272 #. `Sean Parent's Papers and Presentations
3273 <http://github.com/sean-parent/sean-parent.github.com/wiki/Papers-and-Presentations>`_
3274 - A GitHub project full of links to slides, video, and sometimes code.
3276 When deciding between creating a type hierarchy (with either tagged or virtual
3277 dispatch) and using templates or concepts-based polymorphism, consider whether
3278 there is some refinement of an abstract base class which is a semantically
3279 meaningful type on an interface boundary. If anything more refined than the
3280 root abstract interface is meaningless to talk about as a partial extension of
3281 the semantic model, then your use case likely fits better with polymorphism and
3282 you should avoid using virtual dispatch. However, there may be some exigent
3283 circumstances that require one technique or the other to be used.
3285 If you do need to introduce a type hierarchy, we prefer to use explicitly
3286 closed type hierarchies with manual tagged dispatch and/or RTTI rather than the
3287 open inheritance model and virtual dispatch that is more common in C++ code.
3288 This is because LLVM rarely encourages library consumers to extend its core
3289 types, and leverages the closed and tag-dispatched nature of its hierarchies to
3290 generate significantly more efficient code. We have also found that a large
3291 amount of our usage of type hierarchies fits better with tag-based pattern
3292 matching rather than dynamic dispatch across a common interface. Within LLVM we
3293 have built custom helpers to facilitate this design. See this document's
3294 section on :ref:`isa and dyn_cast <isa>` and our :doc:`detailed document
3295 <HowToSetUpLLVMStyleRTTI>` which describes how you can implement this
3296 pattern for use with the LLVM helpers.
3298 .. _abi_breaking_checks:
3303 Checks and asserts that alter the LLVM C++ ABI are predicated on the
3304 preprocessor symbol `LLVM_ENABLE_ABI_BREAKING_CHECKS` -- LLVM
3305 libraries built with `LLVM_ENABLE_ABI_BREAKING_CHECKS` are not ABI
3306 compatible LLVM libraries built without it defined. By default,
3307 turning on assertions also turns on `LLVM_ENABLE_ABI_BREAKING_CHECKS`
3308 so a default +Asserts build is not ABI compatible with a
3309 default -Asserts build. Clients that want ABI compatibility
3310 between +Asserts and -Asserts builds should use the CMake build system
3311 to set `LLVM_ENABLE_ABI_BREAKING_CHECKS` independently
3312 of `LLVM_ENABLE_ASSERTIONS`.
3316 The Core LLVM Class Hierarchy Reference
3317 =======================================
3319 ``#include "llvm/IR/Type.h"``
3321 header source: `Type.h <https://llvm.org/doxygen/Type_8h_source.html>`_
3323 doxygen info: `Type Classes <https://llvm.org/doxygen/classllvm_1_1Type.html>`_
3325 The Core LLVM classes are the primary means of representing the program being
3326 inspected or transformed. The core LLVM classes are defined in header files in
3327 the ``include/llvm/IR`` directory, and implemented in the ``lib/IR``
3328 directory. It's worth noting that, for historical reasons, this library is
3329 called ``libLLVMCore.so``, not ``libLLVMIR.so`` as you might expect.
3333 The Type class and Derived Types
3334 --------------------------------
3336 ``Type`` is a superclass of all type classes. Every ``Value`` has a ``Type``.
3337 ``Type`` cannot be instantiated directly but only through its subclasses.
3338 Certain primitive types (``VoidType``, ``LabelType``, ``FloatType`` and
3339 ``DoubleType``) have hidden subclasses. They are hidden because they offer no
3340 useful functionality beyond what the ``Type`` class offers except to distinguish
3341 themselves from other subclasses of ``Type``.
3343 All other types are subclasses of ``DerivedType``. Types can be named, but this
3344 is not a requirement. There exists exactly one instance of a given shape at any
3345 one time. This allows type equality to be performed with address equality of
3346 the Type Instance. That is, given two ``Type*`` values, the types are identical
3347 if the pointers are identical.
3351 Important Public Methods
3352 ^^^^^^^^^^^^^^^^^^^^^^^^
3354 * ``bool isIntegerTy() const``: Returns true for any integer type.
3356 * ``bool isFloatingPointTy()``: Return true if this is one of the five
3357 floating point types.
3359 * ``bool isSized()``: Return true if the type has known size. Things
3360 that don't have a size are abstract types, labels and void.
3364 Important Derived Types
3365 ^^^^^^^^^^^^^^^^^^^^^^^
3368 Subclass of DerivedType that represents integer types of any bit width. Any
3369 bit width between ``IntegerType::MIN_INT_BITS`` (1) and
3370 ``IntegerType::MAX_INT_BITS`` (~8 million) can be represented.
3372 * ``static const IntegerType* get(unsigned NumBits)``: get an integer
3373 type of a specific bit width.
3375 * ``unsigned getBitWidth() const``: Get the bit width of an integer type.
3378 This is subclassed by ArrayType and VectorType.
3380 * ``const Type * getElementType() const``: Returns the type of each
3381 of the elements in the sequential type.
3383 * ``uint64_t getNumElements() const``: Returns the number of elements
3384 in the sequential type.
3387 This is a subclass of SequentialType and defines the interface for array
3391 Subclass of Type for pointer types.
3394 Subclass of SequentialType for vector types. A vector type is similar to an
3395 ArrayType but is distinguished because it is a first class type whereas
3396 ArrayType is not. Vector types are used for vector operations and are usually
3397 small vectors of an integer or floating point type.
3400 Subclass of DerivedTypes for struct types.
3405 Subclass of DerivedTypes for function types.
3407 * ``bool isVarArg() const``: Returns true if it's a vararg function.
3409 * ``const Type * getReturnType() const``: Returns the return type of the
3412 * ``const Type * getParamType (unsigned i)``: Returns the type of the ith
3415 * ``const unsigned getNumParams() const``: Returns the number of formal
3420 The ``Module`` class
3421 --------------------
3423 ``#include "llvm/IR/Module.h"``
3425 header source: `Module.h <https://llvm.org/doxygen/Module_8h_source.html>`_
3427 doxygen info: `Module Class <https://llvm.org/doxygen/classllvm_1_1Module.html>`_
3429 The ``Module`` class represents the top level structure present in LLVM
3430 programs. An LLVM module is effectively either a translation unit of the
3431 original program or a combination of several translation units merged by the
3432 linker. The ``Module`` class keeps track of a list of :ref:`Function
3433 <c_Function>`\ s, a list of GlobalVariable_\ s, and a SymbolTable_.
3434 Additionally, it contains a few helpful member functions that try to make common
3439 Important Public Members of the ``Module`` class
3440 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3442 * ``Module::Module(std::string name = "")``
3444 Constructing a Module_ is easy. You can optionally provide a name for it
3445 (probably based on the name of the translation unit).
3447 * | ``Module::iterator`` - Typedef for function list iterator
3448 | ``Module::const_iterator`` - Typedef for const_iterator.
3449 | ``begin()``, ``end()``, ``size()``, ``empty()``
3451 These are forwarding methods that make it easy to access the contents of a
3452 ``Module`` object's :ref:`Function <c_Function>` list.
3454 * ``Module::FunctionListType &getFunctionList()``
3456 Returns the list of :ref:`Function <c_Function>`\ s. This is necessary to use
3457 when you need to update the list or perform a complex action that doesn't have
3458 a forwarding method.
3462 * | ``Module::global_iterator`` - Typedef for global variable list iterator
3463 | ``Module::const_global_iterator`` - Typedef for const_iterator.
3464 | ``global_begin()``, ``global_end()``, ``global_size()``, ``global_empty()``
3466 These are forwarding methods that make it easy to access the contents of a
3467 ``Module`` object's GlobalVariable_ list.
3469 * ``Module::GlobalListType &getGlobalList()``
3471 Returns the list of GlobalVariable_\ s. This is necessary to use when you
3472 need to update the list or perform a complex action that doesn't have a
3477 * ``SymbolTable *getSymbolTable()``
3479 Return a reference to the SymbolTable_ for this ``Module``.
3483 * ``Function *getFunction(StringRef Name) const``
3485 Look up the specified function in the ``Module`` SymbolTable_. If it does not
3486 exist, return ``null``.
3488 * ``FunctionCallee getOrInsertFunction(const std::string &Name,
3489 const FunctionType *T)``
3491 Look up the specified function in the ``Module`` SymbolTable_. If
3492 it does not exist, add an external declaration for the function and
3493 return it. Note that the function signature already present may not
3494 match the requested signature. Thus, in order to enable the common
3495 usage of passing the result directly to EmitCall, the return type is
3496 a struct of ``{FunctionType *T, Constant *FunctionPtr}``, rather
3497 than simply the ``Function*`` with potentially an unexpected
3500 * ``std::string getTypeName(const Type *Ty)``
3502 If there is at least one entry in the SymbolTable_ for the specified Type_,
3503 return it. Otherwise return the empty string.
3505 * ``bool addTypeName(const std::string &Name, const Type *Ty)``
3507 Insert an entry in the SymbolTable_ mapping ``Name`` to ``Ty``. If there is
3508 already an entry for this name, true is returned and the SymbolTable_ is not
3516 ``#include "llvm/IR/Value.h"``
3518 header source: `Value.h <https://llvm.org/doxygen/Value_8h_source.html>`_
3520 doxygen info: `Value Class <https://llvm.org/doxygen/classllvm_1_1Value.html>`_
3522 The ``Value`` class is the most important class in the LLVM Source base. It
3523 represents a typed value that may be used (among other things) as an operand to
3524 an instruction. There are many different types of ``Value``\ s, such as
3525 Constant_\ s, Argument_\ s. Even Instruction_\ s and :ref:`Function
3526 <c_Function>`\ s are ``Value``\ s.
3528 A particular ``Value`` may be used many times in the LLVM representation for a
3529 program. For example, an incoming argument to a function (represented with an
3530 instance of the Argument_ class) is "used" by every instruction in the function
3531 that references the argument. To keep track of this relationship, the ``Value``
3532 class keeps a list of all of the ``User``\ s that is using it (the User_ class
3533 is a base class for all nodes in the LLVM graph that can refer to ``Value``\ s).
3534 This use list is how LLVM represents def-use information in the program, and is
3535 accessible through the ``use_*`` methods, shown below.
3537 Because LLVM is a typed representation, every LLVM ``Value`` is typed, and this
3538 Type_ is available through the ``getType()`` method. In addition, all LLVM
3539 values can be named. The "name" of the ``Value`` is a symbolic string printed
3542 .. code-block:: llvm
3548 The name of this instruction is "foo". **NOTE** that the name of any value may
3549 be missing (an empty string), so names should **ONLY** be used for debugging
3550 (making the source code easier to read, debugging printouts), they should not be
3551 used to keep track of values or map between them. For this purpose, use a
3552 ``std::map`` of pointers to the ``Value`` itself instead.
3554 One important aspect of LLVM is that there is no distinction between an SSA
3555 variable and the operation that produces it. Because of this, any reference to
3556 the value produced by an instruction (or the value available as an incoming
3557 argument, for example) is represented as a direct pointer to the instance of the
3558 class that represents this value. Although this may take some getting used to,
3559 it simplifies the representation and makes it easier to manipulate.
3563 Important Public Members of the ``Value`` class
3564 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3566 * | ``Value::use_iterator`` - Typedef for iterator over the use-list
3567 | ``Value::const_use_iterator`` - Typedef for const_iterator over the
3569 | ``unsigned use_size()`` - Returns the number of users of the value.
3570 | ``bool use_empty()`` - Returns true if there are no users.
3571 | ``use_iterator use_begin()`` - Get an iterator to the start of the
3573 | ``use_iterator use_end()`` - Get an iterator to the end of the use-list.
3574 | ``User *use_back()`` - Returns the last element in the list.
3576 These methods are the interface to access the def-use information in LLVM.
3577 As with all other iterators in LLVM, the naming conventions follow the
3578 conventions defined by the STL_.
3580 * ``Type *getType() const``
3581 This method returns the Type of the Value.
3583 * | ``bool hasName() const``
3584 | ``std::string getName() const``
3585 | ``void setName(const std::string &Name)``
3587 This family of methods is used to access and assign a name to a ``Value``, be
3588 aware of the :ref:`precaution above <nameWarning>`.
3590 * ``void replaceAllUsesWith(Value *V)``
3592 This method traverses the use list of a ``Value`` changing all User_\ s of the
3593 current value to refer to "``V``" instead. For example, if you detect that an
3594 instruction always produces a constant value (for example through constant
3595 folding), you can replace all uses of the instruction with the constant like
3600 Inst->replaceAllUsesWith(ConstVal);
3607 ``#include "llvm/IR/User.h"``
3609 header source: `User.h <https://llvm.org/doxygen/User_8h_source.html>`_
3611 doxygen info: `User Class <https://llvm.org/doxygen/classllvm_1_1User.html>`_
3615 The ``User`` class is the common base class of all LLVM nodes that may refer to
3616 ``Value``\ s. It exposes a list of "Operands" that are all of the ``Value``\ s
3617 that the User is referring to. The ``User`` class itself is a subclass of
3620 The operands of a ``User`` point directly to the LLVM ``Value`` that it refers
3621 to. Because LLVM uses Static Single Assignment (SSA) form, there can only be
3622 one definition referred to, allowing this direct connection. This connection
3623 provides the use-def information in LLVM.
3627 Important Public Members of the ``User`` class
3628 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3630 The ``User`` class exposes the operand list in two ways: through an index access
3631 interface and through an iterator based interface.
3633 * | ``Value *getOperand(unsigned i)``
3634 | ``unsigned getNumOperands()``
3636 These two methods expose the operands of the ``User`` in a convenient form for
3639 * | ``User::op_iterator`` - Typedef for iterator over the operand list
3640 | ``op_iterator op_begin()`` - Get an iterator to the start of the operand
3642 | ``op_iterator op_end()`` - Get an iterator to the end of the operand list.
3644 Together, these methods make up the iterator based interface to the operands
3650 The ``Instruction`` class
3651 -------------------------
3653 ``#include "llvm/IR/Instruction.h"``
3655 header source: `Instruction.h
3656 <https://llvm.org/doxygen/Instruction_8h_source.html>`_
3658 doxygen info: `Instruction Class
3659 <https://llvm.org/doxygen/classllvm_1_1Instruction.html>`_
3661 Superclasses: User_, Value_
3663 The ``Instruction`` class is the common base class for all LLVM instructions.
3664 It provides only a few methods, but is a very commonly used class. The primary
3665 data tracked by the ``Instruction`` class itself is the opcode (instruction
3666 type) and the parent BasicBlock_ the ``Instruction`` is embedded into. To
3667 represent a specific type of instruction, one of many subclasses of
3668 ``Instruction`` are used.
3670 Because the ``Instruction`` class subclasses the User_ class, its operands can
3671 be accessed in the same way as for other ``User``\ s (with the
3672 ``getOperand()``/``getNumOperands()`` and ``op_begin()``/``op_end()`` methods).
3673 An important file for the ``Instruction`` class is the ``llvm/Instruction.def``
3674 file. This file contains some meta-data about the various different types of
3675 instructions in LLVM. It describes the enum values that are used as opcodes
3676 (for example ``Instruction::Add`` and ``Instruction::ICmp``), as well as the
3677 concrete sub-classes of ``Instruction`` that implement the instruction (for
3678 example BinaryOperator_ and CmpInst_). Unfortunately, the use of macros in this
3679 file confuses doxygen, so these enum values don't show up correctly in the
3680 `doxygen output <https://llvm.org/doxygen/classllvm_1_1Instruction.html>`_.
3684 Important Subclasses of the ``Instruction`` class
3685 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3689 * ``BinaryOperator``
3691 This subclasses represents all two operand instructions whose operands must be
3692 the same type, except for the comparison instructions.
3697 This subclass is the parent of the 12 casting instructions. It provides
3698 common operations on cast instructions.
3704 This subclass represents the two comparison instructions,
3705 `ICmpInst <LangRef.html#i_icmp>`_ (integer operands), and
3706 `FCmpInst <LangRef.html#i_fcmp>`_ (floating point operands).
3710 Important Public Members of the ``Instruction`` class
3711 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3713 * ``BasicBlock *getParent()``
3715 Returns the BasicBlock_ that this
3716 ``Instruction`` is embedded into.
3718 * ``bool mayWriteToMemory()``
3720 Returns true if the instruction writes to memory, i.e. it is a ``call``,
3721 ``free``, ``invoke``, or ``store``.
3723 * ``unsigned getOpcode()``
3725 Returns the opcode for the ``Instruction``.
3727 * ``Instruction *clone() const``
3729 Returns another instance of the specified instruction, identical in all ways
3730 to the original except that the instruction has no parent (i.e. it's not
3731 embedded into a BasicBlock_), and it has no name.
3735 The ``Constant`` class and subclasses
3736 -------------------------------------
3738 Constant represents a base class for different types of constants. It is
3739 subclassed by ConstantInt, ConstantArray, etc. for representing the various
3740 types of Constants. GlobalValue_ is also a subclass, which represents the
3741 address of a global variable or function.
3745 Important Subclasses of Constant
3746 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3748 * ConstantInt : This subclass of Constant represents an integer constant of
3751 * ``const APInt& getValue() const``: Returns the underlying
3752 value of this constant, an APInt value.
3754 * ``int64_t getSExtValue() const``: Converts the underlying APInt value to an
3755 int64_t via sign extension. If the value (not the bit width) of the APInt
3756 is too large to fit in an int64_t, an assertion will result. For this
3757 reason, use of this method is discouraged.
3759 * ``uint64_t getZExtValue() const``: Converts the underlying APInt value
3760 to a uint64_t via zero extension. IF the value (not the bit width) of the
3761 APInt is too large to fit in a uint64_t, an assertion will result. For this
3762 reason, use of this method is discouraged.
3764 * ``static ConstantInt* get(const APInt& Val)``: Returns the ConstantInt
3765 object that represents the value provided by ``Val``. The type is implied
3766 as the IntegerType that corresponds to the bit width of ``Val``.
3768 * ``static ConstantInt* get(const Type *Ty, uint64_t Val)``: Returns the
3769 ConstantInt object that represents the value provided by ``Val`` for integer
3772 * ConstantFP : This class represents a floating point constant.
3774 * ``double getValue() const``: Returns the underlying value of this constant.
3776 * ConstantArray : This represents a constant array.
3778 * ``const std::vector<Use> &getValues() const``: Returns a vector of
3779 component constants that makeup this array.
3781 * ConstantStruct : This represents a constant struct.
3783 * ``const std::vector<Use> &getValues() const``: Returns a vector of
3784 component constants that makeup this array.
3786 * GlobalValue : This represents either a global variable or a function. In
3787 either case, the value is a constant fixed address (after linking).
3791 The ``GlobalValue`` class
3792 -------------------------
3794 ``#include "llvm/IR/GlobalValue.h"``
3796 header source: `GlobalValue.h
3797 <https://llvm.org/doxygen/GlobalValue_8h_source.html>`_
3799 doxygen info: `GlobalValue Class
3800 <https://llvm.org/doxygen/classllvm_1_1GlobalValue.html>`_
3802 Superclasses: Constant_, User_, Value_
3804 Global values ( GlobalVariable_\ s or :ref:`Function <c_Function>`\ s) are the
3805 only LLVM values that are visible in the bodies of all :ref:`Function
3806 <c_Function>`\ s. Because they are visible at global scope, they are also
3807 subject to linking with other globals defined in different translation units.
3808 To control the linking process, ``GlobalValue``\ s know their linkage rules.
3809 Specifically, ``GlobalValue``\ s know whether they have internal or external
3810 linkage, as defined by the ``LinkageTypes`` enumeration.
3812 If a ``GlobalValue`` has internal linkage (equivalent to being ``static`` in C),
3813 it is not visible to code outside the current translation unit, and does not
3814 participate in linking. If it has external linkage, it is visible to external
3815 code, and does participate in linking. In addition to linkage information,
3816 ``GlobalValue``\ s keep track of which Module_ they are currently part of.
3818 Because ``GlobalValue``\ s are memory objects, they are always referred to by
3819 their **address**. As such, the Type_ of a global is always a pointer to its
3820 contents. It is important to remember this when using the ``GetElementPtrInst``
3821 instruction because this pointer must be dereferenced first. For example, if
3822 you have a ``GlobalVariable`` (a subclass of ``GlobalValue)`` that is an array
3823 of 24 ints, type ``[24 x i32]``, then the ``GlobalVariable`` is a pointer to
3824 that array. Although the address of the first element of this array and the
3825 value of the ``GlobalVariable`` are the same, they have different types. The
3826 ``GlobalVariable``'s type is ``[24 x i32]``. The first element's type is
3827 ``i32.`` Because of this, accessing a global value requires you to dereference
3828 the pointer with ``GetElementPtrInst`` first, then its elements can be accessed.
3829 This is explained in the `LLVM Language Reference Manual
3830 <LangRef.html#globalvars>`_.
3834 Important Public Members of the ``GlobalValue`` class
3835 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3837 * | ``bool hasInternalLinkage() const``
3838 | ``bool hasExternalLinkage() const``
3839 | ``void setInternalLinkage(bool HasInternalLinkage)``
3841 These methods manipulate the linkage characteristics of the ``GlobalValue``.
3843 * ``Module *getParent()``
3845 This returns the Module_ that the
3846 GlobalValue is currently embedded into.
3850 The ``Function`` class
3851 ----------------------
3853 ``#include "llvm/IR/Function.h"``
3855 header source: `Function.h <https://llvm.org/doxygen/Function_8h_source.html>`_
3857 doxygen info: `Function Class
3858 <https://llvm.org/doxygen/classllvm_1_1Function.html>`_
3860 Superclasses: GlobalValue_, Constant_, User_, Value_
3862 The ``Function`` class represents a single procedure in LLVM. It is actually
3863 one of the more complex classes in the LLVM hierarchy because it must keep track
3864 of a large amount of data. The ``Function`` class keeps track of a list of
3865 BasicBlock_\ s, a list of formal Argument_\ s, and a SymbolTable_.
3867 The list of BasicBlock_\ s is the most commonly used part of ``Function``
3868 objects. The list imposes an implicit ordering of the blocks in the function,
3869 which indicate how the code will be laid out by the backend. Additionally, the
3870 first BasicBlock_ is the implicit entry node for the ``Function``. It is not
3871 legal in LLVM to explicitly branch to this initial block. There are no implicit
3872 exit nodes, and in fact there may be multiple exit nodes from a single
3873 ``Function``. If the BasicBlock_ list is empty, this indicates that the
3874 ``Function`` is actually a function declaration: the actual body of the function
3875 hasn't been linked in yet.
3877 In addition to a list of BasicBlock_\ s, the ``Function`` class also keeps track
3878 of the list of formal Argument_\ s that the function receives. This container
3879 manages the lifetime of the Argument_ nodes, just like the BasicBlock_ list does
3880 for the BasicBlock_\ s.
3882 The SymbolTable_ is a very rarely used LLVM feature that is only used when you
3883 have to look up a value by name. Aside from that, the SymbolTable_ is used
3884 internally to make sure that there are not conflicts between the names of
3885 Instruction_\ s, BasicBlock_\ s, or Argument_\ s in the function body.
3887 Note that ``Function`` is a GlobalValue_ and therefore also a Constant_. The
3888 value of the function is its address (after linking) which is guaranteed to be
3893 Important Public Members of the ``Function``
3894 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3896 * ``Function(const FunctionType *Ty, LinkageTypes Linkage,
3897 const std::string &N = "", Module* Parent = 0)``
3899 Constructor used when you need to create new ``Function``\ s to add the
3900 program. The constructor must specify the type of the function to create and
3901 what type of linkage the function should have. The FunctionType_ argument
3902 specifies the formal arguments and return value for the function. The same
3903 FunctionType_ value can be used to create multiple functions. The ``Parent``
3904 argument specifies the Module in which the function is defined. If this
3905 argument is provided, the function will automatically be inserted into that
3906 module's list of functions.
3908 * ``bool isDeclaration()``
3910 Return whether or not the ``Function`` has a body defined. If the function is
3911 "external", it does not have a body, and thus must be resolved by linking with
3912 a function defined in a different translation unit.
3914 * | ``Function::iterator`` - Typedef for basic block list iterator
3915 | ``Function::const_iterator`` - Typedef for const_iterator.
3916 | ``begin()``, ``end()``, ``size()``, ``empty()``
3918 These are forwarding methods that make it easy to access the contents of a
3919 ``Function`` object's BasicBlock_ list.
3921 * ``Function::BasicBlockListType &getBasicBlockList()``
3923 Returns the list of BasicBlock_\ s. This is necessary to use when you need to
3924 update the list or perform a complex action that doesn't have a forwarding
3927 * | ``Function::arg_iterator`` - Typedef for the argument list iterator
3928 | ``Function::const_arg_iterator`` - Typedef for const_iterator.
3929 | ``arg_begin()``, ``arg_end()``, ``arg_size()``, ``arg_empty()``
3931 These are forwarding methods that make it easy to access the contents of a
3932 ``Function`` object's Argument_ list.
3934 * ``Function::ArgumentListType &getArgumentList()``
3936 Returns the list of Argument_. This is necessary to use when you need to
3937 update the list or perform a complex action that doesn't have a forwarding
3940 * ``BasicBlock &getEntryBlock()``
3942 Returns the entry ``BasicBlock`` for the function. Because the entry block
3943 for the function is always the first block, this returns the first block of
3946 * | ``Type *getReturnType()``
3947 | ``FunctionType *getFunctionType()``
3949 This traverses the Type_ of the ``Function`` and returns the return type of
3950 the function, or the FunctionType_ of the actual function.
3952 * ``SymbolTable *getSymbolTable()``
3954 Return a pointer to the SymbolTable_ for this ``Function``.
3958 The ``GlobalVariable`` class
3959 ----------------------------
3961 ``#include "llvm/IR/GlobalVariable.h"``
3963 header source: `GlobalVariable.h
3964 <https://llvm.org/doxygen/GlobalVariable_8h_source.html>`_
3966 doxygen info: `GlobalVariable Class
3967 <https://llvm.org/doxygen/classllvm_1_1GlobalVariable.html>`_
3969 Superclasses: GlobalValue_, Constant_, User_, Value_
3971 Global variables are represented with the (surprise surprise) ``GlobalVariable``
3972 class. Like functions, ``GlobalVariable``\ s are also subclasses of
3973 GlobalValue_, and as such are always referenced by their address (global values
3974 must live in memory, so their "name" refers to their constant address). See
3975 GlobalValue_ for more on this. Global variables may have an initial value
3976 (which must be a Constant_), and if they have an initializer, they may be marked
3977 as "constant" themselves (indicating that their contents never change at
3980 .. _m_GlobalVariable:
3982 Important Public Members of the ``GlobalVariable`` class
3983 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
3985 * ``GlobalVariable(const Type *Ty, bool isConstant, LinkageTypes &Linkage,
3986 Constant *Initializer = 0, const std::string &Name = "", Module* Parent = 0)``
3988 Create a new global variable of the specified type. If ``isConstant`` is true
3989 then the global variable will be marked as unchanging for the program. The
3990 Linkage parameter specifies the type of linkage (internal, external, weak,
3991 linkonce, appending) for the variable. If the linkage is InternalLinkage,
3992 WeakAnyLinkage, WeakODRLinkage, LinkOnceAnyLinkage or LinkOnceODRLinkage, then
3993 the resultant global variable will have internal linkage. AppendingLinkage
3994 concatenates together all instances (in different translation units) of the
3995 variable into a single variable but is only applicable to arrays. See the
3996 `LLVM Language Reference <LangRef.html#modulestructure>`_ for further details
3997 on linkage types. Optionally an initializer, a name, and the module to put
3998 the variable into may be specified for the global variable as well.
4000 * ``bool isConstant() const``
4002 Returns true if this is a global variable that is known not to be modified at
4005 * ``bool hasInitializer()``
4007 Returns true if this ``GlobalVariable`` has an initializer.
4009 * ``Constant *getInitializer()``
4011 Returns the initial value for a ``GlobalVariable``. It is not legal to call
4012 this method if there is no initializer.
4016 The ``BasicBlock`` class
4017 ------------------------
4019 ``#include "llvm/IR/BasicBlock.h"``
4021 header source: `BasicBlock.h
4022 <https://llvm.org/doxygen/BasicBlock_8h_source.html>`_
4024 doxygen info: `BasicBlock Class
4025 <https://llvm.org/doxygen/classllvm_1_1BasicBlock.html>`_
4029 This class represents a single entry single exit section of the code, commonly
4030 known as a basic block by the compiler community. The ``BasicBlock`` class
4031 maintains a list of Instruction_\ s, which form the body of the block. Matching
4032 the language definition, the last element of this list of instructions is always
4033 a terminator instruction.
4035 In addition to tracking the list of instructions that make up the block, the
4036 ``BasicBlock`` class also keeps track of the :ref:`Function <c_Function>` that
4037 it is embedded into.
4039 Note that ``BasicBlock``\ s themselves are Value_\ s, because they are
4040 referenced by instructions like branches and can go in the switch tables.
4041 ``BasicBlock``\ s have type ``label``.
4045 Important Public Members of the ``BasicBlock`` class
4046 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
4048 * ``BasicBlock(const std::string &Name = "", Function *Parent = 0)``
4050 The ``BasicBlock`` constructor is used to create new basic blocks for
4051 insertion into a function. The constructor optionally takes a name for the
4052 new block, and a :ref:`Function <c_Function>` to insert it into. If the
4053 ``Parent`` parameter is specified, the new ``BasicBlock`` is automatically
4054 inserted at the end of the specified :ref:`Function <c_Function>`, if not
4055 specified, the BasicBlock must be manually inserted into the :ref:`Function
4058 * | ``BasicBlock::iterator`` - Typedef for instruction list iterator
4059 | ``BasicBlock::const_iterator`` - Typedef for const_iterator.
4060 | ``begin()``, ``end()``, ``front()``, ``back()``,
4061 ``size()``, ``empty()``
4062 STL-style functions for accessing the instruction list.
4064 These methods and typedefs are forwarding functions that have the same
4065 semantics as the standard library methods of the same names. These methods
4066 expose the underlying instruction list of a basic block in a way that is easy
4067 to manipulate. To get the full complement of container operations (including
4068 operations to update the list), you must use the ``getInstList()`` method.
4070 * ``BasicBlock::InstListType &getInstList()``
4072 This method is used to get access to the underlying container that actually
4073 holds the Instructions. This method must be used when there isn't a
4074 forwarding function in the ``BasicBlock`` class for the operation that you
4075 would like to perform. Because there are no forwarding functions for
4076 "updating" operations, you need to use this if you want to update the contents
4077 of a ``BasicBlock``.
4079 * ``Function *getParent()``
4081 Returns a pointer to :ref:`Function <c_Function>` the block is embedded into,
4082 or a null pointer if it is homeless.
4084 * ``Instruction *getTerminator()``
4086 Returns a pointer to the terminator instruction that appears at the end of the
4087 ``BasicBlock``. If there is no terminator instruction, or if the last
4088 instruction in the block is not a terminator, then a null pointer is returned.
4092 The ``Argument`` class
4093 ----------------------
4095 This subclass of Value defines the interface for incoming formal arguments to a
4096 function. A Function maintains a list of its formal arguments. An argument has
4097 a pointer to the parent Function.